Evidence · strength 1 of 5 · searched to 2 Nov 2026
Does AI guidance for tutors and teaching assistants help students?
Not enough evidence yet to say whether AI guidance for tutors helps their students learn. 1 of 2 studies found a clear gain in student learning.
What trials found when human tutors or teaching assistants worked with AI suggestions during tutoring and feedback, and what happened to their students.
VerdictNot enough evidence yet
No good evidence
ImpactNo clear effect
1 of 2 studies found a clear gain in student learning.
Evidence strength1 of 5
We have very low confidence in this estimate.
Cost per user a year$
The tools ran on general AI models; no study counts the full cost.
Pilot rating under our v0.3 rules. The rules are published on our method page but not yet independently reviewed, so ratings may change as the method is checked and new studies arrive. Our method.
What the studies found
Learning (headline)
No clear effect- Wang et al. (2026): +4 points Students passing the session exit ticket (topic mastery), tutor with AI co-pilot vs without (p < 0.01; 62% to 66%).
Students working with treatment tutors were 4 percentage points (p.p.) more likely to pass their exit tickets
- Wang et al. (2026): not significant End-of-year NWEA MAP maths score (initial preregistered outcome). Not statistically clear.
we did not find statistically significant improvements in end-of-year math test scores
- Lu et al. (2026): −0.7 points Post-test score after revision (blind-graded optional quiz), AI-mediated v human-only feedback (coefficient -0.007, t = -1.02, p = 0.31). Not statistically clear.
we found that being in the two different conditions did not affect students' post-test scores (β = -0.007, t = -1.02, p = 0.31).
Quality
Moderate gain- Lu et al. (2026): +0.5 SD Final (revised) essay quality, AI-mediated TA feedback v handwritten TA feedback (mixed-effects model) (coefficient 0.05 (5 points on a 0-100% rubric score), p < 0.001).
The effect size is Cohen's d = 0.5, roughly equilavent to moving a student from the 50th to the 70th percentile.
A minus sign on time means the task took less time. Bands are provisional; see how impact is banded. Effects are not pooled yet: each band is set from one result per study, and where studies use different units we show only the direction.
How sure are we?
- Starting level2 of 5, from 2 studies graded A, B or C.
- Down one: old models0 of 2 studies show they used a model released since May 2025 (the 18-month window); the rest used older models or do not say which.
- Consistencywith 2 studies, never downgraded on this test alone.
- Publication biasnot testable with fewer than 10 studies.
- Result1 of 5. We have very low confidence in this estimate.
Behind the average
- Students of lower-rated tutors (Wang et al. (2026)): +9 percentage points (56% to 65%) passing exit tickets.
How, where and why it works
Why it works
2 of 4. The AI suggests what an expert tutor might say; in the largest trial, students of lower-rated tutors gained most.
For whom and where
2 of 4. Students of lower-rated tutors gained most in the largest trial.
What it takes
2 of 4. Tutors chose whether to use each suggestion during live sessions.
The full cost
1 of 4. Costs are mentioned but not counted in full.
Four separate scores from 0 to 4, never added up: how much the studies tell us about why it works, for whom, what it takes and what it really costs (the EMMIE method).
What to do next
Measure your pilot →If you are trying it yourself
- Use suggestions as prompts, not scripts.
- Check students' understanding without the AI.
If you are rolling it out
- Start with less experienced or lower-rated tutors, where the largest trial found most gain.
- Measure student learning, not just session quality.
What it takes
- Tutors see suggestions during the session and decide what to use.
What can go wrong
- Better-looking work during the session did not always mean more learning on a later test.
The studies
| Study | Grade | Design | People | AI model (released) | Company role |
|---|---|---|---|---|---|
| Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise Rose E. Wang, Ana T. Ribeiro, Carly D. Robinson, Susanna Loeb, Dora Demszky, 2026. arXiv 2410.03017 (preprint) | B | field RCT, deployed work | 783 tutors (388 treatment, 395 control) and 1,013 students in grades 3-6, nine schools in a southern US district | GPT-4 (OpenAI), inside a tutor-facing assistant (Mar 2023) | none |
| AI-Mediated Feedback Improves Student Revisions: A Randomized Trial with FeedbackWriter in a Large Undergraduate Course Xinyi Lu, Kexin Phyllis Ju, Mitchell Dudley, Larissa Sano, Xu Wang, 2026. CHI '26 (journal) | C | RCT, deployed work | 354 students in 11 discussion sections of an introductory economics course, 11 TAs; 1,366 essays graded | OpenAI GPT-4o | none |
Grades: A, a well-run randomised trial in real work; B, a strong natural experiment in real work or a well-run trial on realistic tasks; C, a trial with clear limits or a matched comparison; D, before-and-after, surveys or case studies. How studies are graded.
Does the rating hold up?
- Run without company-run studiesNo change: no such studies.
- Run without working papers and preprints1 studies: strength 0, verdict no good evidence, headline band no clear effect.
How up to date is this evidence?
Method →Evidence searched to 2 Nov 2026. Studies cover models released Mar 2023 to Mar 2023. Rating computed 5 Nov 2026 under Module 15 v0.3 (pilot).
How this page was made: AI research agents searched for studies and read each paper; a second pass checked every number on this page against the paper itself, with the quote and where it appears. The rating is computed by fixed rules, not chosen. No person outside AI Adoption has reviewed it yet. Editorial owner: AI Adoption editors. Spot a mistake? Corrections.
Disclosure: AI Adoption uses AI tools, including Claude from Anthropic, in its research and editorial work. No company pays for or has any say in a rating.