The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2763
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T17:20:35.868809Z
status: published
lens: trace
sector: education
headline: Evidence of impact and interpretational limits of generative AI in STEM education: a systematic review and meta-analysis on cognitive learning outcomes
dek: Abstract This systematic review and meta-analysis examines the impact of generative artificial intelligence (GAI) on cognitive learning outcomes in STEM education. Prior research is growing but remains fragmented, often focusing on usability or single tools like ChatGPT rather than domain-specific cognitive effects. We therefore address these gaps by examining (1) the extent to which interactions with GAI enhance learning effectiveness and possible moderators, (2) what challenges learners face when interacting w…
gain_title: A meta-analysis of 49 studies found that generative AI looked beneficial for STEM learning overall, and a bias-corrected model still pointed to promise when AI augmented student activities and targeted knowledge gains.
problem_title: A systematic review and meta-analysis of 49 STEM education studies found the overall positive effect of generative AI was largely attributable to publication bias (bias-corrected effect 0.076 ± 0.254), with huge variation between studies and possible harm when AI substituted for student activity.
trace_subject: Generative AI and cognitive learning outcomes in STEM education
gain_reading: A meta-analysis of 49 studies found that generative AI looked beneficial for STEM learning overall, and a bias-corrected model still pointed to promise when AI augmented student activities and targeted knowledge gains.
gain_evidence: GAI appears promising in STEM if knowledge gains are targeted as learning outcomes, and when GAI is used to augment students’ learning activities
problem_reading: A systematic review and meta-analysis of 49 STEM education studies found the overall positive effect of generative AI was largely attributable to publication bias (bias-corrected effect 0.076 ± 0.254), with huge variation between studies and possible harm when AI substituted for student activity.
problem_evidence: the overall positive effect of GAI in STEM education can be largely attributed to publication bias
quick_read: A systematic review and meta-analysis pooled 59 effect sizes from 49 controlled studies on how generative AI affects cognitive learning in STEM subjects. A standard random-effects model showed an overall positive effect, but results varied enormously between studies (I² = 96.32%), with a prediction interval running from Hedges' g of -1.52 to 3.20.

After correcting for publication bias with a Robust Bayesian Meta-Analysis, the average effect shrank to 0.076 ± 0.254, so the headline benefit is largely a product of which studies get published. Funnel plot asymmetry also pointed to possible publication bias.

The size of the effect depended on what was measured and how AI was used. Knowledge outcomes did better than skills, and AI that augmented student activities did better than AI that replaced them. Thirty-three studies reported large effects only because intervention and control groups did cognitively different activities, so the comparisons were not like for like.
limitation: Substantial unexplained heterogeneity remains, and evidence on learner challenges and supportive interventions was limited and inconsistently reported, so those findings are qualitative rather than generalizable effect estimates.
tag: Dual reading
key_points: The review screened studies published after 2017 with a comparison or control group and found 85 eligible studies; 49 studies with 59 effect sizes qualified for meta-analysis. | Searches covered ERIC, PsycINFO and Web of Science, updated May 7, 2026, plus citation tracking; most studies were in higher education and used text-based tools such as ChatGPT. | The random-effects model showed a positive overall effect, with heterogeneity of I² = 96.32% and a prediction interval from g = -1.52 to g = 3.20. | The bias-corrected estimate was 0.076 ± 0.254, with remaining heterogeneity (τ = 1.190 ± 0.166) not explained by publication bias. | Learning outcome type (knowledge versus skills) explained 12.5% of heterogeneity and the level of cognitive activity in the AI condition versus control explained 13.0%; both held up after bias correction. | Knowledge as an outcome was almost, but not strictly, a necessary condition for large effects, and no combination of outcome type and activity level was sufficient for them. | Thirty-three studies reported large effects only because student activities in the two groups were not comparable. | The authors propose six testable hypotheses and an integrative theoretical framework; the protocol was preregistered on AsPredicted.
rundown: The authors meta-analyzed externally assessed cognitive outcomes and narratively synthesized learner challenges and supportive interventions where pooling was not feasible. Two pairs of raters independently screened and coded the studies, which had to be peer-reviewed, quantitative, and include a comparison group.

The moderator analysis found that the type of learning outcome and the interactive-constructive-active-passive level of the intervention versus the control explained parts of the variation. A coincidence analysis found no combination of the two that was enough on its own to produce large effects.

The review reports that evidence on learner challenges and effective interventions was limited and inconsistently reported. Studies systematically underreported learner-level variables such as AI literacy and metacognitive skills, and process-level factors such as task delegation, prompt quality and verification behaviours.
sources:
- peer_reviewed | Artificial Intelligence Review | https://doi.org/10.1007/s10462-026-11665-9 | 2026-08-25
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

ee482cc30edc2938a6ec99d711c86e649fd6630766d5b6981d23cb049a6b44bc

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000