Truace
Subscribe
70/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Students using AI chatbots showed more motivation to learn, a review of 42 studies finds

Study title: Effects of generative artificial intelligence on students' learning motivation: evidence from a meta-analysis

Evidence type
Systematic review or meta-analysis
Sample
n = 6,059
Sources
1 sourcePeer-reviewed
Published
2 min read
Truvace record
TRV-2026-2752cite or check it

Study title: Effects of generative artificial intelligence on students' learning motivation: evidence from a meta-analysis

Researchers at East China Normal University and Qingdao University pooled 56 effect sizes from 42 controlled and quasi-experimental studies, covering 6,059 students, published between 2023 and 2025. Students who learned with generative AI showed higher motivation than comparison groups, with an average Hedges' g of 0.764 (95% CI 0.566 to 0.962), which the authors describe as moderate to large.

Read the study abstract

Background: The use of generative artificial intelligence (GenAI) in the education setting has expanded quickly. GenAI may enhance learning motivation by providing immediate feedback, adaptive support, and multimodal resources, yet empirical findings remain inconsistent. Aims: To address the controversy of current studies, a meta-analysis was conducted to examine the overall effect of GenAI on students' learning motivation and test whether effects differed across study contexts. Methods: A random-effects meta-an…

The gain

A meta-analysis of 42 controlled studies with 6,059 students found those learning with generative AI scored higher on motivation than comparison groups (Hedges' g = 0.764), though effects varied widely between studies.

In brief

Results varied a great deal between studies (I2 = 93.7%), and the 95% prediction interval ran from -0.689 to 2.217. The authors say the pooled figure is a positive average rather than proof of a stable effect in every classroom, and that the findings do not support the claim that GenAI reliably raises motivation everywhere.

Effects looked larger among university students (g = 0.886) than secondary (0.360) or primary pupils (0.311), larger in language learning (0.946) than science (0.109), and larger on mobile devices (0.782) than computers (0.303). These subgroup comparisons are exploratory, because the factors are study-level traits rather than manipulated conditions.

Main points

  1. The review pooled 56 effect sizes from 42 studies with 6,059 participants; sample sizes ranged from 32 to 486 students.
  2. Average effect on learning motivation was g = 0.764 (95% CI 0.566 to 0.962), with heterogeneity I2 = 93.7% and a prediction interval from -0.689 to 2.217.
  3. Effects were larger in collectivist contexts (g = 0.815, 49 effect sizes) than individualist ones (g = 0.367, 7 effect sizes).
  4. University students showed g = 0.886, versus 0.360 for secondary and 0.311 for primary students; language studies (0.946) outpaced science (0.109, not significant).
  5. Mobile-device use (g = 0.782) beat computer use (g = 0.303); GenAI function, intervention length, motivation type, student gender and publication year were not significant moderators.
  6. Leave-one-out checks kept the pooled effect between 0.695 and 0.840, and Egger's test (p = 0.057) gave no clear evidence of publication bias.
  7. Included studies scored 18 to 23 out of 24 on a 12-item quality scale (mean 20.38).

What the review found when it combined the studies

These are the review’s own pooled results, copied from its abstract. Truvace did not recalculate them.

  • Student learning motivation with generative AI compared with other learning conditions

    Hedges' g 0.764 (95% CI 0.566 to 0.962)

    42 studies · 6,059 people

    The review’s own words
    As Hedge's g = 0.764, 95% CI = [0.566, 0.962], 95% PI = [−0.689, 2.217]; z = 7.571, p < 0.001, students using GenAI showed more learning motivation than other students.

  • Learning motivation with generative AI among secondary school students

    Hedges' g 0.360 (95% CI 0.157 to 0.563)

    The review’s own words
    The effect was stronger among university students ( g = 0.886, 95% CI = [0.637, 1.136]), followed by secondary school students ( g = 0.360, 95% CI = [0.157, 0.563]), and finally primary school students ( g = 0.311, 95% CI = [0.077, 0.544]).

  • Learning motivation with generative AI in collectivist-culture countries

    Hedges' g 0.815 (95% CI 0.585 to 1.046)

    The review’s own words
    This effect was larger in countries with predominantly collectivist cultures ( g = 0.815, 95% CI = [0.585, 1.046]) than those with individualistic ones ( g = 0.367, 95% CI = [0.147, 0.587]).

  • Learning motivation with generative AI used on computers

    Hedges' g 0.303 (95% CI 0.169 to 0.438)

    The review’s own words
    Using GenAI on mobile devices ( g = 0.782, 95% CI = [0.545, 1.018]) had a stronger effect on learning motivation than using it on computers ( g = 0.303, 95% CI = [0.169, 0.438]).

