2026
A randomized controlled trial at the University of Maryland, College Park, in fall 2025 gave some instructors' course sections access to a GPT-4o-based virtual study assistant built into the campus learning management system, with roughly 2,400 undergraduates across many disciplines. Access to the AI tutor lowered final course grades and sharply reduced students' participation in existing online course activities, with larger estimated grade losses for first-generation students.
Findings
- Among sections of the same course, AI tutor access reduced final grades by 0.37 standard deviations and learning management system participation by 0.90 standard deviations.
“Among sections of the same course, tutor access reduced final grades by 0.37 standard deviations and learning management system participation by 0.90 standard deviations.”
Abstract - Estimated effects on final grades ranged from -0.27 to -0.37 standard deviations across samples, and effects on homework scores were null or negative.
“We found that access to the AI tutor negatively affected student final course grades by 0.27–0.37 standard deviations (SDs), with the clearest evidence in the exact-match sample.”
Introduction, p. 3 - Estimated final-grade losses for first-generation students were more than twice those for continuing-generation students.
“the estimated final-grade losses among first-generation students were more than twice those among continuing-generation students in both samples”
Introduction, p. 4
Method
Randomized controlled trial with instructors randomized within blocks of identical or similar courses; 2,379 undergraduate students and 30 instructors across STEM, social sciences, humanities, and education at the University of Maryland, College Park, fall 2025; outcomes from course grades, learning management system activity, and surveys.
- Published
- Oct 2026
- Publisher
- Annenberg Institute at Brown University (EdWorkingPapers)
- Authors
- Jing Liu, Tracy Sweet, Marty Haoyuan Chen, Jonathan Engelberg, Megan Masters, Martyn Clark, Axel Persaud, Alia Lancaster, Jeffrey Hollingsworth, Jennifer King Rice
- Type
- Research report
- Level
- Higher ed
- Covers
- Maryland
- Topics
- Equity & Access, Instructional Use, Research
The Effects of Course-Integrated AI Tutoring on Student Performance and Engagement: A Randomized University Trial (EdWorkingPaper 26-1598)The Effects of Course-Integrated AI Tutoring on Student Performance and Engagement: A Randomized University Trial (full text PDF)
A randomized field experiment with more than 6,000 middle school students in Hamilton County Schools, Tennessee, compared math practice on a computer-assisted learning platform with and without a large language model tutor, and with and without mastery-based progression, followed by a delayed test a week later. Students with AI support progressed more slowly but recovered from mistakes more effectively, and the most encouraging delayed-test gains, which were only marginally significant, appeared when the AI was embedded in the mastery workflow.
Findings
- Students with AI support progressed more slowly and attempted fewer questions, but answered more accurately when they did attempt, and the AI improved correctness on the attempt following a mistake.
“AI students progressed more slowly and attempted fewer questions, but answered more accurately conditional on reaching an attempt. The clearest mechanism appeared after mistakes: AI improved next-attempt correctness”
Abstract - Mastery-based progression increased three-correct-in-a-row attainment but did not by itself improve delayed learning.
“Mastery increased three-correct-in-a-row attainment but did not by itself improve delayed learning.”
Abstract - The strongest delayed-test evidence came from combining AI with the mastery workflow, with marginally significant gains concentrated on practiced material.
“The most encouraging delayed-test evidence appears when AI is embedded in the mastery workflow, with marginally significant gains concentrated on practiced Exercise 1 material.”
Abstract
Method
Registered randomized field experiment (AEA RCT Registry AEARCTR-0018678); more than 6,000 students in grades 6-8 across 20 schools in Hamilton County Schools, Tennessee, March-April 2026; cross-randomized AI vs. computer-assisted-learning-only support, mastery vs. non-mastery progression, and math topic; delayed assessment about one week later.
- Published
- Aug 2026
- Publisher
- National Bureau of Economic Research
- Authors
- Philip Oreopoulos, Michael Liut, Alp Sungu, Nina Low
- Type
- Research report
- Level
- K-12
- Covers
- Tennessee
- Topics
- Instructional Use, Research
Making AI Tutoring Productive: Evidence from a Mastery-Based Math Practice Experiment (NBER Working Paper 35621)Making AI Tutoring Productive: Evidence from a Mastery-Based Math Practice Experiment (full text PDF)
A two-year cluster randomized trial in 18 middle schools in Hamilton County, Tennessee, randomly assigned grades within each school to use Khan Academy with its Khanmigo AI tutor, set to coach rather than give answers, during daily remedial math sessions. Assignment produced small math gains similar to those from Khan Academy practice without AI, and the authors attribute the modest effect to students rarely engaging the tutor in substantive dialogue.
Findings
- Assignment to Khanmigo raised math achievement by 1.3 national percentile ranks per term, about 0.06 to 0.08 standard deviations over a school year, with an implied effect of 0.14 standard deviations for a full year of active participation.
“Assignment raises math achievement by 1.3 national percentile ranks per term, or about 0.06 to 0.08 standard deviations over a school year; the implied effect of a full year of active participation reaches 0.14 standard deviations.”
Abstract - These gains resembled those from Khan Academy practice without AI assistance.
“These gains resemble those from Khan Academy practice without AI assistance.”
Abstract - Although 96% of students tried Khanmigo, the median student messaged it on only a third of practice days and in only 17% of sessions in which they made a mistake.
“96 percent of students tried Khanmigo at least once, but the median student messaged it on only a third of the days they practiced, and in only 17 percent of the exercise sessions in which they made a mistake.”
Abstract
Method
Preregistered cluster randomized trial over the 2024-25 and 2025-26 school years; one or two of three grades (6-8) randomly assigned within each of 18 middle schools in Hamilton County Schools, Tennessee; below-grade-level students in daily remedial math sessions; 6,902 student-term observations with MAP math scores.
- Published
- Aug 2026
- Publisher
- Annenberg Institute at Brown University (EdWorkingPapers)
- Authors
- Philip Oreopoulos, Nina Low
- Type
- Research report
- Level
- K-12
- Covers
- Tennessee
- Topics
- Instructional Use, Research
One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment (EdWorkingPaper 26-1551)One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment (full text PDF)
Researchers studied California State University, Northridge's use of an AI-enabled text-messaging chatbot to support undergraduates, combining system observation, administrator discussions, and a randomized controlled trial following the 2018 and 2019 entering first-year cohorts over four years. Students stayed receptive to the messages, and effects were concentrated in completing time-sensitive administrative tasks, with no detectable effects on academic performance, persistence, or graduation.
Findings
- Effects were concentrated in improved completion of time-sensitive administrative tasks, with no detectable effects on academic performance or persistence.
“Students remain receptive over time, and impacts are concentrated in improved completion of time-sensitive administrative tasks, with no detectable effects on academic performance or persistence.”
Abstract - There was no statistically significant difference in enrollment and persistence between treatment and control students, and no evidence of a difference in graduation rates.
“We find no statistically significant difference in enrollment and persistence between students in the treatment and control groups”
Results, enrollment and persistence (Table 4) - Centralized ownership of the tool and flexible communication were identified as critical for sustaining the program.
“We find that centralized ownership and flexible communication are critical for sustainability.”
Abstract
Method
Mixed-methods implementation study with a randomized controlled trial; entering first-year cohorts of 2018 and 2019 at California State University, Northridge (about 4,361 treatment and 4,347 control students; 8,708 observations), followed for eight semesters; intent-to-treat estimates.
- Published
- Jul 2026
- Publisher
- Annenberg Institute at Brown University (EdWorkingPapers)
- Authors
- Catherine Mata, Emily Russell, Lindsay C. Page
- Type
- Research report
- Level
- Higher ed
- Covers
- California
- Topics
- Institutional Operations, Research
Sustaining AI-Enabled Student Support: A Four-Year Implementation and Impact Study (EdWorkingPaper 26-1409)Sustaining AI-Enabled Student Support: A Four-Year Implementation and Impact Study (full text PDF)
Annual national surveys of students, instructors, and administrators at US colleges, fielded in April and May 2026, on AI adoption, assessment practices, institutional AI policy, and workforce readiness. More than half of each group reported at least weekly AI use, faculty who redesigned assessments around AI reported better engagement than those who restricted it, and few institutions had an institution-wide AI policy.
Findings
- For the first time, more than half of administrators, faculty, and students reported using AI at least weekly.
“For the first time, more than half of administrators, faculty, and students use AI at least weekly, with daily GenAI use at its highest point since Spring 2023 (43% of administrators, 32% of students, and 25% of faculty).”
Key insights - Only 13% of faculty at institutions that ban AI said their policy was effective, compared with 30% where AI is integrated into assignments.
“Only 13% of faculty at institutions that ban AI entirely say their policy is effective, versus 30% where AI is integrated into assignments.”
Key insights - 32% of administrators reported an institution-wide AI policy; where one existed, 22% of faculty considered it effective versus 44% of administrators.
“Just 32% of administrators report an institution-wide AI policy, and where one exists, only 22% of faculty believe it's effective compared to 44% of administrators.”
Key insights
Method
National surveys of more than 3,000 students, instructors, and administrators at over 750 US colleges and universities, fielded April-May 2026.
- Published
- 12 Jun 2026
- Publisher
- Tyton Partners
- Authors
- Catherine Shaw, Hadley Dorn, Amrita Narayanan, Louise Hay, Gates Bryant
- Type
- Survey or poll
- Level
- Higher ed
- Topics
- Academic Integrity, Assessment, Instructional Use
Time for Class 2026: The AI Tipping Point: From Monitoring Students to Engaging Them
RAND's December 2025 American Youth Panel survey of 1,214 people ages 12 to 29 asked how and why middle school, high school, and college students use AI for school, with trends from earlier 2025 surveys. The share using AI for homework help rose over 2025, and a growing majority agreed that more AI use for schoolwork will harm critical thinking.
Findings
- The share of students in middle school and up who used AI for homework help rose from 48 percent in May 2025 to 62 percent in December 2025, driven by middle and high schoolers.
“The share of students in middle school grades and up who reported using AI for help with their homework increased from 48 percent in May 2025 to 62 percent in December 2025.”
Key Takeaways - By December 2025, 67 percent of students agreed that more AI use for schoolwork will harm their critical thinking skills, up more than 10 percentage points in ten months.
“As of December 2025, 67 percent of students endorsed the statement "The more students use AI for their schoolwork, the more it will harm their critical thinking skills"”
Key Takeaways - Except for using AI to get homework answers, most students did not consider their school-related AI uses to be cheating.
“With the exception of using AI to get answers for homework, most students did not feel that their use of AI for other school-related purposes constituted cheating.”
Key Takeaways
Method
Nationally representative survey of 1,214 youth ages 12-29 (RAND American Youth Panel, December 2025), with trend comparisons to February and May 2025 waves.
- Published
- 17 Mar 2026
- Publisher
- RAND Corporation
- Authors
- Heather L. Schwartz, Melissa Kay Diliberti
- Type
- Survey or poll
- Level
- K-12, Higher ed
- Topics
- Academic Integrity, Instructional Use
More Students Use AI for Homework, and More Believe It Harms Critical Thinking: Selected Findings from the American Youth PanelArchived copy
CRPE released a database of AI-related actions by 20 early adopter states in 2024 and 2025, covering guidance, legislation, executive orders, professional learning, pilots, and partnerships, with a preliminary, non-evaluative analysis. Most of these states issued non-binding guidance and invested in educator training, about half ran pilots, and approaches looked similar across political contexts.
Findings
- Of the 20 states, 18 provided professional learning, 17 used partnerships, 16 issued guidance, and 11 piloted AI strategies or tools.
“Of the 20 states we examined, they commonly provide professional learning opportunities (18 of 20), leverage partnerships to support AI goals (17 of 20), issue guidance (16 of 20), and pilot AI strategies or tools (11 of 20).”
p. 2 - All of the guidance issued by the 16 states was non-binding, and many states updated it regularly.
“Sixteen of 20 have issued AI guidance, all of which is non-binding”
Flexibility over Directives, p. 2 - In 19 of the 20 states, legislatures introduced at least one AI bill involving K-12 education or the state education agency in 2024 or 2025.
“In 19 of the 20 states (95%) in the database, legislatures introduced at least one AI-related bill in 2024 or 2025 that involved K-12 education and/or impacted the SEA.”
p. 3
Method
Database of 20 early adopter states compiled from publicly available information captured November 2025 through January 2026; states chosen through expert interviews, observed AI actions, and geographic and political variety.
- Published
- Mar 2026
- Publisher
- Center on Reinventing Public Education
- Authors
- Dana Harrison, Bree Dusseault
- Type
- Policy brief
- Level
- K-12
- Topics
- Educator Training, Local Policy Requirement, Procurement & Vendors, Task Forces & Study
States and AI: An Early Look at How Early Adopters Are Approaching AI in EducationArchived copyStates and AI: An Early Look at How States Are Approaching AI in Education (brief)Archived copy
The Brookings Global Task Force on AI in Education conducted a yearlong global "premortem" study of the risks generative AI poses to children's learning and development, drawing on consultations with more than 500 people in 50 countries, a review of more than 400 studies, and a Delphi panel. It concludes that at this point the risks to children's education overshadow the benefits, and it offers recommendations under three pillars, Prosper, Prepare, and Protect.
Findings
- The study concludes that, at this point in its development, the risks of generative AI in children's education overshadow its benefits, largely because the risks undermine foundational development.
“we find that at this point in its trajectory, the risks of utilizing generative AI in children’s education overshadow its benefits. This is largely because the risks of AI differ in nature from its benefits—that is, these risks undermine children’s foundational development”
Source - The authors find that well-designed AI tools can benefit learning when used within a pedagogically sound approach, while overreliance can put children's learning capacity, wellbeing, relationships, safety, and privacy at risk.
“Overreliance on AI tools and platforms can put children and youth’s fundamental learning capacity at risk.”
Source - Recommendations include comprehensive regulatory frameworks for educational AI and procurement that protects student privacy, safety, and security.
