Does AI Make Kids Worse at Math? What Research Shows
Quick answer: AI can hurt math learning when students use it to skip problem-solving. In a 2025 experiment at a Turkish high school, students using an unrestricted GPT-4 chatbot scored about 17% lower than students without AI on immediate tests taken without AI, while a tutor with safeguards showed no detectable difference. Structured AI-supported programs have helped some students, and a 2026 two-year trial of Khanmigo found modest gains limited mostly by how rarely students used it. The key question is what your child can solve independently afterward.
Research checked November 22, 2026.
Parents and teachers are worried that AI will do kids’ thinking for them. Others say AI tutors could give every child personalized help. Both can be true. The research so far points to one clear pattern: the effect depends on whether AI replaces the student’s effort or supports it.
The Key Studies at a Glance
| Study | Who | Finding | Limits |
|---|---|---|---|
| Bastani et al., PNAS, 2025 | Nearly 1,000 high school students, one school in Turkey, four sessions | Basic GPT-4 chatbot: assisted practice up 48%, immediate unaided exams about 17% lower (relative). Safeguarded tutor: practice up 127%, no statistically detectable exam difference. | Short-term; one school |
| Khanmigo two-year trial (NBER working paper, 2026) | 18 Tennessee middle schools, remedial math | Khan Academy with Khanmigo set to coach, not give answers: about 0.06 to 0.08 standard deviations per school year, similar to Khan Academy practice without AI | Students rarely used the tutor; working paper |
| Henkel et al., 2024 (Rori) | Students in grades 3–9, 11 schools in Ghana (477 completed both tests in the published version) | WhatsApp AI math tutor, two 30-minute sessions a week for 8 months: effect size about 0.36 on math growth | Year one; school-level assignment; extra practice time not separated from AI |
| Wang et al., 2024–2025 (Tutor CoPilot) | 900 tutors randomized; about 1,000 students in grades 3–6 analyzed | Students whose human tutors had AI guidance were 4 percentage points more likely to master topics; 9 points for lower-rated tutors | AI supported human tutors, not students directly |
| Kestin et al., Scientific Reports, 2025 | 194 Harvard physics students | A purpose-built AI tutor beat an active-learning class on two lessons (median 4.5 vs 3.5) | College physics; short-term |
When AI Hurts Learning
How to read these numbers: the studies use different measures. A 17% relative decline, a 4-percentage-point gain and an effect size of 0.36 cannot be compared directly, and “no detectable difference” does not prove two approaches are equal.
The clearest warning comes from the 2025 PNAS study by Hamsa Bastani and colleagues. Students who practiced with a basic GPT-4 chatbot solved more practice problems, often by getting answers, but scored lower than students with no AI when tested on their own. The researchers’ explanation: students used the chatbot as a crutch and skipped the effort that builds skill. A small 2025 MIT-led preprint on adults writing essays (“Your Brain on ChatGPT,” 54 participants) points the same way, though it did not study children or math.
When AI Helps Learning
The studies with gains share a design: AI guided the student’s own work instead of replacing it.
- Safeguarded tutors. The PNAS tutor version used teacher-written solutions and gave hints, and avoided the test-score drop.
- Extra practice where help is scarce. Rori gave Ghanaian students regular tutoring by text on basic phones, with sizeable gains (the study could not separate the AI from the extra practice time).
- AI helping human tutors. Tutor CoPilot suggested better teaching moves to tutors, at an estimated AI cost of about 20 dollars per tutor per year.
Engagement is the catch. In the two-year Khanmigo trial, 96% of students tried the tutor, but the median student messaged it on only a third of practice days, and mostly with bare answers. Access alone did not create learning.
What Kids Themselves Report
In a Pew Research Center survey of US teens in fall 2025, 54% said they had used AI chatbots for schoolwork help and 43% had used one to help solve a math problem; 59% said students at their school use chatbots to cheat at least somewhat often. See is using ChatGPT for math homework cheating?
What We Don’t Know Yet
- Long-term effects. Most studies last weeks or months, not years.
- Young children. Few studies cover early elementary students.
- Everyday use. Lab and school studies use designed tools; most kids use general chatbots at home.
- Newer models. Many studies used 2023–2024 models.
