Skip to the main content
PeabodyMake a free story

2026 in review

The 12 most important studies of 2026 on AI and learning

This year brought the first big classroom trials of AI tutors, plus national surveys of how kids really use chatbots. The headline is not "AI works" or "AI fails." It is that design and engagement decide the outcome: tutors that make kids review mistakes help a little, while open access to a chatbot can lower grades. We list each study in plain words with its size, what it found and its limits. Several are working papers, not yet peer reviewed. Read them as strong clues, not final answers, and share them with your child's teacher.

We chose studies with large samples, randomized designs or national survey methods, published or released in 2026 (or late 2025 where they shaped the year). Links were checked in October 2026. Checked 2026-10-24. Nobody paid to be here and nothing is an affiliate link.

  1. Making AI Tutoring Productive (NBER 35621)

    Oreopoulos and colleagues

    A randomized trial with 6,000+ Tennessee middle schoolers. Students with AI help went slower and tried fewer problems, but recovered better after mistakes. The best delayed-test results came when AI was paired with mastery practice.

    Watch for: Working paper; gains were modest.

  2. One Click Away: Khanmigo in a two-year school experiment (NBER 35620)

    NBER

    A two-year trial in the same Tennessee district found math scores rose about 1.3 national percentile points per term with Khanmigo. The authors say student engagement, not the tutor, was the limit.

    Watch for: Small effect; not yet peer reviewed.

  3. University AI tutor trial that lowered grades (IZA DP 18989)

    IZA Institute of Labor Economics

    In a trial with 2,379 undergraduates, access to an AI tutor cut final grades by 0.37 standard deviations and sharply reduced use of other course activities. A warning about tools that replace effort.

    Watch for: College students, not children; results may differ for kids.

  4. Access is Not Enough: human support and AI reading tutors

    Stanford National Student Support Accelerator

    Two randomized trials in elementary reading found kids barely used the AI tutor on their own and reading did not improve. Adding human tutors raised engagement.

    Watch for: Shows the tool alone is not a plan.

  5. OECD Digital Education Outlook 2026

    OECD

    The OECD's big January review found general chatbots often improve the work students hand in but not what they learn, and the edge can vanish or reverse on exams without AI. It warns of "metacognitive laziness."

    Watch for: Linked summary; the OECD site may block some browsers.

  6. A New Direction for Students in an AI World

    Brookings Center for Universal Education

    A "premortem" based on interviews in 50 countries and hundreds of papers. It concludes AI's risks to kids' learning and social growth currently outweigh benefits, but says the harm is fixable.

    Watch for: Interviews and literature review, not a trial.

  7. How Teens Use and View AI

    Pew Research Center · Ages 13-17

    A national survey of 1,458 teens and parents. 54% of teens have used chatbots for schoolwork and 12% for emotional support or advice. About 1 in 10 do most of their assignments with AI.

    Watch for: Fielded fall 2025, so use is likely higher now.

  8. Common Sense Census: AI Use by Tweens and Teens

    Common Sense Media · Ages 9-17

    The first yearly census of 1,204 kids ages 9-17. 86% use AI, 24% daily, and 44% say no parent has ever talked to them about using it safely. One in six chatbot users saw inappropriate material.

    Watch for: Survey shows links with loneliness, not cause.

  9. School phone bans and test scores (two NBER studies)

    Stanford, Penn, Duke, Michigan researchers

    Data from about 4,600 schools found bans had close to zero average effect on scores, attendance or bullying, though in-class phone use dropped sharply. A separate Florida district study found second-year score gains.

    Watch for: Results vary by setting and how long the ban has run.

  10. Public Montessori preschool lottery study

    University of Virginia (PNAS) · Ages 3-6

    A national lottery study of 588 children at 24 public Montessori schools found winners ahead in reading, executive function and social understanding by the end of kindergarten, at lower cost per child.

    Watch for: Published late 2025; not an AI study, but a strong reminder of what works for young kids.

  11. Young children's brains on ChatGPT (Technion preprint)

    Technion researchers via NYT Magazine · Ages 6-7

    Brain scans of 6- and 7-year-olds using ChatGPT showed weaker coordination in attention areas than adults, and creative kids told fewer imaginative stories afterward. The lead author says 6 is probably too early.

    Watch for: Small preprint, not yet peer reviewed.

  12. Gen Z's AI adoption steady, skepticism climbs

    Gallup, Walton Family Foundation, GSV · Ages 14+

    A spring survey of 1,572 people ages 14-29 found weekly AI use flat at about half, anger toward AI up to 31%, and 74% of students saying their school now has AI rules.

    Watch for: Covers older teens and young adults.

More lists

All 112 lists →