Free ChatGPT use boosts originality on the assisted creative task but the advantage disappears when assistance is removed¶

Claim · All claims
Evidence · 2 studies · 1 causal r3, 1 theoretical r2 · q2–q3 · i0 negligible
Currency · newest study 2026 · checked against its sources 2026-10-15 (automated, 2 of 2 entries) · not reviewed by a person

Subclaims¶

q3 i0 On the product improvement task with ChatGPT available, the general-AI group produced more original solutions than the human-only group (d = 0.79), but on the subsequent unassisted product invention task the general-AI group did not differ from the human-only group (d = 0.04); the i0 code reads this unassisted result. → Wong 2026

q2 i? The paper reports, citing Wong and Qiu (2026), that university students using ChatGPT freely produced the most creative work on the assisted task, but on a later task with no ChatGPT the think-first group outperformed the free-ChatGPT group. → Halani 2026

Evidence¶

Wong 2026¶

Wong, S. S. H., & Qiu, S. X. (2026). Think First, ChatGPT Later: Guiding Human–AI Collaboration for Learning Gains in Independent Human Creativity. Educational Psychology Review. https://doi.org/10.1007/s10648-026-10118-7

q3 · i0 · negligible difference on the unassisted task, d = 0.04 (the assisted-task d = 0.79 is not what this code reads) · causal · r3

A 3 (Learning Group) × 2 (Creative Task) mixed ANOVA on originality scores from 196 university students randomly assigned to human-only, general-AI, or regulated-AI groups. The general-AI group "produced more original solutions than both the human-only" and regulated-AI groups on the assisted task. However, "the general-AI group did not differ from the human-only group in their originality after the intervention, p = 1.00, d = 0.04."

"the general-AI ( M = 4.08, SD = 1.22) group produced more original solutions than both the human-only ( M = 3.06, SD = 1.37) and regulated-AI (M = 3.30, SD = 1.30) groups, p <.001 and p =.002, d = 0.79 and − 0.62"

Halani 2026¶

Halani, A. (2026). Designing for Reach: Seven Levers and the Student Alone with AI. Tang Institute at Phillips Academy. https://doi.org/10.35542/osf.io/t4cd2_v1

q2 · i? · theoretical · r2

The paper reports this secondhand, citing Wong and Qiu (2026): university students improved a creative product alone, with free ChatGPT, or under a think-first regimen; on the assisted task the free-ChatGPT group produced the most creative work and the think-first group showed no advantage.

"On a later task with no ChatGPT, the pattern reversed: the free-ChatGPT group fell back to the level of those who had worked alone, while the think-first group produced more creative responses than the other groups (Wong & Qiu, 2026)."

Studies bearing on this claim¶

Does using generative AI while producing creative work leave learners more original when they later work without it? This table gathers the studies behind this claim and the 5 narrower claims that bear on it: 6 distinct studies (2 review, 1 causal, 1 theoretical, 1 design, 1 associational), q2–q3, 1 of 6 with an effect size coded. One row per study, codes as each entry records them; no effect is averaged and no verdict is drawn. Built by scripts/synthesize_claim.py from eval/synthesis/; a row that looks wrong is fixed on the claim it comes from.

Discussion¶

Learner Variables¶

• Creativity — outcome: instruction changes it

Related Claims¶

• The educational value of GAI is contingent on pedagogical integration, not on the technology itself — the general claim this finding bears on, with the studies behind it

• AI severs the correlation between artefact quality and cognitive engagement, breaking the artefact-as-proxy for learning — the general claim this finding bears on, with the studies behind it

• The regulated 'think first, ChatGPT later' group outperformed both other groups in independent originality on the unassisted task despite no assisted-task advantage — related

• Collaborative prompt use mediated the regulated-AI group's advantage over the general-AI group in independent originality — related
• The regulated-AI group sustained originality across tasks while both other groups declined from the easier to the more complex task — related
• Unguarded answer-giving AI harmed unaided exam performance while a guarded version of the same model erased the harm (Bastani et al., 2025, as reported) — related
• AI-assisted peer feedback enhanced lexical diversity only in the first writing task, with the group advantage disappearing by the second task — related
• The general-AI group most often dictated ChatGPT to directly generate solutions while the regulated-AI group more frequently collaborated to improve self-generated ideas — related
• ChatGPT sustains coherent, human-like multi-topic conversation and produces convincing creative content, as shown by the authors' own prompting test — related
• ChatGPT use improves the quality, elaboration, and originality of university students' solutions to a complex ill-defined problem-solving task — related
• ChatGPT use does not change absolute self-evaluation accuracy, but self-evaluation judgments fail to correlate with performance under ChatGPT assistance — related
• AI-assisted performance during skill-building does not transfer to subsequent unassisted performance — a broader claim this one bears on
• Primary studies report possible benefits of GenAI for children's divergent thinking, narrative creativity and creative self-efficacy, without confirmed efficacy — related
• Generative AI use shows divergent effects on creativity: weakened creative writing but increased creative problem-solving and idea quality — related