“Procure technology that protects students’ privacy, safety, and security.”
Protect pillar
Method
Yearlong global qualitative study with interviews, focus groups, and consultations with more than 500 students, teachers, parents, education leaders, and technologists in 50 countries; review of more than 400 studies; Delphi panel.
- Published
- 14 Jan 2026
- Publisher
- Brookings Institution, Center for Universal Education
- Authors
- Mary Burns, Rebecca Winthrop, Natasha Luther, Emma Venetis, Rida Karim
- Type
- Research report
- Level
- K-12
- Topics
- AI Literacy, Equity & Access, Procurement & Vendors, Student Safety & Wellbeing, Student Privacy
A new direction for students in an AI world: Prosper, prepare, protectArchived copyA New Direction for Students in an AI World: Prosper, Prepare, Protect (full report)Archived copy
A global survey drawing on 45,398 responses from students and faculty in 35 countries on AI adoption, AI literacy, course integration, and attitudes toward AI in higher education. AI integration in courses was uneven and its learning value mixed, and respondents in the US and Canada were more cautious than other regions, with faculty there reporting declining intent to use AI in teaching.
Findings
- Only 15% of students said AI is integrated into many of their courses, and 43% said they had not experienced any AI integration.
“Only 15% of students say AI is integrated into many of their courses, while 43% say it appears in a few. Another 43% say they have not experienced any AI integration in their courses.”
Key insight 02 - Faculty intent to use AI in teaching in the US and Canada fell from 76% in 2025 to 67% in 2026, the lowest of any region.
“Faculty intent to use AI has declined by 9 percentage points, from 76% in 2025 to 67% in 2026.”
Key insight 05 - 60% of students worldwide, and 73% in the US and Canada, worried that classmates might misuse AI for unfair advantage.
“60% of students globally worry that their classmates might misuse AI for unfair advantage, rising to 73% in US & Canada.”
Key insight 07 - Only 29% of students believed their instructors were well equipped to guide them on AI use, falling to 17% in the US and Canada.
“Only 29% of students believe their instructors are well equipped to guide them on AI use. In the US & Canada, this number is considerably lower at 17%.”
Key insight 03
Method
Survey of 27,284 students and 18,114 faculty (45,398 responses) across 35 countries, with regional breakdowns including US and Canada.
- Published
- 2026
- Publisher
- Digital Education Council
- Authors
- Digital Education Council
- Type
- Survey or poll
- Level
- Higher ed
- Topics
- Academic Integrity, AI Literacy, Educator Training, Instructional Use
2025
The Southern Regional Education Board's Commission on AI in Education, which convened more than 75 education and business leaders in 2024 and 2025, approved 12 final policy recommendations for states. They include statewide AI networks, targeted guidance, professional development, AI in K-12 standards, needs assessments, procurement and safety policies, postsecondary adoption playbooks, and evaluation frameworks.
Findings
- The Commission approved 12 final recommendations for states, schools, and colleges, with the last three covering safety, postsecondary adoption, and evaluation.
“The SREB Commission on AI in Education has approved 12 final recommendations to help states, schools and colleges as they adopt artificial intelligence tools.”
Source - It recommends that states regularly review education technology safety policies to protect minors from AI-related harm.
“States should regularly review, and revise if necessary, their educational technology safety policies to ensure that minors are protected from harm related to AI use.”
Recommendation 10 - It recommends that states integrate AI knowledge and skills into K-12 standards and curricula.
“States should integrate AI knowledge and skills into K-12 standards and curricula”
Source
Method
Policy recommendations developed by a regional commission of more than 75 education and business leaders during 2024 and 2025.
- Published
- 2 Dec 2025
- Publisher
- Southern Regional Education Board
- Authors
- SREB Commission on AI in Education
- Type
- Policy brief
- Level
- K-12, Higher ed
- Topics
- Curriculum & Standards, Educator Training, Procurement & Vendors, Student Safety & Wellbeing, Task Forces & Study, Workforce & Careers
SREB AI in Education Commission Approves Final RecommendationsAI Commission RecommendationsArchived copy
Update to CRPE's database of early adopter school districts for the 2025-26 school year, compiled from public information. The database grew from 40 to 79 districts, more of them published AI guidance and offered teacher training, and more used student-facing AI tools, though few changed courses or learning standards.
Findings
- The early adopter database nearly doubled from 40 to 79 districts; professional development offerings rose to 86% from 63%, and published guidance rose to 78% from 65%.
“Our Early Adopters database has nearly doubled in a year, from 40 to 79 districts.”
More Early Adopters Are Piloting Systemwide AI Strategies - More than a third (37%) of districts were implementing five or more AI strategies, up from a few initiatives per district the prior year.
“this year, we see more than a third (37%) implementing five or more strategies”
More Early Adopters Are Piloting Systemwide AI Strategies - Nearly two-thirds of districts reported student-centered AI tools, but only 27% shared information about AI courses and six districts (8%) reported changes to learning standards.
“Only a quarter (27%) share information about AI courses, and just six districts (8%) communicate any adjustments to learning standards.”
Early Adopters Focus More on Students, Mostly with New Tools
Method
Database of 79 early adopter districts and charter management organizations for 2025-26 compiled from publicly available information on AI initiatives; not a representative sample.
- Published
- 13 Nov 2025
- Publisher
- Center on Reinventing Public Education
- Authors
- Bree Dusseault, Jared Hurwitz, Anagha Mandayam
- Type
- Policy brief
- Level
- K-12
- Topics
- AI Literacy, Curriculum & Standards, Educator Training, Instructional Use, Local Policy Requirement
Districts and AI: Early Adopters Focus More on Students in 2025-26Archived copy
The Center for Democracy & Technology surveyed nationally representative samples of high school students and middle and high school teachers and parents in summer 2025 about AI use in schools and related risks. Most respondents had used AI, and students and teachers in schools that use AI for more purposes were more likely to report risks such as data breaches, deepfake harassment, and troubling student interactions with chatbots.
Findings
- Eighty-five percent of teachers and 86 percent of students reported using AI during the 2024-25 school year.
“85 percent of teachers report they have used AI in at least one way in their classroom in the past school year (2024-25), and 86 percent of students say they have used AI for either school or personal use.”
AI Literacy section, p. 34 - Teachers and students reporting many school-related AI uses were more likely to report exposure to risks including data breaches, tech-enabled harassment, AI systems that do not work as intended, and troubling student-technology interactions.
“Teachers and students who report many school-related uses are also more likely to report exposure to risks like:”
Emerging Risks, p. 11 - Forty-two percent of students said they or a friend had used AI for mental health support, and 42 percent as a friend or companion, in the past school year.
“As a friend/companion 42%”
Table 3, p. 25 - Forty-eight percent of teachers and of students said they had received AI training or information from their school.
“Almost half of teachers and students report that they have received some training or information about AI from someone at their school, and most found it helpful.”
Figure 30, p. 35
Method
Online surveys of nationally representative samples of 1,030 9th-12th grade students, 806 6th-12th grade teachers, and 1,018 6th-12th grade parents, fielded June to August 2025 with quotas and weighting to national demographics.
- Published
- 8 Oct 2025
- Publisher
- Center for Democracy & Technology
- Authors
- Elizabeth Laird, Maddy Dwyer, Hannah Quay-de la Vallee
- Type
- Survey or poll
- Level
- K-12
- Topics
- AI Literacy, Instructional Use, Student Safety & Wellbeing, Student Privacy, Deepfakes & Synthetic Media
Hand in Hand: Schools’ Embrace of AI Connected to Increased Risks to StudentsArchived copyHand in Hand (full report PDF)Archived copy
RAND combined 2024-25 surveys of nationally representative samples of students, parents, teachers, principals, and district leaders to update the picture of AI in K-12 schools. Student and teacher AI use rose by 15 percentage points or more in one to two years, while training for teachers and students and school AI policies lagged behind.
Findings
- In 2025, 54 percent of students and 53 percent of ELA, math, and science teachers said they used AI for school, up more than 15 percentage points from surveys one to two years earlier.
“In 2025, 54 percent of students and 53 percent of English language arts, math, and science teachers indicated that they used AI for school.”
Key Takeaways - Parents and students were more likely than district leaders to agree that greater AI use will harm students' critical-thinking skills, and half of students worried about being falsely accused of cheating with AI.
“Sixty-one percent of parents, 48 percent of middle schoolers, 55 percent of high schoolers, and only 22 percent of district leaders agreed that greater use of AI will harm students’ critical-thinking skills.”
Key Takeaways - Thirty-five percent of district leaders reported providing students with AI training, and over 80 percent of students said teachers had not explicitly taught them how to use AI for schoolwork.
“As of spring 2025, 35 percent of district leaders reported that they provide students with training on AI.”
Key Takeaways - Forty-five percent of principals reported school or district AI policies or guidance, and 34 percent of teachers reported policies on AI and academic integrity.
“Forty-five percent of principals reported having school or district policies or guidance on the use of AI in schools, and 34 percent of teachers reported having school or district policies on the use of AI related to academic integrity.”
Key Takeaways
Method
Weighted analysis of eight nationally representative RAND panel samples collected October 2024 to May 2025: middle and high school students (n=1,261 and n=852, American Youth Panel), 984 parents of 12-17-year-olds, 967 teachers of all subjects and 8,601 ELA/math/science teachers (American Teacher Panel), 3,668 principals, and 289 and 232 district leaders; plus interviews with ten district leaders.
- Published
- 30 Sep 2025
- Publisher
- RAND Corporation
- Authors
- Christopher Joseph Doss, Robert Bozick, Heather L. Schwartz, Lisa Chu, Lydia R. Rainey, Ashley Woo, Justin Reich, Jesse Dukes
- Type
- Survey or poll
- Level
- K-12
- Topics
- Academic Integrity, AI Literacy, Educator Training, Instructional Use
AI Use in Schools Is Quickly Increasing but Guidance Lags Behind: Findings from the RAND Survey PanelsArchived copyAI Use in Schools Is Quickly Increasing but Guidance Lags Behind (PDF)Archived copy
Common Sense Media and NORC surveyed a nationally representative sample of U.S. teens ages 13 to 17 in spring 2025 about their use of AI companions such as Character.AI and Replika. Most teens had used AI companions and about half used them regularly; a third of users had discussed serious matters with an AI companion instead of a person, and a third had felt uncomfortable with something one said or did.
Findings
- Seventy-two percent of teens had used AI companions at least once, and 52 percent were regular users.
“Seventy-two percent of teens have used AI companions at least once, and over half (52%) qualify as regular users who interact with these platforms at least a few times a month.”
Key Findings, finding 1 - Thirty-three percent of AI companion users had chosen to discuss important or serious matters with an AI companion instead of real people.
“Among AI companion users, 33% have chosen to discuss important or serious matters with AI companions instead of real people”
Key Findings, finding 9 - Thirty-four percent of users reported feeling uncomfortable with something an AI companion had said or done.
“One-third of users experience discomfort with AI companions.”
Key Findings, finding 8
Method
Nationally representative survey of 1,060 U.S. teens ages 13-17 conducted by NORC at the University of Chicago, April 30 to May 14, 2025, combining the probability-based AmeriSpeak Teen Omnibus with a nonprobability sample.
- Published
- 16 Jul 2025
- Publisher
- Common Sense Media
- Authors
- Michael B. Robb, Supreet Mann
- Type
- Survey or poll
- Level
- K-12
- Topics
- Student Safety & Wellbeing, Student Privacy
Talk, Trust, and Trade-Offs: How and Why Teens Use AI CompanionsArchived copyTalk, Trust, and Trade-Offs (full report PDF)Archived copy
CRPE studied 27 "early adopter" school districts during the 2024-25 school year through focus groups, interviews, a survey, and roundtables to learn how they approached system-wide AI adoption. Most districts were still piloting AI in fragmented ways, mainly to save teachers time, and leaders cited data privacy, sustainable funding, limited capacity, and uneven state guidance as the main barriers.
Findings
- About 80% of the districts studied were classified as "Dabblers" or "Emerging Users," still in early, ad hoc stages of AI adoption.
“About 80% of the districts we studied fell into the first two categories of Dabblers and Emerging Users”
p. 4 - Data privacy and sustainable funding were each rated as considerable or extreme challenges by 35% of surveyed districts.
“Two challenges stood out most prominently: data privacy and sustainable funding, each cited by more than a third (35%) of respondents as posing “considerable” or “extreme” difficulty.”
p. 12 - Among surveyed districts in states that had not yet issued AI guidance, 80% called the lack of state guidance an extremely difficult barrier.
“80% of surveyed districts that operate in states without guidance (at the time of this survey) reported the lack of state guidance as an “extremely difficult” barrier”
p. 13
Method
Qualitative and survey study of 27 early adopter districts recruited from 51 identified districts; 22 focus groups and interviews with 45 leaders from 22 districts, survey responses from 17 districts, 10 follow-up group interviews in six districts, and three roundtables; not nationally representative.
- Published
- Jul 2025
- Publisher
- Center on Reinventing Public Education
- Authors
- Bree Dusseault, Maddy Sims, Michael Berardino
- Type
- Research report
- Level
- K-12
- Topics
- Educator Training, Funding & Grants, Local Policy Requirement, Procurement & Vendors, Student Privacy
AI Early Adopter Districts: The Promises and Challenges of Using AI to Transform EducationAI Early Adopter Districts: The Promises and Challenges of Using AI to Transform Education (report PDF)
A preregistered field experiment at a high school in Turkey gave nearly 1,000 math students either a standard GPT-4 chat interface, a GPT-4 tutor with teacher-designed safeguards, or no AI during practice sessions, then tested them without AI. Both AI versions raised practice scores, but students who used the unrestricted chatbot scored lower on the later unassisted exam than students who never had access, while the safeguarded tutor largely avoided that drop.
