Reducing stress, but limiting novelty and diversity: exploring generative AI’s roles in writing

Abstract Learning expository writing is critical to students’ articulation of complex concepts, and thus to their long-term academic and professional success. Developed collaboratively with educators, this five-week quasi-experimental classroom intervention compared two intact eighth-grade classes: one in a generative AI (GenAI)-supported condition ( n = 36) and one in a search-engine condition ( n = 37). Comparison of the two groups’ pre- and post-intervention compositions suggested that students in the GenAI-supported condition performed better in terms of focus, structure, content, and language, but tended to write less-complex sentences and showed smaller gains in novelty. Heart-rate variability (HRV) data collected during writing indicated stress fluctuations in both groups, but a net reduction in stress among those students who used GenAI. Stimulated-recall interview data suggested that GenAI reduced stress in four ways: by aligning task demands with students’ capabilities, providing immediate and explanatory feedback, supporting learner autonomy, and easing the mental effort required to complete tasks. Lastly, exploratory partial least squares structural equation modeling showed that the GenAI-supported condition was associated with both higher HRV and greater writing improvement. Methodologically, this study’s key contribution is its combination of HRV-based visualization of stress patterns during writing with automated textual analysis, drawn from computational linguistics, to examine changes in lexical diversity and sentence complexity. Theoretically, it highlights HRV-based stress as a possible pathway linking GenAI-supported writing conditions to writing-performance improvement.