Personalized Learning with AI Chatbots May Boost Students’ Creative Thinking, Study of 533 Students Suggests

Large language models are rapidly reshaping classrooms around the world, but one of the most contested questions in education research has been whether these tools genuinely make students think more creatively or simply encourage them to outsource their thinking. A new observational study published in PLOS One adds a notable data point to that debate. 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 most pronounced gains appearing in the higher-order thinking abilities of analyzing and creating. The findings, published on 14 October 2026, come at a moment when schools and universities are still deciding how, and how much, generative artificial intelligence should be woven into everyday instruction.

The research team framed their investigation around a long-standing question in educational psychology: what does creativity actually look like in an era when a machine can generate essays, solutions, and ideas on demand? Rather than treating creativity as a single, vague quality, the study anchored its analysis in Bloom’s taxonomy, the widely used framework that organizes cognitive skills into levels ranging from remembering and understanding to applying, analyzing, evaluating, and creating. By mapping creative performance onto this hierarchy, the researchers could distinguish between lower-order engagement with AI tools and the higher-order skills that educators most want to cultivate. The study’s focus area, illustrated in figures accompanying the paper, centers on how this taxonomy interacts with peer interaction and AI-supported personalization to elevate thinking skills.

Personalized learning sits at the heart of the study’s rationale. Traditional classrooms often force students to move through material at the same pace, which can leave some learners behind while others grow bored. Large language models, by contrast, can adapt explanations, examples, and difficulty to an individual student’s needs in real time, offering hints rather than answers, rephrasing concepts until they click, and generating practice problems matched to a learner’s current level. The researchers hypothesized that this adaptive scaffolding could free students to spend more cognitive effort on the upper tiers of Bloom’s taxonomy, where analysis and creation happen, instead of getting stuck on foundational recall and comprehension.

The observational design of the study is important for interpreting its conclusions. The researchers followed 533 Indonesian students and assessed how engagement with large language model-supported personalized learning related to their creative abilities. Because the study was observational rather than a randomized controlled trial, it captures associations rather than definitive proof of cause and effect, a limitation the authors acknowledge. Nevertheless, the scale of the sample and the structured framework used to evaluate creativity give the findings weight in a field where much of the existing evidence consists of small-scale experiments or anecdotal classroom reports. The pattern that emerged pointed toward a meaningful enhancement of creativity, particularly in the analyzing and creating dimensions of higher-order thinking.

Why would analyzing and creating benefit most? The researchers’ framework suggests an answer rooted in how AI tools change the division of cognitive labor. When a language model handles routine drafting, summarizing, and information retrieval, students are pushed to evaluate the output critically, question its accuracy, and refine it toward their own goals. Those activities correspond directly to the analyzing and evaluating levels of Bloom’s taxonomy. Similarly, creating, the taxonomy’s highest tier, may be stimulated when students use AI-generated material as raw material for original work rather than as a finished product. In this view, the model acts less like an answer machine and more like a collaborator whose contributions must be interrogated, recombined, and improved.

Peer interaction emerges as the second pillar of the study’s framework. The paper’s title, which references both Bloom’s taxonomy and peer interaction, signals that the researchers see creative growth as a social process as well as a technological one. When students discuss AI-generated content with classmates, compare their interpretations, and critique one another’s reasoning, they engage in exactly the kind of evaluative dialogue that educational theorists have long associated with deeper learning. The combination of adaptive AI support and collaborative peer exchange may therefore be more powerful than either element alone, with technology personalizing the input and peers sharpening the thinking that follows.

The context of the study is also significant. Indonesia’s education system has been actively pursuing digital transformation, and the research team’s funding reflects institutional investment in this direction. The publication charges were supported by Universitas Padjadjaran through the Indonesian Endowment Fund for Education on behalf of the Indonesian Ministry of Higher Education, Science and Technology, managed under the EQUITY Program. Research from middle-income countries is underrepresented in the global literature on educational AI, which has been dominated by studies from North America, Europe, and East Asia. Evidence from Indonesian classrooms helps address that gap and suggests that the potential creative benefits of large language models are not confined to well-resourced Western institutions.

The authors, affiliated with institutions in Indonesia and Taiwan, published their work open access in PLOS One under a CC-BY 4.0 license, and they declared no competing interests. The journal’s peer review process means the study’s methods and claims have undergone external scrutiny, though as with any observational research, readers should weigh the findings alongside the design’s inherent limits. Observational studies cannot fully rule out the possibility that students who already possess stronger creative dispositions are simply more likely to embrace AI-supported personalized learning in the first place. Randomized interventions will be needed to establish causality with confidence, and the authors’ framework provides a testable structure for such future trials.

For educators, the practical implications are cautiously encouraging. The study suggests that banning or restricting large language models outright may forgo genuine learning benefits, while unstructured use may not automatically produce gains either. The more promising path indicated by the research is deliberate integration: designing assignments that require students to analyze and improve AI output, encouraging peer discussion around that work, and using the adaptive capabilities of these models to tailor support to individual learners. In such environments, the technology becomes a scaffold for higher-order thinking rather than a substitute for it, and the creative capacities at the top of Bloom’s taxonomy, analyzing, evaluating, and creating, become the explicit targets of instruction.

As generative AI continues to spread through education systems worldwide, studies like this one help move the conversation beyond polarized narratives of either technological salvation or cognitive decline. The evidence from 533 Indonesian students indicates that the outcome may depend less on the technology itself than on the pedagogical structures wrapped around it. When personalization, critical engagement, and peer collaboration are combined, large language models might do more than help students learn faster; they might help them think more creatively. Confirming that promise will require experimental follow-up across diverse educational settings, but this study offers a concrete, theoretically grounded starting point for the next phase of research into how artificial intelligence and human creativity can develop together.

Subject of Research: The effect of large language model-supported personalized learning on students' creativity and higher-order thinking skills

Article Title: When LLMs enable personalized learning, they might enhance students' creativity, particularly by boosting analyzing and creating abilities, per observational study of 533 Indonesian students

Article References: When LLMs enable personalized learning, they might enhance students' creativity, particularly by boosting analyzing and creating abilities, per observational study of 533 Indonesian students. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: large language models, personalized learning, creativity, Bloom's taxonomy, higher-order thinking, peer interaction, education technology, observational study, Indonesia, PLOS One, generative AI, student learning