Findings
- Access to GPT-4 during practice raised practice-problem grades by 48% for the standard chat interface and 127% for the safeguarded tutor, relative to the control group.
“having GPT-4 access while solving problems significantly improves performance (48% improvement in grades for GPT Base and 127% for GPT Tutor)”
Abstract - When AI access was removed, students who had used the standard chat interface performed 17% worse on the exam than students who never had access; the safeguarded tutor's effect on the exam was statistically indistinguishable from the control group.
“on the exam, students in the GPT Base arm perform statistically significantly worse than students in the control arm by 17%; this negative effect is essentially eradicated in the GPT Tutor arm, though we still do not observe a positive effect.”
Main results - Message analysis indicated that most students in the standard chat arm used it to obtain answers, and students did not perceive the reduction in their own learning.
“we find evidence that students do not perceive any reduction in their learning or subsequent performance as a consequence of copying solutions”
Introduction
Method
Preregistered classroom-randomized controlled trial; about fifty 9th-11th grade math classes (nearly 1,000 students) at one high school in Turkey, fall 2023-24; three arms (control, GPT Base, GPT Tutor) over four 90-minute sessions.
- Published
- 25 Jun 2025
- Publisher
- Proceedings of the National Academy of Sciences
- Authors
- Hamsa Bastani, Osbert Bastani, Alp Sungu, Haosen Ge, Özge Kabakcı, Rei Mariman
- Type
- Peer-reviewed article
- Level
- K-12
- Topics
- Instructional Use, Research
Generative AI without guardrails can harm learning: Evidence from high school mathematics (PubMed Central full text)Archived copy
Education Commission of the States blog post describing state and district AI pilot programs in K-12 schools, including Connecticut's classroom pilot, Indiana's AI platform grant, Iowa's AI reading tutor rollout, Kentucky's early warning tool, and a New Mexico attendance pilot. It also notes equity concerns about access and about the accuracy of predictive early warning systems.
Findings
- At least five states had implemented or were developing AI pilot programs; Connecticut launched one in seven districts in spring 2025 for students in grades 7-12.
“At least five states have implemented or are in the process of developing pilot programs to foster integration and responsible use in the classroom.”
Source - Indiana's 2023-24 AI-Powered Platform Pilot Grant used $2 million in federal COVID relief funds, and 53% of participating teachers rated their experience positive or very positive.
“Of the teachers who participated, 53% indicated that their experience was either positive or very positive.”
Source - Iowa announced a $3 million investment to provide all elementary schools with an AI reading tutor tool.
“The Iowa Department of Education announced a $3 million investment to provide all Iowa elementary schools (public and nonpublic) with an AI reading tutor tool”
Source
Method
Descriptive review of publicly reported state and district AI pilot programs as of spring 2025.
- Published
- 24 Jun 2025
- Publisher
- Education Commission of the States
- Authors
- Sam Comai
- Type
- Policy brief
- Level
- K-12
- Topics
- Equity & Access, Funding & Grants, Instructional Use
Gallup and the Walton Family Foundation surveyed U.S. public K-12 teachers in spring 2025 about how often they use AI tools, for which tasks, and how much time it saves. Six in 10 teachers used AI for work, weekly users estimated saving about six hours a week, and teachers in schools with an AI policy reported more use and larger time savings.
Findings
- Six in 10 teachers used an AI tool for their work in 2024-25, and 32 percent used AI at least weekly.
“In the 2024-25 school year, six in 10 teachers reported using an AI tool for their work.”
Source - Teachers who used AI at least weekly estimated saving 5.9 hours per week, equivalent to about six weeks over a school year.
“Teachers who use AI tools at least weekly estimate they save 5.9 hours per week, on average. Over the course of a 37.4-week school year, these time savings add up to the equivalent of six weeks per school year .”
Regular Users Earn an Estimated “AI Dividend” of 5.9 Hours per Week - Only 19 percent of teachers worked at schools with an AI policy; those schools saw higher AI use and a 26 percent larger average time savings per teacher.
“schools with an AI policy are earning an AI dividend that is 26% greater than schools without an AI policy (2.3 vs. 1.7 hours saved per week per teacher).”
p. 17 - Most teachers (68 percent) had not received any school- or district-provided AI training.
“most teachers (68%) have not engaged in any training provided by their school or district on how to use AI or AI tools.”
p. 18
Method
Web survey of 2,232 U.S. public K-12 teachers recruited from the RAND American Teacher Panel, March 18 to April 11, 2025, weighted to national teacher and school demographics; margin of error ±2.5 points.
- Published
- 24 Jun 2025
- Publisher
- Gallup
- Authors
- Gallup, Walton Family Foundation
- Type
- Survey or poll
- Level
- K-12
- Topics
- Educator Training, Instructional Use
Three in 10 Teachers Use AI Weekly, Saving Six Weeks a YearArchived copyTeaching for Tomorrow: Unlocking Six Weeks a Year With AI (report PDF)Archived copy
Education Commission of the States blog post reviewing state AI-in-education policy through spring 2025, including K-12 guidance, legislation, oversight and sandbox proposals, and state task force reports. It finds that at least 28 states had published K-12 AI guidance and that many states had introduced AI education bills, although only one had been enacted in 2025 as of April 1.
Findings
- At least 28 states had published guidance on AI in K-12 settings as of April 2025.
“Fast forward to April 2025, and at least 28 states have published guidance on AI in K-12 settings.”
Source - As of April 1, 2025, one AI-related education bill had been enacted that year (Mississippi S.B. 2426, creating a state AI task force), while at least 20 states had introduced AI education bills.
“As of April 1, 2025, only one AI-related education bill has been enacted this year: Mississippi ’s S.B. 2426 , which creates a state AI task force.”
Source - Task force reports from Arkansas, Georgia, and Illinois shared priorities including AI literacy frameworks, educator professional development, equitable access, student data protection, and implementation support.
“Task force reports from Arkansas ( 2025 ), Georgia ( 2024 ) and Illinois ( 2024 ) highlight similar priorities like creating curricular frameworks for AI literacy, investing in educator professional development”
Source
Method
Review of state legislation, guidance documents, and task force reports through April 2025.
- Published
- 17 Jun 2025
- Publisher
- Education Commission of the States
- Authors
- Jenny McCann
- Type
- Policy brief
- Level
- K-12
- Topics
- AI Literacy, Curriculum & Standards, Local Policy Requirement, Task Forces & Study
How States Are Responding to the Rise of AI in EducationArchived copy
Annual national surveys of administrators, instructors, and students at US colleges, focused on introductory and developmental courses, covering teaching modality, student engagement, digital learning tools, and generative AI. Weekly or daily generative AI use rose sharply among all three groups since spring 2023, while preventing student cheating ranked as a top instructional challenge.
Findings
- Daily or weekly personal use of generative AI reached 40% of administrators, 30% of instructors, and 42% of students in spring 2025, up from 2%, 4%, and 14% in spring 2023.
“daily or weekly use up to 40% among administrators (from 2% in spring 2023), 30% among instructors (from 4% in spring 2023), and 42% among students in spring 2025 (from 14% in spring 2023)”
p. 6, Key findings - Preventing student cheating was the top instructional challenge reported by instructors of introductory or developmental courses (45%).
“Preventing student cheating and increasing student attendance are the top instructional challenges among instructors teaching introductory or developmental courses (45% and 44%, respectively)”
p. 6, Key findings - Instructor preference for face-to-face teaching rose from 55% in 2023 to 64% in 2025.
“Instructor preference for in-person teaching jumped to 64%, up from 55% in 2023.”
Source
Method
National surveys of administrators (n=312), instructors (n=1,572), and students (n=1,528) at more than 900 US colleges and universities, spring 2025; data weighted by institutional sector.
- Published
- 11 Jun 2025
- Publisher
- Tyton Partners
- Authors
- Catherine Shaw, Hadley Dorn, Gates Bryant
- Type
- Survey or poll
- Level
- Higher ed
- Topics
- Academic Integrity, Instructional Use
Time for Class 2025: Empowering Educators, Engaging StudentsArchived copyTime for Class 2025 (full report PDF)Archived copy
An MIT Media Lab study assigned 54 participants from five Boston-area universities to write essays over several sessions using ChatGPT, a search engine, or no tools, while recording brain activity with EEG; in a fourth session some participants switched conditions. Participants using the language model showed the weakest brain connectivity, the lowest sense of ownership of their essays, and the most difficulty quoting their own work, which the authors interpret as a possible cognitive cost of relying on AI for writing.
Findings
- EEG measures showed the strongest and most distributed brain networks in the no-tools group, moderate engagement for search engine users, and the weakest connectivity for language model users.
“EEG revealed significant differences in brain connectivity: Brain-only participants exhibited the strongest, most distributed networks; Search Engine users showed moderate engagement; and LLM users displayed the weakest connectivity.”
Abstract - In the first session, 83.3% of participants in the language model group could not correctly quote from the essay they had just written, compared with 11.1% in each of the other groups.
“In the LLM ‑ assisted group, 83.3 % of participants (15/18) failed to provide a correct quotation, whereas only 11.1 % (2/18) in both the Search ‑ Engine and Brain ‑ Only groups encountered the same difficulty.”
Results, Session 1 interview, Question 3 (Ability to Quote) - Self-reported ownership of essays was lowest in the language model group and highest in the no-tools group.
“Self-reported ownership of essays was the lowest in the LLM group and the highest in the Brain-only group.”
Abstract
Method
Laboratory experiment with three assigned conditions (LLM, search engine, no tools) over three sessions plus a crossover fourth session; 54 participants aged 18-39 from MIT, Wellesley, Harvard, Tufts, and Northeastern (18 completed session 4); EEG, NLP analysis of essays, teacher and AI scoring, and interviews.
- Published
- 10 Jun 2025
- Publisher
- arXiv
- Authors
- Nataliya Kosmyna, Eugene Hauptmann, Ye Tong Yuan, Jessica Situ, Xian-Hao Liao, Ashly Vivian Beresnitzky, Iris Braunstein, Pattie Maes
- Type
- Research report
- Level
- Higher ed
- Covers
- Massachusetts
- Topics
- Academic Integrity, Instructional Use, Research
Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing TaskYour Brain on ChatGPT (full text PDF, v2)
A randomized crossover experiment in a large introductory physics course at Harvard University compared lessons delivered through a custom GPT-4 tutor, built around pedagogical best practices and pre-written solutions, with the same lessons taught through in-class active learning. Students using the AI tutor at home scored higher on post-tests, spent less time on task on average, and reported feeling more engaged and motivated.
Findings
- Median learning gains for students in the AI-tutored condition were more than double those for students in the in-class active learning condition.
“The median learning gains for students, relative to the pre-test baseline (M = 2.75, N = 316), in the AI-tutored group were over double”
Results, Learning gains - Controlling for prior knowledge and other factors, the estimated effect size was 0.63 in linear regression and 0.73 to 1.3 standard deviations in a quantile regression that accounts for a ceiling effect.
“While the linear regression suggests an effect size of 0.63, this is an underestimation due to ceiling effect”
Results, regression analysis - The median time on task with the AI tutor was 49 minutes, compared with about 60 minutes of in-class learning time, and 70% of AI-group students spent under 60 minutes.
“70% of students in the AI group spent less than 60 minutes on task, while 30% spent more than 60 minutes on task. The median time on task for students in the AI group was 49 minutes.”
Results, Time on task - Students in the AI condition reported higher engagement and motivation than students in the in-class condition.Results, students' perceptions of learning
Method
Randomized controlled crossover trial; 194 eligible students in an introductory physics course for the life sciences at Harvard University, fall 2023; two lessons (surface tension, fluid flow) with pre- and post-tests.
- Published
- 3 Jun 2025
- Publisher
- Scientific Reports
- Authors
- Greg Kestin, Kelly Miller, Anna Klales, Timothy Milbourne, Gregorio Ponti
- Type
- Peer-reviewed article
- Level
- Higher ed
- Covers
- Massachusetts
- Topics
- Instructional Use, Research
AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting (PubMed Central full text)Archived copy
A World Bank policy research working paper reports a randomized controlled trial of a six-week after-school program in Benin City, Nigeria, in which first-year senior secondary students used Microsoft Copilot (powered by GPT-4) with teacher guidance to study English. Students selected for the program scored higher on an end-of-program assessment and on their regular end-of-term English exam, with larger effects for students who attended more sessions.
Findings
- Students selected for the program scored 0.31 standard deviations higher on the end-of-intervention assessment, with an effect of 0.23 standard deviations on English, the main outcome.
“The intervention demonstrated a significant improvement of 0.31 standard deviation on an assessment that included English topics aligned with the Nigerian curriculum, knowledge of artificial intelligence and digital skills.”
Abstract - Being selected for the program raised performance on the regular third-term English curricular exam by 0.21 standard deviations.
“We calculate an ITT effect of being selected for the program on the performance in the third-term exam of 0.21 standard deviations.”
Introduction, p. 2 - Effects were positive across baseline ability levels but largest for female students and for students with higher prior performance; each additional day attended was associated with about 0.031 standard deviations of improvement.
“with an effect size of approximately 0.031 standard deviation per additional day of attendance”
Introduction, p. 3
Method
Randomized controlled trial; first-year senior secondary students from nine public schools in Benin City, Nigeria, randomly selected from volunteers (657 assigned to treatment, 671 to control; 759 completed the final assessment); twelve 90-minute after-school sessions over six weeks; intent-to-treat and attendance-based estimates.
- Published
- 19 May 2025
- Publisher
- World Bank
- Authors
- Martin De Simone, Federico Tiberti, Maria Barron Rodriguez, Federico Manolio, Wuraola Mosuro, Eliot Jolomi Dikoru
- Type
- Research report
- Level
- K-12
- Topics
- AI Literacy, Instructional Use, Research
RAND compared nationally representative district surveys from fall 2023 and fall 2024 and interviewed district leaders to track how many districts train teachers on generative AI. The share of districts reporting such training roughly doubled to about half, but low-poverty districts remained well ahead of high-poverty districts.