  • Learning motivation with generative AI among university students

    Hedges' g 0.886 (95% CI 0.637 to 1.136)

    The review’s own words
    The effect was stronger among university students ( g = 0.886, 95% CI = [0.637, 1.136]), followed by secondary school students ( g = 0.360, 95% CI = [0.157, 0.563]), and finally primary school students ( g = 0.311, 95% CI = [0.077, 0.544]).

  • Learning motivation with generative AI used on mobile devices

    Hedges' g 0.782 (95% CI 0.545 to 1.018)

    The review’s own words
    Using GenAI on mobile devices ( g = 0.782, 95% CI = [0.545, 1.018]) had a stronger effect on learning motivation than using it on computers ( g = 0.303, 95% CI = [0.169, 0.438]).

  • Learning motivation with generative AI among primary school students

    Hedges' g 0.311 (95% CI 0.077 to 0.544)

    The review’s own words
    The effect was stronger among university students ( g = 0.886, 95% CI = [0.637, 1.136]), followed by secondary school students ( g = 0.360, 95% CI = [0.157, 0.563]), and finally primary school students ( g = 0.311, 95% CI = [0.077, 0.544]).

  • Learning motivation with generative AI in language-related subjects

    Hedges' g 0.946 (95% CI 0.630 to 1.262)

    The review’s own words
    Observed between-subgroup differences across subject domains were found ( Q BET = 23.785, df = 4, p < 0.05), with language-related studies showing the largest pooled effect ( g = 0.946, 95% CI = [0.630, 1.262]), followed by mathematics ( g = 0.711, 95% CI = [0.356, 1.066]) and pedagogy ( g = 0.697, 95% CI = [0.261, 1.133]), then information ( g = 0.291, 95% CI = [0.047, 0.535]), and lowest in science ( g = 0.109, 95% CI = [−0.107, 0.325]).

  • Learning motivation with generative AI in individualistic-culture countries

    Hedges' g 0.367 (95% CI 0.147 to 0.587)

    The review’s own words
    This effect was larger in countries with predominantly collectivist cultures ( g = 0.815, 95% CI = [0.585, 1.046]) than those with individualistic ones ( g = 0.367, 95% CI = [0.147, 0.587]).

Code checks every number above against the review’s text before it is shown. The short labels are an AI model’s reading of that text, so the review’s own sentence sits under each one. A pooled figure is only as strong as the studies behind it. How Truvace uses pooled results

The full story

The authors searched Web of Science, ERIC, ProQuest, EBSCO and Scopus for studies from January 2023 to November 2025, finding 2,365 records plus 74 from citation chasing. After removing 541 duplicates and screening, 42 studies met the criteria: students from primary school to university, a GenAI-supported group, a non-GenAI comparison group and a quantitative motivation score.

Two authors independently coded every study and rated quality (Cohen's kappa 0.83 for quality ratings, 0.91 for coding). Effects were computed from post-test means and standard deviations. A three-level model was tested for dependent effect sizes, but a simpler two-level random-effects model was kept because the extra variance component did not improve fit.

The pooled effect was positive for both intrinsic and extrinsic motivation, but the difference between the two was not significant. Language-learning studies made up 32 of the 56 effect sizes and university samples 43, so the average leans heavily on those settings.

The authors caution that the moderator results show association, not cause, and call for future work on the instructional, learner and contextual conditions in which GenAI best supports motivation.

Sources

1 on file
  1. Frontiers in Psychologydoi:10.3389/fpsyg.2026.1832451
    Peer-reviewed2026-09-24

Talk it over

Four questions to start onWho gets this benefit, and who was left out of the study or story?

Start with a question

Students using AI chatbots showed more motivation to learn, a review of 42 studies findshttps://truvace.com/item/effects-of-generative-artificial-intelligence-on-students-learning-motivation-ev

  1. Who gets this benefit, and who was left out of the study or story?

    The gain A meta-analysis of 42 controlled studies with 6,059 students found those learning with generative AI scored higher on motivation than comparison groups (Hedges' g = 0.764), though effects varied widely between studies.

    Answer in the thread

  2. How do we know? What can this kind of evidence show, and what can’t it?

    The evidence This rests on a systematic review or meta-analysis with a sample of 6,059, from Frontiers in Psychology.

    Answer in the thread

  3. What does it leave open? What else would you want to know before acting on it?

    The record’s own caveat The very high heterogeneity (I2 = 93.7%) means the average effect does not hold across all contexts, and the subgroup differences are exploratory because they reflect study-level characteristics rather than manipulated conditions.

    Answer in the thread

  4. Have you seen AI used this way where you study, work or live? Did it go the same way?

    Answer in the thread

The questions are the same on every record. The details under them are taken from this record as written.

What readers say

More from Education