- Who it helps
- Educators selecting AI tools and learners checking whether assistance improves understanding.
- Availability
- The working paper is publicly readable. It studies a fall-2025 deployment using GPT-4o and course materials; it does not release a generally available tutoring product.
What the trial found
The authors report lower grades and learning-platform activity in the matched-course treatment comparison. Around 15% of students offered the tutor used it at least once. Students in both conditions could also use other chat tools.
Those details matter when interpreting the effect. The paper does not isolate every possible tutoring design or establish a single mechanism for the outcome.
A practical consequence for study sessions
Use a question you can answer without assistance before the session. After the explanation, solve a changed problem with the chat closed, and repeat a fresh question the following day. Keep the score and your explanation of the reasoning.
The existing one-chapter study guide already includes a baseline, a changed problem and delayed recall. Those checks remain useful; this study evaluated a different tutor and supplies no product-specific verdict on that guide.
What you can do
- Read the paper’s intervention and usage sections before applying its headline finding to another tutoring setup. Open reference ↗
Where the evidence stops
A personal before-and-after check cannot establish causality. The paper was reviewed as a working paper, and its analysis was not independently reproduced.
Sources read. Pages 0–6 were read on October 17 UTC. The linked study, our study exercise and the separate Tutor application are distinct.
Sources & further reading
Web · paper
AI tutor randomized trial working paper ↗
Working-paper pages 0–6 read, including the treatment, usage and course outcome limits.
AI tutor trial authors / EdWorkingPapers · page checked 2026-10-17
More sources reviewed (1)
Web · analysis
AI in Education Weekly — October 13 ↗
Independent comparison read after discovery; the audit covers its five Main Stories.
Sidaard Gunasekaran · page checked 2026-10-17
Coverage note: Primary and linked sources are limited to those listed. Independent execution, original X coverage and comparable cross-platform measurements were not established in this run.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
A working paper studies a course-specific AI tutor offered to 2,379 undergraduates across 30 instructors. The result concerns that intervention and its setting.