Findings
- Forty-eight percent of districts reported having trained teachers on AI by fall 2024, up from 23 percent in fall 2023, and another 26 percent planned to do so during 2024-25.
“In fall 2024, 48 percent of districts reported that they had trained teachers on the use of AI, which is a substantial, 25-percentage-point increase from the percentage of districts that reported implementing training on AI in fall 2023.”
Survey Results - By fall 2024, 67 percent of low-poverty districts had trained teachers on AI, compared with 42 percent of middle-poverty and 39 percent of high-poverty districts.
“By fall 2024, 67 percent of low-poverty districts reported having provided training for teachers on AI use, compared with 42 percent of middle-poverty districts and 39 percent of high-poverty districts.”
Survey Results - District leaders said initial trainings focused on addressing teachers' fears and discomfort with AI rather than on instructional tools.
“In interviews, district leaders said that they focused initial teacher trainings on addressing teachers' fear and discomfort with AI rather than jumping right into instructional tools.”
Key Findings
Method
Weighted, nationally representative surveys of K-12 public school districts in the American School District Panel (fall 2023, n=224; fall 2024, n=290) plus interviews with 14 district leaders.
- Published
- 8 Apr 2025
- Publisher
- RAND Corporation
- Authors
- Melissa Kay Diliberti, Robin J. Lake, Steven R. Weiner
- Type
- Survey or poll
- Level
- K-12
- Topics
- Educator Training, Equity & Access
More Districts Are Training Teachers on Artificial Intelligence: Findings from the American School District PanelArchived copy
Gallup, the Walton Family Foundation, and GSV Ventures surveyed nearly 3,500 Americans ages 13 to 28 in March 2025 about their use of and feelings toward generative AI, including at school and work. Most used generative AI, more said it would harm than help their critical thinking, and only about a quarter of students said their school explicitly permits AI while about half said their school lacks clear AI policies.
Findings
- Seventy-nine percent of Gen Z respondents used generative AI products, including 47 percent weekly, but more said the tools made them anxious than hopeful or excited.
“Most Gen Zers (79%) say they use generative artificial intelligence products, including 47% who do so on a weekly basis”
Key findings, 01 - Gen Z respondents were more than twice as likely to say AI will harm rather than help their critical thinking skills (49 percent to 22 percent).
“Gen Zers are also more than twice as likely to say AI will harm, rather than help, their critical thinking skills (49% to 22%).”
Key findings, 02 - About half of students said schools should permit AI and teach students to use it, but 28 percent said their school explicitly permits AI and 49 percent said it has no clear AI policies.
“just 28% of students say their school explicitly permits the use of artificial intelligence, while nearly half (49%) say their school does not have clear policies regarding AI.”
Key findings, 03
Method
Gallup Panel web survey, March 6-13, 2025, of 3,465 people ages 13-28 in all 50 states and DC, including 1,460 children ages 13-18 reached through panel households; margin of error ±2.4 points overall, ±3.3 for K-12 students.
- Published
- 1 Apr 2025
- Publisher
- Gallup
- Authors
- Gallup, Walton Family Foundation, GSV Ventures
- Type
- Survey or poll
- Level
- K-12
- Topics
- AI Literacy, Instructional Use
Voices of Gen Z: How American Youth View and Use Artificial IntelligenceArchived copyVoices of Gen Z: How American Youth View and Use Artificial Intelligence (report PDF)
The education chapter of Stanford HAI's annual AI Index, prepared with the Kapor Foundation, the Computer Science Teachers Association, and the Public Interest Technology University Network, compiles data on K-12 computer science and AI education and on computing degrees. It reports that most U.S. computer science teachers think AI belongs in foundational CS education but fewer than half of high school CS teachers feel equipped to teach it.
Findings
- Although 81% of U.S. computer science teachers agreed that using and learning about AI should be part of foundational CS education, fewer than half of high school CS teachers felt equipped to teach AI.
“Despite 81% of CS teachers agreeing that using AI and learning about AI should be included in a foundational CS learning experience, less than half of high school CS teachers felt equipped to teach AI.”
Chapter 7 Highlights, p. 368 - Access to and enrollment in high school CS courses in the U.S. rose slightly from the prior school year, with gaps by state, race and ethnicity, school size, geography, income, gender, and disability.
“Student participation varies by state, race/ethnicity, school size, geography, income, gender, and disability.”
Chapter 7 Highlights, p. 368 - The number of U.S. graduates earning master's degrees in AI nearly doubled between 2022 and 2023.
“Graduates who earned their master’s degree in AI in the U.S. nearly doubled between 2022 and 2023.”
Chapter 7 Highlights, p. 368
Method
Compilation of secondary data, including Code.org, CSTA, and ECEP Alliance K-12 computer science data, the 2024 Computer Science Teacher Landscape Survey, and NCES IPEDS degree completion data.
- Published
- Apr 2025
- Publisher
- Stanford Institute for Human-Centered Artificial Intelligence
- Authors
- Nestor Maslej, Loredana Fattorini, Raymond Perrault, Yolanda Gil, Vanessa Parli, Njenga Kariuki, Emily Capstick, Anka Reuel, Erik Brynjolfsson, John Etchemendy, Katrina Ligett, Terah Lyons, James Manyika, Juan Carlos Niebles, Yoav Shoham, Russell Wald, Toby Walsh, Armin Hamrah, Lapo Santarlasci, Julia Betts Lotufo, Alexandra Rome, Andrew Shi, Sukrut Oak
- Type
- Research report
- Level
- K-12, Higher ed
- Topics
- AI Literacy, Curriculum & Standards, Educator Training, Workforce & Careers
The 2025 AI Index ReportArchived copyThe AI Index 2025 Annual Report (full PDF)Archived copy
A survey of 788 higher education staff, faculty, and leaders in November 2024 on institutional AI strategy, policies, use cases, workforce, and differences between smaller and larger institutions. More respondents than a year earlier called AI a strategic priority and reported AI acceptable-use policies, while few said their cybersecurity and privacy policies were adequate for AI risks and smaller institutions reported fewer resources for AI adoption.
Findings
- The share of respondents reporting institutional AI-related acceptable use policies rose from 23% to 39% in one year.
“The proportion of respondents reporting that their institution has AI-related AUPs increased from 23% last year to 39% this year”
Key Findings, Policies and Guidelines - Only 9% said their institution's cybersecurity and privacy policies were adequate for AI-related risks.
“Only 9% of respondents reported that their institution's cybersecurity and privacy policies are adequate for addressing AI-related risks to the institution.”
Key Findings, Policies and Guidelines - Teaching and learning was the area most focused on AI, especially academic integrity (74%), and 68% said students use AI more than faculty.
“Teaching and learning is the functional area at the institution most focused on using AI, with particular focus on the areas of academic integrity (74%), coursework (65%), assessment practices (54%), and curriculum design (54%).”
Key Findings, Use Cases
Method
Online survey of EDUCAUSE community members distributed November 4-19, 2024; 788 responses from people working at higher education institutions.
- Published
- 17 Feb 2025
- Publisher
- EDUCAUSE
- Authors
- Jenay Robert, Mark McCormack
- Type
- Survey or poll
- Level
- Higher ed
- Topics
- Academic Integrity, Data Governance, Educator Training, Equity & Access, Local Policy Requirement
2025 EDUCAUSE AI Landscape Study: Into the Digital AI DivideArchived copy2025 EDUCAUSE AI Landscape Study, Introduction and Key FindingsArchived copy2025 EDUCAUSE AI Landscape Study, Methodology and AcknowledgmentsArchived copy
Using spring 2024 surveys from the RAND American Educator Panels, the report examines how many K-12 teachers and principals used AI tools and whether schools provided AI guidance. One quarter of teachers and nearly 60 percent of principals reported using AI, with lower use and less guidance in higher-poverty schools.
Findings
- One quarter of ELA, math, and science teachers used AI tools for instructional planning or teaching in 2023-24, and nearly 60 percent of principals used AI tools for their work.
“one-quarter of ELA, math, and science teachers used AI tools for instructional planning or teaching in the 2023–2024 school year. Nearly 60 percent of surveyed American School Leader Panel principals also reported using AI tools for their work in 2023–2024.”
Key Takeaways - Almost 40 percent of ELA or science teachers reported using AI, compared with 20 percent of general elementary or math teachers.
“almost 40 percent of ELA or science teachers reported using AI compared with 20 percent of general elementary education or math teachers.”
Key Takeaways - Eighteen percent of principals said their school or district provided AI guidance; principals in the highest-poverty schools were about half as likely as those in the lowest-poverty schools to report it.
“Eighteen percent of principals reported that their schools or districts provided guidance on the use of AI by staff, teachers, or students.”
Key Takeaways
Method
Weighted, nationally representative surveys: spring 2024 American Instructional Resources Survey of 9,126 K-12 ELA, math, and science teachers (American Teacher Panel) and spring 2024 American Mathematics Educator Study survey of 3,631 principals (American School Leader Panel).
- Published
- 11 Feb 2025
- Publisher
- RAND Corporation
- Authors
- Julia H. Kaufman, Ashley Woo, Joshua Eagan, Sabrina Lee, Emma B. Kassan
- Type
- Survey or poll
- Level
- K-12
- Topics
- Equity & Access, Institutional Operations, Instructional Use
Uneven Adoption of Artificial Intelligence Tools Among U.S. Teachers and Principals in the 2023–2024 School YearArchived copyUneven Adoption of Artificial Intelligence Tools Among U.S. Teachers and Principals in the 2023–2024 School Year (PDF)Archived copy
A survey of 1,681 faculty at 52 higher education institutions in 28 countries on their use of AI, readiness, and views of institutional support. Most faculty expected to use AI in teaching and worried about students' ability to evaluate AI output, while few said their institutions' guidelines or training resources were sufficient.
Findings
- 83% of faculty were concerned about students' ability to critically evaluate AI-generated output.
“The survey revealed widespread concerns about students' ability to critically evaluate AI-generated outputs, with 83% of faculty members expressing concern.”
Faculty Concerns - 54% of faculty said current student evaluation methods are no longer adequate, with 13% calling for a complete revamp.
“Over half of faculty (54%) believe current student evaluation methods are no longer adequate in the age of AI, with 13% calling for an urgent, complete revamp.”
Source - Only 4% of faculty were fully aware of their institution's AI guidelines and found them comprehensive, and 6% fully agreed their institution provided sufficient AI literacy training resources.
“only 4% of faculty are fully aware of their institutional AI guidelines and feel they are fully comprehensive. Just 6% fully agree that their institutions have provided sufficient resources to train faculty's AI literacy.”
Source
Method
Survey of 1,681 faculty members at 52 higher education institutions in 28 countries.
- Published
- 20 Jan 2025
- Publisher
- Digital Education Council
- Authors
- Digital Education Council
- Type
- Survey or poll
- Level
- Higher ed
- Topics
- Assessment, Educator Training, Instructional Use
What Faculty Want: Key Results from the Global AI Faculty Survey 2025Archived copy
Pew Research Center surveyed U.S. teens ages 13 to 17 in fall 2024 about whether they use ChatGPT for schoolwork and which uses they consider acceptable. The share using it for schoolwork doubled from 2023 to 26 percent, and teens were more accepting of using it to research topics than to write essays.
Findings
- Twenty-six percent of teens said they had used ChatGPT for schoolwork, up from 13 percent in 2023.
“The share of teens who say they use ChatGPT for their schoolwork has risen to 26%, according to a Pew Research Center survey of U.S. teens ages 13 to 17. That’s up from 13% in 2023.”
Source - Black and Hispanic teens (31 percent each) were more likely than White teens (22 percent) to report using ChatGPT for schoolwork.
“Black and Hispanic teens (31% each) are more likely than White teens (22%) to say they have used ChatGPT for their schoolwork.”
By race and ethnicity - Fifty-four percent of teens said using ChatGPT to research new topics is acceptable, compared with 18 percent for writing essays.
“18% say it’s acceptable to use ChatGPT to write essays, and 42% say it’s not acceptable.”
Source
Method
Survey of 1,391 U.S. teens ages 13-17, Sept. 18 to Oct. 10, 2024, recruited through Ipsos KnowledgePanel parents; weighted to be representative of U.S. teens living with their parents.
- Published
- 15 Jan 2025
- Publisher
- Pew Research Center
- Authors
- Olivia Sidoti, Eugenie Park, Jeffrey Gottfried
- Type
- Survey or poll
- Level
- K-12
- Topics
- Academic Integrity, Equity & Access, Instructional Use
About a quarter of U.S. teens have used ChatGPT for schoolwork – double the share in 2023Archived copy
A committee report based on a December 2024 survey of about 500 randomly sampled AAUP members on how AI and other education technology are affecting faculty work. Respondents reported that AI initiatives are largely led by administrators with little faculty input and that AI has worsened several aspects of their work; the report recommends shared governance over technology procurement and use.
Findings
- 71% of respondents said AI decision-making and initiatives are overwhelmingly led by college or university administrations.
“Seventy-one percent of respondents said decision-making and AI initiatives are overwhelmingly led by college or university administrations”
Findings, concern 2 - 91% of respondents were at least somewhat concerned about preventing academic dishonesty enabled by generative AI.
“Respondents were overwhelmingly concerned with student plagiarism made possible by generative AI. Ninety-one percent noted that they were at least somewhat concerned about preventing academic dishonesty.”
Findings, concern 3 - Respondents said AI had led to at least somewhat worse outcomes for the teaching environment (62%), job enthusiasm (76%), and student success (69%).
“AI has generally led to at least somewhat worse outcomes for the teaching environment (according to 62 percent of respondents), pay equity (30 percent), job enthusiasm (76 percent), academic freedom (40 percent), and student success (69 percent).”
Findings
Method
Online survey of AAUP members in December 2024; 5,000 members randomly selected from the active membership list, about 500 responses over two weeks; descriptive statistics and thematic coding of open-ended items. Results are not representative of all US academic workers.
