The Chatbot Did Not Do the Thinking. The Students Around It Might Have.

Does a chatbot make students more creative, or does it just do the creative bit for them? A new observational study published in PLOS One on 14 October 2026 offers a careful answer: it depends what the students are doing with it. Researchers from Indonesia and Taiwan examined 533 Indonesian students and found that when large language models are used to enable personalized learning, students’ creativity may be enhanced — with the strongest gains in analyzing and creating, the top end of thinking skills. Information verified through 21 October 2026.

Measured against Bloom, not vibes

The team refused to treat creativity as one vague quality. They anchored the work in Bloom’s taxonomy, the framework that runs from remembering and understanding up through applying, analyzing, evaluating and creating. That choice lets the study say something precise. The gains did not sit at the recall end, where a model can simply hand a student a fact. They sat where a student has to pull material apart and build something new from it.

The proposed mechanism is a division of labour. A language model can adapt explanations, examples and difficulty to one learner in real time, offering hints rather than answers and rephrasing concepts until they click. If routine drafting, summarizing and retrieval move to the machine, the student’s effort is pushed upward — evaluating output, questioning its accuracy, refining it toward a goal. Those activities are analyzing and evaluating, almost by definition.

Peer interaction is the study’s second pillar, and it may be the load-bearing one. When students discuss AI-generated content with classmates, compare interpretations and critique one another’s reasoning, they do the evaluative work out loud, together. Personalization tailors the input. Peers sharpen what follows.

Now the caveats, which are real

This was an observational study, not a randomized controlled trial. It captures associations, not proof of cause and effect, and the authors acknowledge that limit. Students who already think creatively may simply be the students most likely to embrace AI-supported personalized learning in the first place. The study cannot rule that out. Anyone quoting it as proof that chatbots boost creativity is quoting past what it says.

A second study, published in Discover Education, shows how far ahead of institutions student practice already runs. Surveying 186 students across Indonesian, English and Hungarian linguistic contexts, it found 78.0 percent reported using generative AI several times per week or daily, and 96.8 percent reported using ChatGPT specifically. Yet the support students asked for was not more technology. The two most requested forms of help were free institution-paid access to AI tools, at 37.1 percent, and clearer institutional AI policies, at 24.2 percent. Classmate support scored a mean of 3.94 on a five-point scale; professor encouragement scored 3.14. Students are teaching each other, in a guidance vacuum.

A third piece of evidence scales the question up. An analysis of PISA 2025 data covering 14,168 Indonesian 15-year-olds in 419 schools found that intensive chatbot use was not associated with lower achievement in science or reading. Rare users scored about 7 points lower than selective users after background was controlled — a gap the authors say may reflect selection rather than any effect of use. In Indonesia, 53% of 15-year-olds reported using AI chatbots at least once a week to help them learn.

What a teacher can actually do with this

Read together, the three studies point the same way. Use is already widespread, at 15 and at university. Heavy use, by itself, shows no achievement penalty in the PISA analysis. The creativity study suggests the gains, where they exist, live in structured work: assignments that require students to analyze and improve AI output, discussion that forces them to defend their changes, personalization that meets a learner at their level instead of a class average.

Analysis: The policy fight in most schools is still framed as allow or ban, and these studies suggest that framing is already obsolete. The variable that matters is design. An unstructured chatbot is a shortcut machine; the same tool, wrapped in tasks that demand evaluation and peer critique, becomes something closer to a sparring partner. The Discover Education survey adds an uncomfortable footnote for institutions: students know the difference, and they are asking — for funded access and clear rules, in that order. Schools that answer with a ban are not protecting thinking. They are declining to teach the skill their students are practising without them every day.

Sources

1. Science Magazine / Scienmag — “Personalized Learning with AI Chatbots May Boost Students’ Creative Thinking — Study of 533 Students” (PLOS One study, published 14 October 2026): https://scienmag.com/personalized-learning-with-ai-chatbots-may-boost-students-creative-thinking-study-of-533-students-suggests/
2. Science Magazine / Scienmag — “Generative AI Is Everywhere on Campus, But Students Say Support Is Lagging” (Discover Education survey, October 2026): https://scienmag.com/generative-ai-is-everywhere-on-campus-but-students-say-support-is-lagging/
3. Research Square — “Profiles of generative AI chatbot use for schoolwork and achievement among Indonesian 15-year-olds: A latent class and multilevel analysis of PISA 2025 data” (2026): https://www.researchsquare.com/article/rs-11408706/v1