DOI: 10.12688/f1000research.192825.1 ISSN: 2046-1402

How Does Generative AI Support Inquiry-Based Learning? A Systematic Review

Fahmi Yahya, Rusijono Rusijono, Utari Dewi, Siti Uswatun Khasanah, Lathifatuddini Binti Rusdi, Dini Amalia, Magfirotul Hamdiah, Nila Syarifun Nisak
Show PDF Cite
Background Teachers and learners now use generative artificial intelligence (GenAI) in inquiry-based learning (IBL), but we know little about which inquiry phases GenAI supports, who uses it and what outcomes studies report across disciplines. Methods Following PRISMA 2020, we searched Scopus on 6 October 2026 for empirical journal articles published in English from 2022 to 2026. Of 267 records screened, we sought 110 reports, assessed 89 in full text and included 66 studies. We coded the studies with a 29-variable codebook and mapped GenAI use onto the five inquiry phases (orientation, conceptualization, investigation, conclusion and discussion). We synthesized the findings descriptively and did not appraise study quality. Results Most studies appeared in 2025–2026 and took place in higher or teacher education; more than half came from fields outside science. Studies used ChatGPT or other OpenAI GPT models in 71.2% of cases, mostly as unmodified chatbots, and GenAI served most often as an information source (60.6%). Studies reported GenAI support most often in the investigation phase (68.2%) and least often in orientation (30.3%); 11 studies covered all five phases, and for most phases about half of the studies did not say whether GenAI was used. Half of the studies reported positive outcomes. Thirteen of the 14 comparisons with instruction without AI were positive, whereas comparisons between GenAI designs split evenly between positive and mixed results. Inaccuracy (31.8%) and over-reliance (24.2%) were the challenges studies reported most often. Conclusions GenAI mainly supplies information during investigation and seldom supports the full inquiry cycle. Some comparative evidence favors GenAI designs that withhold direct answers or add metacognitive scaffolds, but this evidence is thin. Future studies should report the phases GenAI supports, measure inquiry skills directly and compare GenAI designs with one another.

More from our Archive