- Published
- 2025
- Publisher
- American Association of University Professors
- Authors
- AAUP Ad Hoc Committee on Artificial Intelligence and Academic Professions
- Type
- Research report
- Level
- Higher ed
- Topics
- Academic Integrity, Staff Employment & Evaluation, Institutional Operations, Procurement & Vendors
Artificial Intelligence and Academic ProfessionsArchived copy
2024
A randomized lab experiment with 117 university students compared support from ChatGPT, a human expert, writing analytics tools, or no extra tool during an essay revision task, tracking motivation, self-regulated learning processes, and performance. The ChatGPT group improved essay scores the most, but did not differ in intrinsic motivation, knowledge gain, or knowledge transfer, and the authors report that reliance on ChatGPT may encourage "metacognitive laziness."
Findings
- Students supported by ChatGPT showed the largest essay score improvement, but their knowledge gain and transfer were not significantly different from other groups.
“ChatGPT group outperformed in the essay score improvement but their knowledge gain and transfer were not significantly different.”
Abstract, Results - Groups with different kinds of support did not differ in post-task intrinsic motivation, but showed different patterns of self-regulated learning processes.
“learners who received different learning support showed no difference in post-task intrinsic motivation; 2) there were significant differences in the frequency and sequences of the self-regulated learning processes among groups”
Abstract, Results - The authors conclude that tools such as ChatGPT may promote dependence on technology and hinder learners' self-regulation.Practitioner Notes, What this paper adds
Method
Randomized laboratory experiment; 117 university students assigned to four support conditions (ChatGPT, human expert, writing analytics checklist tools, no extra tool) on a writing task; multi-channel learning process, performance, and motivation data.
- Published
- 10 Dec 2024
- Publisher
- British Journal of Educational Technology
- Authors
- Yizhou Fan, Luzhen Tang, Huixiao Le, Kejie Shen, Shufang Tan, Yueying Zhao, Yuan Shen, Xinyu Li, Dragan Gašević
- Type
- Peer-reviewed article
- Level
- Higher ed
- Topics
- Instructional Use, Research
CRPE built a database of 40 "early adopter" school districts and charter networks with system-wide AI strategies, compiled from public information, and sorted their approaches into guidance, tools for students and staff, changes to systems and structures, and back-office uses. About two-thirds had published AI guidance, and most were pursuing only a few of the eight strategies tracked.
Findings
- Of the 40 early adopter systems, 26 (65%) had published guidance on AI use for staff, students, and families, leaving a third pursuing AI strategies without formal guidance.
“Of the 40 school systems we studied, 26 (65%) have provided publicly available guidance to staff, students, and families on using AI.”
Beyond tools section - The most common strategies were AI tools for teachers, students, and parents (70%, 58%, and 38%) and professional development (63%); new AI courses and learning standards were less common (30% and 8%).
“The most common strategies include using AI-enabled tools for key stakeholders (teachers, students, and parents at 70%, 58%, and 38%, respectively) and providing professional development opportunities (63%).”
Beyond tools section - The 40 systems endorsed or partnered with 65 tools, most often MagicSchool, SchoolAI, and Writable; six systems had comprehensive approaches with five or more strategies.
“These 40 school systems collectively endorse or partner with 65 tools”
Beyond tools section
Method
Database of 40 early adopter school districts and charter management organizations compiled from publicly available news stories and referrals from AI support organizations; not a representative sample.
- Published
- 24 Oct 2024
- Publisher
- Center on Reinventing Public Education
- Authors
- Bree Dusseault, Jared Hurwitz, Michael Berardino, Nadja Michel-Herf
- Type
- Policy brief
- Level
- K-12
- Topics
- AI Literacy, Educator Training, Instructional Use, Local Policy Requirement
Districts and AI: Tracking Early Adopters and Implications for the 2024-25 School YearArchived copy
Stanford researchers ran a preregistered randomized controlled trial in which virtual tutors working with elementary and middle school students in a large southern U.S. school district were randomly given access to Tutor CoPilot, a large language model tool that suggests expert-like responses in real time. Students whose tutors had the tool were more likely to master lesson topics, with the largest gains for students of lower-rated or less experienced tutors, at an estimated cost of about $20 per tutor per year.
Findings
- Students whose tutors used Tutor CoPilot were 4 percentage points more likely to master mathematics topics, as measured by session exit tickets.
“students whose tutors used Tutor CoPilot were 4 percentage points more likely to master mathematics topics (p<.01)”
Abstract - Gains were largest for students of lower-rated tutors (9 percentage points) and less experienced tutors (7 percentage points).
“lower-rated tutors experienced a 9 p.p. increase in students passing their exit ticket (56%→ 65%student passing rate from control to treatment). In Figure 2b, less experienced tutors showed a 7 p.p. increase (61%→ 68%).”
Results, RQ2 - Analysis of more than 350,000 messages showed tutors with the tool asked more probing questions and gave less generic praise; estimated annual cost was about $20 per tutor.
“Analysis of more than 350,000 messages indicates that Tutor CoPilot promotes effective pedagogy, increasing probing questions and reducing generic praise.”
Abstract
Method
Preregistered randomized controlled trial with tutors as the unit of randomization; 900 tutors initially assigned (783 active in the study) and 1,013 students in grades 3-6 from nine schools in one large southern U.S. district (79% Hispanic); outcome is session exit-ticket mastery.
- Published
- 3 Oct 2024
- Publisher
- arXiv
- Authors
- Rose E. Wang, Ana T. Ribeiro, Carly D. Robinson, Susanna Loeb, Dora Demszky
- Type
- Research report
- Level
- K-12
- Topics
- Educator Training, Equity & Access, Instructional Use, Research
Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise (arXiv v3 PDF)Archived copyTutor CoPilot: A Human-AI Approach for Scaling Real-Time ExpertiseArchived copy
CRPE surveyed and interviewed leaders of U.S. schools of education about how teacher preparation programs are responding to AI. Most programs were in the early stages, AI instruction for future teachers centered on preventing plagiarism, and few leaders said their faculty were confident using AI.
Findings
- About 59% of respondents said their programs offered some AI-related instruction to preservice teachers, mostly within existing coursework and focused on plagiarism prevention; about 25% trained future teachers on new ways of teaching with AI.
“Only about 25% of programs surveyed provide training on ways that AI can support new ways of teaching”
p. 3 - Only 10% of education school leaders reported that their faculty were confident using AI.
“only 10% of the education school leaders surveyed reported that their faculty members are confident in using AI, and most do not use it in their instruction.”
p. 4 - More than two-thirds of respondents said their institutions had no policies on AI tool use.
“More than two-thirds of survey respondents reported that their institutions have no policies around AI tool use”
p. 3
Method
Survey and interviews of U.S. education school leaders; more than 500 contacted, 14% response rate.
- Published
- Oct 2024
- Publisher
- Center on Reinventing Public Education
- Authors
- Steven Weiner, Robin Lake, Jessica Rosner
- Type
- Survey or poll
- Level
- K-12, Higher ed
- Topics
- Academic Integrity, AI Literacy, Educator Training
AI Is Evolving, but Teacher Prep Is Lagging: A First Look at Teacher Preparation Program Responses to AIArchived copyAI is Evolving, but Teacher Prep is Lagging (report PDF)Archived copy
Common Sense Media surveyed a nationally representative sample of U.S. teens ages 13 to 18 and their parents in spring 2024 about generative AI use at home and school. Seven in 10 teens had used a generative AI tool, many were unsure of their school's AI rules, and Black teens were more likely than peers to report having schoolwork wrongly flagged as AI-generated.
Findings
- Seventy percent of teens ages 13 to 18 had used at least one type of generative AI tool.
“while 70% of teens say they have used at least one type of gen AI tool”
Key Findings, p. 4 - Thirty-seven percent of teens were not sure whether their school had generative AI rules, 35 percent said it did, and 27 percent said it had none.
“37% of teens are not sure whether their school has gen AI rules.”
Key Findings, finding 6 - Ten percent of teens said a teacher had flagged their work as AI-generated when it was not; Black teens reported this at about twice the rate of White or Latino teens.
“Black teens are more than twice as likely as White or Latino teens to say that teachers flagged their schoolwork as being created by generative AI when it was not (20% vs. 7% and 10%, respectively).”
Key Findings, finding 7, p. 10 - Eighty-three percent of parents said their child's school had not communicated with them about generative AI.
“The large majority of parents of teens (83%) say that their child's school has not communicated with them about”
Key Findings, finding 6
Method
Nationally representative paired survey of 1,045 U.S. parents and their teens ages 13-18, with oversamples of Black and Latino teens, collected online in English or Spanish by Ipsos, March to May 2024.
- Published
- 18 Sep 2024
- Publisher
- Common Sense Media
- Authors
- Mary Madden, Angela Calvin, Alexa Hasse, Amanda Lenhart
- Type
- Survey or poll
- Level
- K-12
- Topics
- Academic Integrity, Equity & Access, Instructional Use
The Dawn of the AI Era: Teens, Parents, and the Adoption of Generative AI at Home and SchoolArchived copyThe Dawn of the AI Era (full report PDF)Archived copy
Researchers generated short texts with GPT-4, Claude 2, and Bard, altered them with six simple evasion techniques, and ran them along with human-written controls through seven AI text detectors. Detectors were already inaccurate on unaltered AI text, became less accurate on altered text, and misjudged some human-written samples, leading the authors to conclude the tools should not be used to decide academic integrity violations.
Findings
- Mean detector accuracy was 39.5% on unaltered AI-generated text, and only 67% of tests on human-written control samples were accurate.
“39.5% when evaluating unmanipulated AI-generated content. Importantly, regarding the human-written control samples, only 67% of the tests were accurate”
Results, baseline testing - Applying adversarial techniques such as adding spelling errors or increasing burstiness reduced average detector accuracy by 17.4 percentage points, with Turnitin showing the largest drop.
“Results show significant reductions in detector accuracy (17.4%) when faced with simple techniques to manipulate the AI generated content.”
Abstract - The authors conclude detectors cannot currently be recommended for determining academic integrity violations, though they may support learning when used non-punitively.
“they cannot currently be recommended for determining academic integrity violations due to accuracy limitations and the potential for false accusation which undermines inclusive and fair assessment practices.”
Abstract
Method
Experimental test in September-October 2023; 15 AI-generated samples, 89 adversarially altered versions, and 10 human-written controls (114 samples) run through seven AI text detectors for 805 tests.
- Published
- 8 Sep 2024
- Publisher
- International Journal of Educational Technology in Higher Education
- Authors
- Mike Perkins, Jasper Roe, Binh H. Vu, Darius Postma, Don Hickerson, James McGaughran, Huy Q. Khuat
- Type
- Peer-reviewed article
- Level
- Higher ed
- Topics
- Academic Integrity, Equity & Access
Simple techniques to bypass GenAI text detectors: implications for inclusive education (published version, James Cook University repository)GenAI Detection Tools, Adversarial Techniques and Implications for Inclusivity in Higher Education (preprint)Archived copy
Researchers observed 21 students in an introductory programming course at a small U.S. research university as they solved a programming problem with generative AI tools, using observation, interviews, and eye tracking. Nearly all students finished, but students who were already doing well used the tools to accelerate, while struggling students had their existing metacognitive difficulties compounded and often believed they had performed better than they had.
Findings
- Although 20 of 21 students completed the programming problem, students split between those who used generative AI to accelerate and those who struggled.
“Although 20 of 21 students completed the assigned programming problem, our findings show an unfortunate divide in the use of GenAI tools between students who accelerated and students who struggled.”
Abstract - For struggling students, known metacognitive difficulties persisted and were compounded by generative AI, and many finished with an illusion of competence.
“struggling students often expressed cognitive dissonance about their problem solving ability, thought they performed better than they did, and finished with an illusion of competence.”
Abstract - Students with better grades and higher self-efficacy were more likely to use the tools to accelerate, which the authors suggest could widen the digital divide.
“Students with better grades and higher self-efficacy were more likely to use GenAI tools to accelerate towards a solution”
Section 1, Introduction
Method
Qualitative lab study replicating earlier work; 21 of 27 students in one CS1 course at a small U.S. research university; 21 lab sessions with participant observation, interviews, and eye tracking.
- Published
- 6 Aug 2024
- Publisher
- ACM Conference on International Computing Education Research (ICER 2024)
- Authors
- James Prather, Brent N. Reeves, Juho Leinonen, Stephen MacNeil, Arisoa S. Randrianasolo, Brett A. Becker, Bailey Kimmel, Jared Wright, Ben Briggs
- Type
- Peer-reviewed article
- Level
- Higher ed
- Topics
- Equity & Access, Instructional Use, Research
The Widening Gap: The Benefits and Harms of Generative AI for Novice Programmers (arXiv author version)The Widening Gap: The Benefits and Harms of Generative AI for Novice Programmers
An international survey of 3,839 bachelor's, master's, and doctoral students in 16 countries, conducted in July 2024, on how students use AI, how prepared they feel, and what they expect from their universities. Most students reported regular AI use, but many said they lacked AI skills and that their universities' AI adoption and guidelines fell short of their expectations.
Findings
- 86% of students reported regularly using AI in their studies, and 54% used it weekly.
“86% of students globally are regularly using AI in their studies, with 54% of them using AI on a weekly basis”
Source - 58% said they did not have sufficient AI knowledge and skills, and 48% did not feel prepared for an AI-enabled workplace.
“58% reported that they do not feel that they had sufficient AI knowledge and skills, and 48% do not feel adequately prepared for an AI-enabled workplace.”
Source - Only 5% of students said they were fully aware of their institution's AI guidelines and considered them comprehensive.
“only 5% of students indicated that they were fully aware of AI guidelines and feel that they are fully comprehensive”
Source
Method
Online survey conducted July 2024; 3,839 responses from bachelor's, master's, and doctoral students across 16 countries and multiple fields of study.