What Parents Can Do
- Use a simple routine. Your child tries the problem first, asks for one hint if stuck, explains the next step, checks the answer, then closes the chatbot and tries a different problem. Revisit the skill the next day without help.
- Encourage hint mode. Use prompts like “don’t give me the answer, give me one hint” or a chatbot’s study mode (see how to use AI to learn math).
- Protect practice time. Some math practice should happen without devices.
- Follow school rules and talk with teachers about what is allowed.
- Teach checking. Estimating and substituting back build skills and catch AI errors (see how to check ChatGPT math).
The Bottom Line
AI does not automatically make kids worse or better at math. Used as an answer machine, it can weaken skills that only grow through effort. Used as a patient tutor that asks questions and gives hints, it can help, especially where good human help is hard to find.
Common Misconceptions
“Research proves AI ruins learning.” No. Some uses hurt; well-designed tutoring helped in several trials.
“Any AI tutor will help.” No. Design matters: answer-giving chatbots hurt test scores in a controlled study.
“The essay EEG study proves AI damages kids’ brains.” No. It studied 54 adults writing essays over a few sessions; it is a warning about offloading, not proof of harm to children’s math ability.
“Giving students an AI tutor is enough.” No. In the Khanmigo trial, most students rarely used the tutor in depth.
Watch: Does AI Help or Hurt Learning?
If you learn better by watching, these videos are worth your time.
Check Your Understanding
Five quick questions to see how much of this has stuck. Pick an answer and press Submit; you will see a short explanation either way.
Answer each question and press Submit to see your result right away.
Frequently Asked Questions
Does AI make kids worse at math?
It can when used to get answers. A 2025 study found students using a basic GPT-4 chatbot scored about 17% lower on later tests without AI. But well-designed AI tutors that give hints showed no drop or real gains in other trials.
Can AI tutors improve math scores?
Some can, modestly. A WhatsApp AI math tutor in Ghana showed an effect size of about 0.36 over 8 months, AI guidance for human tutors raised topic mastery by 4 percentage points, and a two-year Khanmigo trial found gains of about 0.06 to 0.08 standard deviations a year, limited by low engagement.
What does “Your Brain on ChatGPT” show?
The 2025 MIT-led preprint found adults writing essays with a chatbot showed weaker brain connectivity on EEG than those writing unaided. It studied 54 adults and essays, not children or math.
How should kids use AI for math?
Try problems first, ask for hints instead of answers, explain the steps back, practice without AI regularly and check answers independently.
Should schools ban AI for math?
Research suggests the use matters more than the tool. Many educators favor clear rules: AI allowed for hints and explanations, not for completing graded work.
Keep Exploring
- Is Math Still Worth Learning If AI Can Do It?
- How to Use AI to Learn Math: 8 ChatGPT Study Habits
- Is Using ChatGPT for Math Homework Cheating?
Sources and Further Reading
- Bastani et al., Generative AI without guardrails can harm learning: Evidence from high school mathematics (PNAS, 2025)
- Henkel et al., Effective and Scalable Math Support: Evidence on the Impact of an AI-Tutor on Math Achievement in Ghana (arXiv 2402.09809)
- Wang et al., Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise (arXiv 2410.03017)
- Kestin et al., AI tutoring outperforms in-class active learning (Scientific Reports, 2025)
- Kosmyna et al., Your Brain on ChatGPT (arXiv 2506.08872)
- NBER: One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment (w35620)
- Pew Research Center: How Teens Use and View AI (Feb 24, 2026)
Related to This Article
More math articles
- Invention, Discovery, Diffusion, and Cultural Change
- Pregnancy Test Calculator (When to Test)
- Benchmark Brilliance: How to Estimate Decimal Sums and Differences with Ease
- Illinois IAR Grade 6 Math Free Worksheets: Printable Grade 6 Math Practice, Answers Included
- 8th Grade MCAS Math FREE Sample Practice Questions
- Overview of the FTCE General Knowledge Math Test
- States of Matter and Changes
- Top 10 ALEKS Math Prep Books (Our 2027 Favorite Picks)
- What Is a Strip Diagram? Multi-Step Word Problem Examples
- Nevada SBAC Grade 6 Math Free Worksheets: 72 Free PDF Worksheets with Step-by-Step Answer Keys


What people say about "Does AI Make Kids Worse at Math? What Research Shows - Effortless Math"?
No one replied yet.