AI wrote strong maths practice exercises—but did it improve exam results?
Six reviewers rated most generated exercises positively, but the Belgian crossover study found no statistically significant adjusted exam advantage in either language cohort.
The short version
- What happened
- The study followed 343 English-course and 236 Dutch-course students taught by the same professor, with a crossover between AI-feedback and conventional-feedback sessions.
- Who it affects
- Students may receive rapid, individualized practice, but an unchecked mathematical error can teach the wrong method with persuasive wording.
- What remains unproven
- This is an unreviewed, non-randomized study at one Belgian university; attendance and feedback use were self-selected, and the language cohorts differed in ways that complicate comparison.
Reviewers liked the practice tasks more than every explanation
Belgian researchers tested a system that reads handwritten mathematics and returns a new practice exercise with feedback. Six reviewers assessed 157 messages twice, producing 314 ratings. The exercise itself received the strongest rating in 300 cases, or 95.5%. Mathematical accuracy was rated strongest in 195 cases, while 26 ratings, or 8.3%, identified a deficiency. The gap matters: a useful-looking task can still contain reasoning that needs checking.[1]
Who it affects
Students may receive rapid, individualized practice, but an unchecked mathematical error can teach the wrong method with persuasive wording.[1]
Keep in mind
This is an unreviewed, non-randomized study at one Belgian university; attendance and feedback use were self-selected, and the language cohorts differed in ways that complicate comparison.
Source analysis · 1 linked source. Original source date: 20 October 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 3 min The evidence, methods, implications and limitations
Reviewers liked the practice tasks more than every explanation
Belgian researchers tested a system that reads handwritten mathematics and returns a new practice exercise with feedback. Six reviewers assessed 157 messages twice, producing 314 ratings. The exercise itself received the strongest rating in 300 cases, or 95.5%. Mathematical accuracy was rated strongest in 195 cases, while 26 ratings, or 8.3%, identified a deficiency. The gap matters: a useful-looking task can still contain reasoning that needs checking.
The classroom study involved 343 students in an English-language course and 236 in a Dutch-language course taught by the same professor. For the first feedback session, 331 English-course and all 236 Dutch-course students were observed. In the crossover analysis, 199 English-course and 147 Dutch-course students took part in both conditions, receiving AI feedback in one session and conventional feedback in another.[1]
The exam analysis did not establish a learning gain
The researchers compared later exam scores for four combinations of course language and feedback sequence, adjusting for earlier marks and several student characteristics. None of the four estimated differences was statistically significant. Attendance and engagement were associated with stronger results in some analyses, but students were not randomly assigned to attend or use feedback, so those patterns do not establish causation.
For teachers, the sensible use is formative support with review: check the mathematics, let students question the explanation and keep consequential grading separate. The cohorts differed in background and teaching-assistant arrangements, the study came from one institution and the preprint is unreviewed. A larger randomized trial with independently marked outcomes, error tracking and comparable implementation across languages would change the assessment.[1]
The distinction that matters
What the study measured—and what it did not
Quality review and later exam performance answer different questions.
Message review
Experts rate exercises and explanations
Identifies content strengths and errors
Crossover
Students receive two feedback conditions
Allows within-course comparison
Exam model
Later scores are adjusted statistically
No significant advantage was found
Watch the background, then check the evidence
Illustrative photo · not a video frame
UNESCO · 8 min 30 sec · Background explainer
Teacher to Teacher: Is AI reshaping education for the better?
YouTube loads only when you choose; playback starts when you press play.
A conversation about how teachers can guide AI use while preserving pupils’ judgement. Watch for the difference between helping a learner think and supplying an answer.
Evidence check: Teachers’ perspectives provide context. They do not establish that a particular AI tool improves learning outcomes.
Published 2025-01-23. Source checked 14 October 2026. Check captions on the publisher’s video page.
Read the education evidencePhoto: Vitaly Gariev / Unsplash · Licence
Evidence check
This is an unreviewed, non-randomized study at one Belgian university; attendance and feedback use were self-selected, and the language cohorts differed in ways that complicate comparison.
Reporting basis: Source analysis. Original source date: 20 October 2026.
Continue the story
Read next
Continue exploring generative ai: compare the finding and its evidence limits.
Does AI help trainee teachers reflect better after metacognitive training?
In a randomized 60-person experiment, a configured Socratic AI improved written reflection only when paired with a 20-minute metacognitive lesson. The immediate, single-university result does not show that generic chatbot access improves teaching practice.
Read the next reportExplore the full source and evidence record
What this means for people
- Students may receive rapid, individualized practice, but an unchecked mathematical error can teach the wrong method with persuasive wording.
Global context
The bilingual design is useful, but it does not establish equivalent effects across other languages, curricula, age groups or handwriting conventions.
What the evidence does not yet show
- This is an unreviewed, non-randomized study at one Belgian university; attendance and feedback use were self-selected, and the language cohorts differed in ways that complicate comparison.
What to watch next
- Randomized replication, independent exam marking, subgroup results, the frequency and severity of feedback errors and evidence that students can recognize incorrect guidance.
Living evidence record
Impact record IAI-1CZN8KE
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Not yet
Record status
Monitoring
Last checked
21 October 2026
Source trail
1 direct source across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Single-source reporting disclosure
This record analyses one direct source. It can establish what arXiv published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 21 October 2026
Research question, topics and regions
Whether language-model feedback on handwritten university mathematics is instructionally sound and followed by better performance on a later exam
Topics: Mathematics education · Formative feedback · Handwriting · Exam performance · Generative AI
Regions: Belgium · Europe
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.