- Published
- 2 Aug 2024
- Publisher
- Digital Education Council
- Authors
- Digital Education Council
- Type
- Survey or poll
- Level
- Higher ed
- Topics
- AI Literacy, Instructional Use
Digital Education Council Global AI Student Survey 2024Archived copyWhat Students Want: Key Results from DEC Global AI Student Survey 2024
Non-regulatory federal guide for companies and organizations building AI products for education, building on the Department's 2023 AI report and developer listening sessions. It frames trust as a responsibility shared among government, developers, and educators and organizes its advice around five recommendations covering design for teaching and learning, evidence, equity and civil rights, safety and security, and transparency.
Findings
- Developers in the Department's listening sessions singled out the 2023 report's recommendation to strengthen trust, and the guide's central message is that trust is a shared responsibility.
“Through listening sessions, the Department learned that one key recommendation in the 2023 AI Report, “Prioritize Strengthening Trust,” resonated with developers as a call to action.”
Key Message: Shared Responsibility for Building Trust (PDF p. 8) - The guide is organized around five areas of shared responsibility, starting with designing for education values such as keeping humans in the loop and incorporating educator and student feedback throughout product development.
“We categorized these concerns and organized this guide around five key overarching areas of shared responsibility”
Introduction (PDF p. 14) - The guidance is explicitly non-regulatory and does not claim to cover federal guidelines exhaustively.
“This guide provides non-regulatory, education-specific guidance that is aligned with federal guidelines and guardrails.”
Introduction (PDF p. 5)
Method
Federal guidance document informed by Department listening sessions with developers and educators and by the 2023 report Artificial Intelligence and the Future of Teaching and Learning.
- Published
- Jul 2024
- Publisher
- U.S. Department of Education
- Authors
- U.S. Department of Education, Office of Educational Technology
- Type
- Research report
- Level
- K-12, Higher ed
- Topics
- Equity & Access, Instructional Use, Procurement & Vendors, Student Privacy
Designing for Education with Artificial Intelligence: An Essential Guide for DevelopersArchived copy
Ithaka S+R analyzed a generative AI module from its 2024 national survey of instructors at US four-year institutions, answered by 2,654 faculty. Most instructors had experimented with generative AI for teaching but few felt confident using it, opinions on its value were divided, and 42% fully prohibited student use.
Findings
- 72% of instructors had used generative AI for at least one instructional activity, but no single use was common; designing course materials was the most frequent, at 22%.
“A large majority (72 percent) of instructors have experimented with using generative AI as an instructional tool. Yet while instructors are using generative AI in many different ways, no individual use case has become particularly well established.”
Finding 2 - Only 14% agreed they felt confident using generative AI in their instruction, and only 19% agreed it would benefit teaching in their field.
“only 14 percent agreed or strongly agreed they feel confident in their ability to use generative AI in their instruction.”
Finding 1 - 42% of instructors completely prohibited students from using generative AI, rising to 53% in the humanities.
“Overall, 42 percent of instructors completely prohibit their students from using generative AI”
Student use of generative AI
Method
Generative AI module of the Ithaka S+R 2024 US Instructor Survey, fielded February 7 to March 10, 2024, to 135,284 faculty at US four-year institutions; 2,654 respondents randomly assigned to the module (3.9% overall response rate).
- Published
- 20 Jun 2024
- Publisher
- Ithaka S+R
- Authors
- Dylan Ruediger, Melissa Blankstein, Sage Jasper Love
- Type
- Survey or poll
- Level
- Higher ed
- Topics
- Academic Integrity, Educator Training, Instructional Use
Generative AI and Postsecondary Instructional Practices: Findings from a National Survey of InstructorsArchived copy
Researchers at the University of California, Berkeley randomly assigned 274 online adult learners to receive no hints, human tutor-written hints, or ChatGPT-generated hints while practicing algebra or statistics problems. Only the ChatGPT condition produced statistically significant learning gains over the no-help control, and gains did not differ significantly between ChatGPT and human tutor hints, though about a third of ChatGPT hints initially failed quality checks.
Findings
- Only the ChatGPT-hint condition produced statistically significant learning gains compared with a no-help control, with no significant difference in gains or time on task between ChatGPT and human tutor hints.
“We find that only the ChatGPT condition produces statistically significant learning gains compared to a no-help control, with no statistically significant differences in gains or time-on-task observed between learners receiving ChatGPT vs human tutor help.”
Abstract - ChatGPT-generated help failed quality checks on 32% of problems; a self-consistency technique reduced errors to near zero for algebra and 13% for statistics.
“Notably, ChatGPT-generated help failed quality checks on 32% of problems. This was, however, reducible to nearly 0% for algebra problems and 13% for statistics problems after applying self-consistency”
Abstract
Method
Randomized 3 x 4 online experiment (hint condition by subject area); 274 adult Amazon Mechanical Turk participants with at least a high school degree; problems from OpenStax Elementary, Intermediate and College Algebra and Statistics; pre-test, five-problem acquisition phase, and post-test.
- Published
- 24 May 2024
- Publisher
- PLOS ONE
- Authors
- Zachary A. Pardos, Shreya Bhandari
- Type
- Peer-reviewed article
- Level
- K-12, Higher ed
- Topics
- Instructional Use, Research
ChatGPT-generated help produces learning gains equivalent to human tutor-authored help on mathematics skills (PubMed Central full text)Archived copy
Pew Research Center surveyed U.S. public K-12 teachers in fall 2023 about whether AI tools do more good or harm in K-12 education, alongside a separate teen survey on ChatGPT use. A quarter of teachers said AI tools do more harm than good, few said they do more good, and many were unsure, with high school teachers the most negative.
Findings
- Twenty-five percent of public K-12 teachers said AI tools do more harm than good in K-12 education, 32 percent said there is an equal mix, 6 percent said more good than harm, and 35 percent were unsure.
“A quarter of public K-12 teachers say using AI tools in K-12 education does more harm than good. About a third (32%) say there is about an equal mix of benefit and harm, while only 6% say it does more good than harm. Another 35% say they aren’t sure.”
Source - High school teachers (35 percent) were more likely than middle (24 percent) and elementary (19 percent) teachers to say AI tools do more harm than good.
“About a third of high school teachers (35%) say these tools do more harm than good. Roughly a quarter of middle school teachers (24%) and 19% of elementary school teachers say the same.”
How teachers’ views differ by school level - Among teens who had heard of ChatGPT, 19 percent said they had used it for schoolwork in fall 2023.
“Among teens who have heard of ChatGPT, 19% say they have used it to help them with schoolwork.”
Source
Method
Survey of 2,531 U.S. public K-12 teachers (RAND American Teacher Panel), Oct. 17 to Nov. 14, 2023, weighted to state and national teacher characteristics; plus a separate survey of 1,453 U.S. teens ages 13-17, Sept. 26 to Oct. 23, 2023.
- Published
- 15 May 2024
- Publisher
- Pew Research Center
- Authors
- Luona Lin
- Type
- Survey or poll
- Level
- K-12
- Topics
- Instructional Use
A quarter of U.S. teachers say AI tools do more harm than good in K-12 educationArchived copy
RAND surveyed nationally representative samples of K-12 teachers and school districts in fall 2023 and interviewed district leaders to describe how teachers were using AI tools and whether districts were providing AI policies, guidance, and training. About one in five teachers reported using AI for teaching, and most districts planned to train teachers on AI by the end of the 2023-24 school year.
Findings
- As of fall 2023, 18 percent of K-12 teachers reported using AI for teaching and another 15 percent had tried it at least once.
“As of fall 2023, 18 percent of K–12 teachers reported using AI for teaching and another 15 percent have tried AI at least once.”
Key Takeaways - Teachers who used AI most often used it to adapt instructional content to students' levels and to generate materials.
“The most common ways that teachers used AI tools were to adapt instructional content to fit the level of their students and to generate materials.”
Key Takeaways - Sixty percent of districts planned to have trained teachers on AI by the end of 2023-24; urban districts were the least likely to deliver such training.
“By the end of the 2023–2024 school year, 60 percent of districts plan to have trained teachers about AI use. Urban districts were the least likely to deliver such training.”
Key Takeaways - District leaders said they were focusing more on increasing teachers' AI use than on writing student-use policies.
“In interviews, leaders described focusing more on increasing teachers' AI use and less on crafting student use policy, primarily because they saw the potential for AI to make teachers' jobs easier.”
Key Takeaways
Method
Nationally representative surveys of 1,020 teachers (RAND American Teacher Panel, fall 2023) and 231 districts (American School District Panel, fall 2023), plus interviews with 11 district leaders in December 2023 and January 2024.
- Published
- 17 Apr 2024
- Publisher
- RAND Corporation
- Authors
- Melissa Kay Diliberti, Heather L. Schwartz, Sy Doan, Anna Shapiro, Lydia R. Rainey, Robin J. Lake
- Type
- Survey or poll
- Level
- K-12
- Topics
- Educator Training, Equity & Access, Instructional Use
Using Artificial Intelligence Tools in K–12 ClassroomsArchived copy
The Center for Democracy & Technology surveyed a nationally representative sample of middle and high school teachers in late 2023 about generative AI use, school policies, training, AI detection tools, and discipline. Teachers reported more school policies and training than the prior year, alongside heavy reliance on AI content detectors and more students disciplined over AI use.
Findings
- Eighty-five percent of teachers said their school had a policy that permits or bans generative AI for schoolwork, and 71 percent of them said it was their school's first such policy.
“Eighty five percent of teachers say that their school has a policy that either generally permits (subject to some conditions or limits) or bans the use of ChatGPT, or other generative AI tools, for schoolwork.”
p. 10 - Sixty-eight percent of teachers reported regularly using an AI content detection tool, a 30-point increase from the prior school year.
“Sixty eight percent of teachers report using an AI content detection tool regularly, a 30 percentage point increase since last school year.”
p. 13 - The share of teachers reporting that students at their school got in trouble for using or being accused of using generative AI rose from 48 percent in 2022-23 to 64 percent in 2023-24.
“Since last school year, student discipline as a result of generative AI use has increased 16 percentage points”
Figure 3, p. 16
Method
Online survey of a nationally representative sample of 460 6th-12th grade public school teachers fielded November-December 2023 by Edge Research, with quotas and weighting to national demographics; compared with CDT's 2022-23 surveys.
- Published
- 27 Mar 2024
- Publisher
- Center for Democracy & Technology
- Authors
- Maddy Dwyer, Elizabeth Laird
- Type
- Survey or poll
- Level
- K-12
- Topics
- Academic Integrity, Assessment, Educator Training, Equity & Access
Report – Up in the Air: Educators Juggling the Potential of Generative AI with Detection, Discipline, and DistrustArchived copyUp in the Air (full report PDF)Archived copy
A preliminary evaluation randomly assigned 11 Rising Academies schools in Ghana to give students in grades 3-9 access to Rori, a WhatsApp-based math tutor that uses language models, during two 30-minute study hall sessions per week, or to continue regular instruction. Over eight months, math growth scores were substantially higher in the treatment group, at a marginal cost of about $5 per student.
Findings
- Math growth scores were substantially higher for the treatment group, with an effect size of 0.36 that the authors describe as roughly equivalent to an extra year of learning.
“We find that the math growth scores were substantially higher for the treatment group and statistically significant (p < 0.001). The effect size of 0.36 is considered large in the context of educational interventions”
Abstract - Rori runs on basic mobile phones over low-bandwidth networks, with a marginal cost of about $5 per student.
“the marginal cost of providing Rori is approximately $5 per student”
Abstract
Method
School-level randomized trial; about 500 students in grades 3-9 at 11 Rising Academies schools in Ghana (five treatment, six control); two 30-minute weekly sessions; math assessments at two points over eight months; difference-in-differences analysis.
- Published
- 15 Feb 2024
- Publisher
- arXiv
- Authors
- Owen Henkel, Hannah Horne-Robinson, Nessie Kozhakhmetova, Amanda Lee
- Type
- Research report
- Level
- K-12
- Topics
- Equity & Access, Instructional Use, Research
Effective and Scalable Math Support: Experimental Evidence on the Impact of an AI-Math Tutor in Ghana (arXiv v2 PDF)Effective and Scalable Math Support: Evidence on the Impact of an AI- Tutor on Math Achievement in GhanaArchived copy
The first EDUCAUSE AI Landscape Study surveyed higher education staff, faculty, and leaders in late 2023 on AI strategy, policy, workforce, and expectations. Most institutions were working on AI strategy, driven largely by student use of AI, and most respondents said AI had already affected academic integrity, while few described their AI policies as restrictive.
Findings
- Only 11% of respondents said nobody at their institution was working on AI-related strategy.
“Only 11% of respondents said that nobody at their institution is working on AI-related strategy.”
Key Findings - 78% said AI had affected academic integrity.
“A majority of respondents (78%) indicated that AI has impacted academic integrity.”
Key Findings - Only 18% described their AI-related policies as somewhat or extremely restrictive, such as bans on student or faculty use.
“Only 18% of respondents said their AI-related policies are somewhat or extremely restrictive—for example, banning student or faculty use.”
Key Findings
Method
Online survey of the EDUCAUSE higher education community distributed November 27 to December 8, 2023.
- Published
- 12 Feb 2024
- Publisher
- EDUCAUSE
- Authors
- Jenay Robert
- Type
- Survey or poll
- Level
- Higher ed
- Topics
- Academic Integrity, Data Governance, Educator Training, Institutional Operations
2024 EDUCAUSE AI Landscape StudyArchived copy2024 EDUCAUSE AI Landscape Study, Introduction and Key FindingsArchived copy
2023
Researchers tested 14 AI-text detection tools, including Turnitin and PlagiarismCheck, on 54 documents of known origin, including human-written, machine-translated, ChatGPT-generated, manually edited, and machine-paraphrased texts. They concluded the tools are neither accurate nor reliable, tend to classify text as human-written, and perform worse when AI text is obfuscated.
