Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 34 of the 34 references it cites were found in a registry or an index; 7 of the 28 sources its claims cite were checked for support.

AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting

Open access

Published version of a preprint

Abstract

Advances in generative artificial intelligence show great potential for improving education. Yet little is known about how this new technology should be used and how effective it can be compared to current best practices. Here we report a randomized, controlled trial measuring college students' learning and their perceptions when content is presented through an AI-powered tutor compared with an active learning class. The novel design of the custom AI tutor is informed by the same pedagogical best practices as employed in the in-class lessons. We find that students learn significantly more in less time when using the AI tutor, compared with the in-class active learning. They also feel more engaged and more motivated. These findings offer empirical evidence for the efficacy of a widely accessible AI-powered pedagogy in significantly enhancing learning outcomes, presenting a compelling case for its broad adoption in learning environments.

Overview

CoChat Summary

TL;DR

In a randomized crossover trial in an undergraduate physics course, students learned more from a custom AI tutor than from an in-class active learning lesson on identical content. The AI condition also produced higher engagement and motivation, while taking less time on task.

Key findings

  • Students in the AI-tutored group had higher median post-test scores than the in-class active learning group: 4.5 (N = 142) vs 3.5 (N = 174), compared with a pre-test baseline median of 2.75 (N = 316).
  • The post-test difference between AI tutoring and in-class active learning was statistically significant: z = -5.6, p < 10–8.
  • In the linear regression model controlling for pre-test score, midterm exam, Force Concept Inventory, prior ChatGPT experience, topic, test version, and time on task, the AI group still performed substantially better; the paper reports p < 10–8 and an effect size estimate of 0.63, with a quantile-regression estimate of 0.73 to 1.3 standard deviations.
  • Students in the AI group reported higher engagement (4.1 vs 3.6; t(311) = -4.5, p < 0.0001) and motivation (3.4 vs 3.1; t(311) = -3.4, p < 0.001); enjoyment and growth mindset were not significantly different.
  • For time on task, the AI group had a median of 49 minutes; 70% spent less than 60 minutes and 30% more than 60 minutes. The paper reports no correlation between time on task and post-test scores.

Methods

  • Randomized, controlled crossover design in Harvard’s introductory physics course (PS2) during Fall 2023; each of the 194 eligible students experienced both conditions across two lessons in consecutive weeks.
  • Two lesson topics were used: surface tension and fluid flow; students took pre-tests and post-tests around each lesson, and the control and experimental conditions used identical content/worksheets, differing only in delivery (in-class active learning vs AI tutor “PS2 Pal”).
  • Main analysis compared post-test scores using a two-sample rank-sum (Mann–Whitney) test and a linear regression model with clustering at the student level and controls for prior knowledge, prior course performance, FCI, ChatGPT experience, topic, test version, and time on task.

Limitations

  • The paper states that the AI tutor’s success may depend on several contextual factors: a heterogeneous student population, high-quality instructional videos, a GPT-4-capable model, expert-crafted prompts, a carefully structured scaffold, and content that lends itself to this format.
  • The authors explicitly say they do not presume structured AI tutoring will always outperform in-class active learning in all contexts, especially those requiring complex synthesis of multiple concepts and higher-order critical thinking.
  • Evident from the text: the study was conducted in a single course at one institution and on two physics topics, so generalizability beyond that setting is not directly tested here.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41598-025-97652-6

Cites

34 of 34 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Comparison of learning gains: A comparison of mean post-test performance between students taught with the in-class active learning and students taught with the AI tutor. Dotted line represents students’ mean baseline knowledge before the lesson (i.e., the pre-test scores of both groups). Error bars show one standard error of the mean.
Fig. 1 Comparison of learning gains: A comparison of mean post-test performance between students taught with the in-class active learning and students taught with the AI tutor. Dotted line represents students’ mean baseline knowledge before the lesson (i.e., the pre-test scores of both groups). Error bars show one standard error of the mean.CC BY-NC-ND · PubMed Central
Fig. 2 AI tutor time on task: Total time students in the AI group spent interacting with the tutor. Dotted line denotes the length of the in-class active learning lesson (60 minutes).
Fig. 2 AI tutor time on task: Total time students in the AI group spent interacting with the tutor. Dotted line denotes the length of the in-class active learning lesson (60 minutes).CC BY-NC-ND · PubMed Central
Fig. 3 Student perception of learning experiences: Level of agreement to statements about perceptions of learning experiences, comparing students taught by in-class active learning and students taught with the AI tutor. Error bars show 1 standard error of the mean. Asterisks above the bars denote P-values generated by dependent t-tests (***p < 0.001).
Fig. 3 Student perception of learning experiences: Level of agreement to statements about perceptions of learning experiences, comparing students taught by in-class active learning and students taught with the AI tutor. Error bars show 1 standard error of the mean. Asterisks above the bars denote P-values generated by dependent t-tests (***p < 0.001).CC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

3 Valid · 2 Not valid · 11 Not checked · 43 Own results and interpretations

18 more claims

Create a free account to see all 59 claims and to verify the 3 references still open.

See all claims — free account
  • Probably not a substantial claim. 26
  • It may state two separate claims as one. 14
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Consistentt(311) = -4.5reported p < .0001 · recomputed p < .001

    “t(311) = − 4.5, p < .0001”

  • Consistentt(311) = -3.4reported p < .001 · recomputed p < .001

    “t(311) = -3.4, p < .001”

2 of 2

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Self-check results appear here once an independent check has confirmed them.

Citations

Cites

Show all 31 references

Research works printed but not linked: 2

  1. Hake, R. R. Interactive-engagement versus traditional methods: A six-thousand-student survey of mechanics test data for introductory physics courses. Am. J. Phys. 66, 64–74 (1998). · DOI · Paper page
  2. Freeman, S. et al. Active learning increases student performance in science, engineering, and mathematics. Proc. Natl. Acad. Sci. 111(23), 8410–8415 (2014). · DOI · Paper page

Cited by

See all 214

Metadata

Authors
Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G.
Affiliations
Harvard University (US)
Corresponding author
Kestin, Greg
Venue
Scientific Reports
Published
2025-06-03
Volume
15
Issue
1
Pages
17458-17458
ISSN
2045-2322 (ISSN-L)
OpenAlex
W4411505840
Cited by
214
References
31
Citations by year
2026: 162 · 2025: 51
FWCI
84.76
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by-nc-nd)
  • Research Square (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
Harvard University
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Intelligent Tutoring Systems and Adaptive Learning
  • Social Sciences › Education › Artificial Intelligence in Education
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
Keywords
AI tutoring, generative AI in education, active learning, university students, learning outcomes, student engagement, student motivation, personalized learning
MeSH terms
Adult, Artificial Intelligence, Female, Humans, Learning, Male, Students, Universities, Problem-Based Learning, Young Adult

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting. https://app.cochat.ai/papers/10.1038/s41598-025-97652-6
The paper
Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G. (2025). AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports. https://doi.org/10.1038/s41598-025-97652-6

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.