Findings
- All 14 tools scored below 80% accuracy and only five scored above 70%.
“Detection tools for AI-generated text do fail, they are neither accurate nor reliable (all scored below 80% of accuracy and only 5 over 70%).”
Discussion - Human-written texts machine-translated from other languages were misclassified more often, and six of the 14 tools produced false positives that would amount to false accusations.
“Six of the fourteen tools tested generated false positives, with the risk increasing dramatically for machine-translated texts.”
Results, Error analysis - Manual editing and machine paraphrasing raised the share of undetected AI-generated texts to about half or more, and no tool correctly classified all obfuscated documents.
“for manually edited texts (case 05-ManEd) the ratio of undetected texts increases to approx. 50% and in the case of machine-paraphrased texts (case 06-Para) rises even higher.”
Results, Error analysis
Method
Controlled test of 12 free and 2 commercial detectors in March-May 2023; 54 English-language test documents prepared by nine researchers in six categories (human-written, machine-translated, ChatGPT-generated, manually edited, paraphrased with Quillbot), for 756 tests in total.
- Published
- 24 Dec 2023
- Publisher
- International Journal for Educational Integrity
- Authors
- Debora Weber-Wulff, Alla Anohina-Naumeca, Sonja Bjelobaba, Tomáš Foltýnek, Jean Gabriel Guerrero-Dib, Olumide Popoola, Petr Šigut, Lorna Waddington
- Type
- Peer-reviewed article
- Level
- Higher ed
- Topics
- Academic Integrity, Assessment
Testing of detection tools for AI-generated textArchived copy
Stanford researchers ran seven widely used GPT detectors on 91 TOEFL essays by non-native English speakers and 88 US eighth-grade essays. The detectors classified the US student essays accurately but labeled more than half of the TOEFL essays as AI-generated, and simple prompting let AI-written text evade detection.
Findings
- Across seven detectors, the average false-positive rate on human-written TOEFL essays was 61.3%, while US eighth-grade essays were classified accurately.
“While the detectors accurately classified the US student essays, they incorrectly labeled more than half of the TOEFL essays as "AI-generated" (average false-positive rate: 61.3%).”
Main text, GPT detectors exhibit bias against non-native English authors - Using ChatGPT to enrich the vocabulary of the TOEFL essays cut the average false-positive rate from 61.3% to 11.6%, pointing to the detectors' reliance on text perplexity.
“This intervention significantly reduced misclassification, with the average false-positive rate dropping by 49.7% (from 61.3% to 11.6%).”
Main text - Asking ChatGPT to rewrite its own college admission essays with more literary language reduced detection rates to near zero.
“upon prompting ChatGPT to self-edit its text with more literary language (prompt: "Elevate the provided text by employing literary language"), detection rates plummeted to near zero”
Bypassing GPT detectors through linguistic diversity enhancement in prompts
Method
Evaluation of seven publicly available GPT detectors on 91 TOEFL essays from a Chinese forum and 88 US eighth-grade essays from the Hewlett Foundation ASAP dataset, plus ChatGPT-generated Common App essays and scientific abstracts before and after prompt-based rewriting.
- Published
- Jul 2023
- Publisher
- Patterns
- Authors
- Weixin Liang, Mert Yüksekgönül, Yining Mao, Eric Wu, James Zou
- Type
- Peer-reviewed article
- Level
- K-12, Higher ed
- Topics
- Academic Integrity, Equity & Access
GPT detectors are biased against non-native English writers (PubMed Central full text)Archived copy
Federal report from the U.S. Department of Education's Office of Educational Technology, informed by four public listening sessions in 2022 with more than 700 attendees, on the opportunities and risks of AI in teaching, learning, assessment, and research. It closes with seven recommendations for education leaders, led by keeping humans in the loop and developing education-specific guidelines and guardrails.
Findings
- The report draws on four listening sessions held in June and August 2022 with more than 700 attendees, who said action was needed ahead of an expected increase of AI in education technology.
“Through a series of four listening sessions conducted in June and August 2022 and attended by more than 700 attendees, it became clear that constituents believe that action is required now in order to get ahead of the expected increase of AI in education technology”
Introduction, p. 2 - Its central recommendation is that "humans in the loop" be a key criterion for educational use of AI.
“Building on this consensus, we call upon all constituents to adopt “humans in the loop” as a key criterion for educational use of AI.”
Recommendation #1: Emphasize Humans in the Loop, p. 53 - The final recommendation asks policymakers to develop education-specific guidelines and guardrails, noting that local leaders find it increasingly hard to make informed decisions about deploying AI.
“With technology growing in complexity at such a rapid pace, it is becoming difficult for local leaders to make informed decisions about the deployment of artificial intelligence.”
Recommendation #7: Develop Education-Specific Guidelines and Guardrails, p. 60
Method
Policy report based on four public listening sessions (June and August 2022, more than 700 attendees) and consultation with internal and external AI policy experts.
- Published
- May 2023
- Publisher
- U.S. Department of Education
- Authors
- U.S. Department of Education, Office of Educational Technology
- Type
- Research report
- Level
- K-12, Higher ed
- Topics
- Educator Training, Equity & Access, Instructional Use, Research
Artificial Intelligence and the Future of Teaching and Learning: Insights and RecommendationsArchived copy
No research matches these filters.
“Among sections of the same course, tutor access reduced final grades by 0.37 standard deviations and learning management system participation by 0.90 standard deviations.”
Abstract
“We found that access to the AI tutor negatively affected student final course grades by 0.27–0.37 standard deviations (SDs), with the clearest evidence in the exact-match sample.”
Introduction, p. 3
“the estimated final-grade losses among first-generation students were more than twice those among continuing-generation students in both samples”
Introduction, p. 4
“AI students progressed more slowly and attempted fewer questions, but answered more accurately conditional on reaching an attempt. The clearest mechanism appeared after mistakes: AI improved next-attempt correctness”
Abstract
“Mastery increased three-correct-in-a-row attainment but did not by itself improve delayed learning.”
Abstract
“The most encouraging delayed-test evidence appears when AI is embedded in the mastery workflow, with marginally significant gains concentrated on practiced Exercise 1 material.”
Abstract
“Assignment raises math achievement by 1.3 national percentile ranks per term, or about 0.06 to 0.08 standard deviations over a school year; the implied effect of a full year of active participation reaches 0.14 standard deviations.”
Abstract
“These gains resemble those from Khan Academy practice without AI assistance.”
Abstract
“96 percent of students tried Khanmigo at least once, but the median student messaged it on only a third of the days they practiced, and in only 17 percent of the exercise sessions in which they made a mistake.”
Abstract
“Students remain receptive over time, and impacts are concentrated in improved completion of time-sensitive administrative tasks, with no detectable effects on academic performance or persistence.”
Abstract
“We find no statistically significant difference in enrollment and persistence between students in the treatment and control groups”
Results, enrollment and persistence (Table 4)
“We find that centralized ownership and flexible communication are critical for sustainability.”
Abstract
“For the first time, more than half of administrators, faculty, and students use AI at least weekly, with daily GenAI use at its highest point since Spring 2023 (43% of administrators, 32% of students, and 25% of faculty).”
Key insights
“Only 13% of faculty at institutions that ban AI entirely say their policy is effective, versus 30% where AI is integrated into assignments.”
Key insights
“Just 32% of administrators report an institution-wide AI policy, and where one exists, only 22% of faculty believe it's effective compared to 44% of administrators.”
Key insights
“The share of students in middle school grades and up who reported using AI for help with their homework increased from 48 percent in May 2025 to 62 percent in December 2025.”
Key Takeaways
“As of December 2025, 67 percent of students endorsed the statement "The more students use AI for their schoolwork, the more it will harm their critical thinking skills"”
Key Takeaways
“With the exception of using AI to get answers for homework, most students did not feel that their use of AI for other school-related purposes constituted cheating.”
Key Takeaways
“Of the 20 states we examined, they commonly provide professional learning opportunities (18 of 20), leverage partnerships to support AI goals (17 of 20), issue guidance (16 of 20), and pilot AI strategies or tools (11 of 20).”
p. 2
“Sixteen of 20 have issued AI guidance, all of which is non-binding”
Flexibility over Directives, p. 2
“In 19 of the 20 states (95%) in the database, legislatures introduced at least one AI-related bill in 2024 or 2025 that involved K-12 education and/or impacted the SEA.”
p. 3
“we find that at this point in its trajectory, the risks of utilizing generative AI in children’s education overshadow its benefits. This is largely because the risks of AI differ in nature from its benefits—that is, these risks undermine children’s foundational development”
Source
“Overreliance on AI tools and platforms can put children and youth’s fundamental learning capacity at risk.”
Source
“Procure technology that protects students’ privacy, safety, and security.”
Protect pillar
“Only 15% of students say AI is integrated into many of their courses, while 43% say it appears in a few. Another 43% say they have not experienced any AI integration in their courses.”
Key insight 02
“Faculty intent to use AI has declined by 9 percentage points, from 76% in 2025 to 67% in 2026.”
Key insight 05
“60% of students globally worry that their classmates might misuse AI for unfair advantage, rising to 73% in US & Canada.”
Key insight 07
“Only 29% of students believe their instructors are well equipped to guide them on AI use. In the US & Canada, this number is considerably lower at 17%.”
Key insight 03
“The SREB Commission on AI in Education has approved 12 final recommendations to help states, schools and colleges as they adopt artificial intelligence tools.”
Source
“States should regularly review, and revise if necessary, their educational technology safety policies to ensure that minors are protected from harm related to AI use.”
Recommendation 10
“States should integrate AI knowledge and skills into K-12 standards and curricula”
Source
“Our Early Adopters database has nearly doubled in a year, from 40 to 79 districts.”
More Early Adopters Are Piloting Systemwide AI Strategies
“this year, we see more than a third (37%) implementing five or more strategies”
More Early Adopters Are Piloting Systemwide AI Strategies
“Only a quarter (27%) share information about AI courses, and just six districts (8%) communicate any adjustments to learning standards.”
Early Adopters Focus More on Students, Mostly with New Tools
“85 percent of teachers report they have used AI in at least one way in their classroom in the past school year (2024-25), and 86 percent of students say they have used AI for either school or personal use.”
AI Literacy section, p. 34
“Teachers and students who report many school-related uses are also more likely to report exposure to risks like:”
Emerging Risks, p. 11
“As a friend/companion 42%”
Table 3, p. 25
“Almost half of teachers and students report that they have received some training or information about AI from someone at their school, and most found it helpful.”
Figure 30, p. 35
“In 2025, 54 percent of students and 53 percent of English language arts, math, and science teachers indicated that they used AI for school.”
Key Takeaways
“Sixty-one percent of parents, 48 percent of middle schoolers, 55 percent of high schoolers, and only 22 percent of district leaders agreed that greater use of AI will harm students’ critical-thinking skills.”
Key Takeaways
“As of spring 2025, 35 percent of district leaders reported that they provide students with training on AI.”
Key Takeaways
“Forty-five percent of principals reported having school or district policies or guidance on the use of AI in schools, and 34 percent of teachers reported having school or district policies on the use of AI related to academic integrity.”
Key Takeaways
“Seventy-two percent of teens have used AI companions at least once, and over half (52%) qualify as regular users who interact with these platforms at least a few times a month.”
Key Findings, finding 1
“Among AI companion users, 33% have chosen to discuss important or serious matters with AI companions instead of real people”
Key Findings, finding 9
“One-third of users experience discomfort with AI companions.”
Key Findings, finding 8
“About 80% of the districts we studied fell into the first two categories of Dabblers and Emerging Users”
p. 4
“Two challenges stood out most prominently: data privacy and sustainable funding, each cited by more than a third (35%) of respondents as posing “considerable” or “extreme” difficulty.”
p. 12
“80% of surveyed districts that operate in states without guidance (at the time of this survey) reported the lack of state guidance as an “extremely difficult” barrier”
p. 13
“having GPT-4 access while solving problems significantly improves performance (48% improvement in grades for GPT Base and 127% for GPT Tutor)”
Abstract
“on the exam, students in the GPT Base arm perform statistically significantly worse than students in the control arm by 17%; this negative effect is essentially eradicated in the GPT Tutor arm, though we still do not observe a positive effect.”
Main results
“we find evidence that students do not perceive any reduction in their learning or subsequent performance as a consequence of copying solutions”
Introduction
“At least five states have implemented or are in the process of developing pilot programs to foster integration and responsible use in the classroom.”
Source
“Of the teachers who participated, 53% indicated that their experience was either positive or very positive.”
Source
“The Iowa Department of Education announced a $3 million investment to provide all Iowa elementary schools (public and nonpublic) with an AI reading tutor tool”
Source
“In the 2024-25 school year, six in 10 teachers reported using an AI tool for their work.”
Source
“Teachers who use AI tools at least weekly estimate they save 5.9 hours per week, on average. Over the course of a 37.4-week school year, these time savings add up to the equivalent of six weeks per school year .”
Regular Users Earn an Estimated “AI Dividend” of 5.9 Hours per Week
“schools with an AI policy are earning an AI dividend that is 26% greater than schools without an AI policy (2.3 vs. 1.7 hours saved per week per teacher).”
p. 17
“most teachers (68%) have not engaged in any training provided by their school or district on how to use AI or AI tools.”
p. 18
“Fast forward to April 2025, and at least 28 states have published guidance on AI in K-12 settings.”
Source
“As of April 1, 2025, only one AI-related education bill has been enacted this year: Mississippi ’s S.B. 2426 , which creates a state AI task force.”
Source
“Task force reports from Arkansas ( 2025 ), Georgia ( 2024 ) and Illinois ( 2024 ) highlight similar priorities like creating curricular frameworks for AI literacy, investing in educator professional development”
Source
“daily or weekly use up to 40% among administrators (from 2% in spring 2023), 30% among instructors (from 4% in spring 2023), and 42% among students in spring 2025 (from 14% in spring 2023)”
p. 6, Key findings
“Preventing student cheating and increasing student attendance are the top instructional challenges among instructors teaching introductory or developmental courses (45% and 44%, respectively)”
p. 6, Key findings
“Instructor preference for in-person teaching jumped to 64%, up from 55% in 2023.”
Source
“EEG revealed significant differences in brain connectivity: Brain-only participants exhibited the strongest, most distributed networks; Search Engine users showed moderate engagement; and LLM users displayed the weakest connectivity.”
Abstract
“In the LLM ‑ assisted group, 83.3 % of participants (15/18) failed to provide a correct quotation, whereas only 11.1 % (2/18) in both the Search ‑ Engine and Brain ‑ Only groups encountered the same difficulty.”
Results, Session 1 interview, Question 3 (Ability to Quote)
“Self-reported ownership of essays was the lowest in the LLM group and the highest in the Brain-only group.”
Abstract
“The median learning gains for students, relative to the pre-test baseline (M = 2.75, N = 316), in the AI-tutored group were over double”
Results, Learning gains
“While the linear regression suggests an effect size of 0.63, this is an underestimation due to ceiling effect”
Results, regression analysis
“70% of students in the AI group spent less than 60 minutes on task, while 30% spent more than 60 minutes on task. The median time on task for students in the AI group was 49 minutes.”
Results, Time on task
Results, students' perceptions of learning
“The intervention demonstrated a significant improvement of 0.31 standard deviation on an assessment that included English topics aligned with the Nigerian curriculum, knowledge of artificial intelligence and digital skills.”
Abstract
“We calculate an ITT effect of being selected for the program on the performance in the third-term exam of 0.21 standard deviations.”
Introduction, p. 2
“with an effect size of approximately 0.031 standard deviation per additional day of attendance”
Introduction, p. 3
“In fall 2024, 48 percent of districts reported that they had trained teachers on the use of AI, which is a substantial, 25-percentage-point increase from the percentage of districts that reported implementing training on AI in fall 2023.”
Survey Results
“By fall 2024, 67 percent of low-poverty districts reported having provided training for teachers on AI use, compared with 42 percent of middle-poverty districts and 39 percent of high-poverty districts.”
Survey Results
“In interviews, district leaders said that they focused initial teacher trainings on addressing teachers' fear and discomfort with AI rather than jumping right into instructional tools.”
Key Findings
“Most Gen Zers (79%) say they use generative artificial intelligence products, including 47% who do so on a weekly basis”
Key findings, 01
“Gen Zers are also more than twice as likely to say AI will harm, rather than help, their critical thinking skills (49% to 22%).”
Key findings, 02
“just 28% of students say their school explicitly permits the use of artificial intelligence, while nearly half (49%) say their school does not have clear policies regarding AI.”
Key findings, 03
“Despite 81% of CS teachers agreeing that using AI and learning about AI should be included in a foundational CS learning experience, less than half of high school CS teachers felt equipped to teach AI.”
Chapter 7 Highlights, p. 368
“Student participation varies by state, race/ethnicity, school size, geography, income, gender, and disability.”
Chapter 7 Highlights, p. 368
“Graduates who earned their master’s degree in AI in the U.S. nearly doubled between 2022 and 2023.”
Chapter 7 Highlights, p. 368
“The proportion of respondents reporting that their institution has AI-related AUPs increased from 23% last year to 39% this year”
Key Findings, Policies and Guidelines
“Only 9% of respondents reported that their institution's cybersecurity and privacy policies are adequate for addressing AI-related risks to the institution.”
Key Findings, Policies and Guidelines
“Teaching and learning is the functional area at the institution most focused on using AI, with particular focus on the areas of academic integrity (74%), coursework (65%), assessment practices (54%), and curriculum design (54%).”
Key Findings, Use Cases
“one-quarter of ELA, math, and science teachers used AI tools for instructional planning or teaching in the 2023–2024 school year. Nearly 60 percent of surveyed American School Leader Panel principals also reported using AI tools for their work in 2023–2024.”
Key Takeaways
“almost 40 percent of ELA or science teachers reported using AI compared with 20 percent of general elementary education or math teachers.”
Key Takeaways
“Eighteen percent of principals reported that their schools or districts provided guidance on the use of AI by staff, teachers, or students.”
Key Takeaways
“The survey revealed widespread concerns about students' ability to critically evaluate AI-generated outputs, with 83% of faculty members expressing concern.”
Faculty Concerns
“Over half of faculty (54%) believe current student evaluation methods are no longer adequate in the age of AI, with 13% calling for an urgent, complete revamp.”
Source
“only 4% of faculty are fully aware of their institutional AI guidelines and feel they are fully comprehensive. Just 6% fully agree that their institutions have provided sufficient resources to train faculty's AI literacy.”
Source
“The share of teens who say they use ChatGPT for their schoolwork has risen to 26%, according to a Pew Research Center survey of U.S. teens ages 13 to 17. That’s up from 13% in 2023.”
Source
“Black and Hispanic teens (31% each) are more likely than White teens (22%) to say they have used ChatGPT for their schoolwork.”
By race and ethnicity
“18% say it’s acceptable to use ChatGPT to write essays, and 42% say it’s not acceptable.”
Source
“Seventy-one percent of respondents said decision-making and AI initiatives are overwhelmingly led by college or university administrations”
Findings, concern 2
“Respondents were overwhelmingly concerned with student plagiarism made possible by generative AI. Ninety-one percent noted that they were at least somewhat concerned about preventing academic dishonesty.”
Findings, concern 3
“AI has generally led to at least somewhat worse outcomes for the teaching environment (according to 62 percent of respondents), pay equity (30 percent), job enthusiasm (76 percent), academic freedom (40 percent), and student success (69 percent).”
Findings
“ChatGPT group outperformed in the essay score improvement but their knowledge gain and transfer were not significantly different.”
Abstract, Results
“learners who received different learning support showed no difference in post-task intrinsic motivation; 2) there were significant differences in the frequency and sequences of the self-regulated learning processes among groups”
Abstract, Results
Practitioner Notes, What this paper adds
“Of the 40 school systems we studied, 26 (65%) have provided publicly available guidance to staff, students, and families on using AI.”
Beyond tools section
“The most common strategies include using AI-enabled tools for key stakeholders (teachers, students, and parents at 70%, 58%, and 38%, respectively) and providing professional development opportunities (63%).”
Beyond tools section
“These 40 school systems collectively endorse or partner with 65 tools”
Beyond tools section
“students whose tutors used Tutor CoPilot were 4 percentage points more likely to master mathematics topics (p<.01)”
Abstract
“lower-rated tutors experienced a 9 p.p. increase in students passing their exit ticket (56%→ 65%student passing rate from control to treatment). In Figure 2b, less experienced tutors showed a 7 p.p. increase (61%→ 68%).”
Results, RQ2
“Analysis of more than 350,000 messages indicates that Tutor CoPilot promotes effective pedagogy, increasing probing questions and reducing generic praise.”
Abstract
“Only about 25% of programs surveyed provide training on ways that AI can support new ways of teaching”
p. 3
“only 10% of the education school leaders surveyed reported that their faculty members are confident in using AI, and most do not use it in their instruction.”
p. 4
“More than two-thirds of survey respondents reported that their institutions have no policies around AI tool use”
p. 3
“while 70% of teens say they have used at least one type of gen AI tool”
Key Findings, p. 4
“37% of teens are not sure whether their school has gen AI rules.”
Key Findings, finding 6
“Black teens are more than twice as likely as White or Latino teens to say that teachers flagged their schoolwork as being created by generative AI when it was not (20% vs. 7% and 10%, respectively).”
Key Findings, finding 7, p. 10
“The large majority of parents of teens (83%) say that their child's school has not communicated with them about”
Key Findings, finding 6
“39.5% when evaluating unmanipulated AI-generated content. Importantly, regarding the human-written control samples, only 67% of the tests were accurate”
Results, baseline testing
“Results show significant reductions in detector accuracy (17.4%) when faced with simple techniques to manipulate the AI generated content.”
Abstract
“they cannot currently be recommended for determining academic integrity violations due to accuracy limitations and the potential for false accusation which undermines inclusive and fair assessment practices.”
Abstract
“Although 20 of 21 students completed the assigned programming problem, our findings show an unfortunate divide in the use of GenAI tools between students who accelerated and students who struggled.”
Abstract
“struggling students often expressed cognitive dissonance about their problem solving ability, thought they performed better than they did, and finished with an illusion of competence.”
Abstract
“Students with better grades and higher self-efficacy were more likely to use GenAI tools to accelerate towards a solution”
Section 1, Introduction
“86% of students globally are regularly using AI in their studies, with 54% of them using AI on a weekly basis”
Source
“58% reported that they do not feel that they had sufficient AI knowledge and skills, and 48% do not feel adequately prepared for an AI-enabled workplace.”
Source
“only 5% of students indicated that they were fully aware of AI guidelines and feel that they are fully comprehensive”
Source
“Through listening sessions, the Department learned that one key recommendation in the 2023 AI Report, “Prioritize Strengthening Trust,” resonated with developers as a call to action.”
Key Message: Shared Responsibility for Building Trust (PDF p. 8)
“We categorized these concerns and organized this guide around five key overarching areas of shared responsibility”
Introduction (PDF p. 14)
“This guide provides non-regulatory, education-specific guidance that is aligned with federal guidelines and guardrails.”
Introduction (PDF p. 5)
“A large majority (72 percent) of instructors have experimented with using generative AI as an instructional tool. Yet while instructors are using generative AI in many different ways, no individual use case has become particularly well established.”
Finding 2
“only 14 percent agreed or strongly agreed they feel confident in their ability to use generative AI in their instruction.”
Finding 1
“Overall, 42 percent of instructors completely prohibit their students from using generative AI”
Student use of generative AI
“We find that only the ChatGPT condition produces statistically significant learning gains compared to a no-help control, with no statistically significant differences in gains or time-on-task observed between learners receiving ChatGPT vs human tutor help.”
Abstract
“Notably, ChatGPT-generated help failed quality checks on 32% of problems. This was, however, reducible to nearly 0% for algebra problems and 13% for statistics problems after applying self-consistency”
Abstract
“A quarter of public K-12 teachers say using AI tools in K-12 education does more harm than good. About a third (32%) say there is about an equal mix of benefit and harm, while only 6% say it does more good than harm. Another 35% say they aren’t sure.”
Source
“About a third of high school teachers (35%) say these tools do more harm than good. Roughly a quarter of middle school teachers (24%) and 19% of elementary school teachers say the same.”
How teachers’ views differ by school level
“Among teens who have heard of ChatGPT, 19% say they have used it to help them with schoolwork.”
Source
“As of fall 2023, 18 percent of K–12 teachers reported using AI for teaching and another 15 percent have tried AI at least once.”
Key Takeaways
“The most common ways that teachers used AI tools were to adapt instructional content to fit the level of their students and to generate materials.”
Key Takeaways
“By the end of the 2023–2024 school year, 60 percent of districts plan to have trained teachers about AI use. Urban districts were the least likely to deliver such training.”
Key Takeaways
“In interviews, leaders described focusing more on increasing teachers' AI use and less on crafting student use policy, primarily because they saw the potential for AI to make teachers' jobs easier.”
Key Takeaways
“Eighty five percent of teachers say that their school has a policy that either generally permits (subject to some conditions or limits) or bans the use of ChatGPT, or other generative AI tools, for schoolwork.”
p. 10
“Sixty eight percent of teachers report using an AI content detection tool regularly, a 30 percentage point increase since last school year.”
p. 13
“Since last school year, student discipline as a result of generative AI use has increased 16 percentage points”
Figure 3, p. 16
“We find that the math growth scores were substantially higher for the treatment group and statistically significant (p < 0.001). The effect size of 0.36 is considered large in the context of educational interventions”
Abstract
“the marginal cost of providing Rori is approximately $5 per student”
Abstract
“Only 11% of respondents said that nobody at their institution is working on AI-related strategy.”
Key Findings
“A majority of respondents (78%) indicated that AI has impacted academic integrity.”
Key Findings
“Only 18% of respondents said their AI-related policies are somewhat or extremely restrictive—for example, banning student or faculty use.”
Key Findings
“Detection tools for AI-generated text do fail, they are neither accurate nor reliable (all scored below 80% of accuracy and only 5 over 70%).”
Discussion
“Six of the fourteen tools tested generated false positives, with the risk increasing dramatically for machine-translated texts.”
Results, Error analysis
“for manually edited texts (case 05-ManEd) the ratio of undetected texts increases to approx. 50% and in the case of machine-paraphrased texts (case 06-Para) rises even higher.”
Results, Error analysis
“While the detectors accurately classified the US student essays, they incorrectly labeled more than half of the TOEFL essays as "AI-generated" (average false-positive rate: 61.3%).”
Main text, GPT detectors exhibit bias against non-native English authors
“This intervention significantly reduced misclassification, with the average false-positive rate dropping by 49.7% (from 61.3% to 11.6%).”
Main text
“upon prompting ChatGPT to self-edit its text with more literary language (prompt: "Elevate the provided text by employing literary language"), detection rates plummeted to near zero”
Bypassing GPT detectors through linguistic diversity enhancement in prompts
“Through a series of four listening sessions conducted in June and August 2022 and attended by more than 700 attendees, it became clear that constituents believe that action is required now in order to get ahead of the expected increase of AI in education technology”
Introduction, p. 2
“Building on this consensus, we call upon all constituents to adopt “humans in the loop” as a key criterion for educational use of AI.”
Recommendation #1: Emphasize Humans in the Loop, p. 53
“With technology growing in complexity at such a rapid pace, it is becoming difficult for local leaders to make informed decisions about the deployment of artificial intelligence.”
Recommendation #7: Develop Education-Specific Guidelines and Guardrails, p. 60
No findings match these filters.