Title: The Impact of Differing LLM Accessibility on Essay Composition Practices

Abstract

Understanding how large language models (LLMs) influence higher education pedagogy is crucial for developing integration strategies that enhance, rather than hinder, student academic outcomes. This research explores the influence of different degrees of LLM support on writing quality, student engagement, and the sense of authorship. In a pilot experiment, 24 university students were randomly allocated to one of three conditions: no LLM access, restricted access (limited to three prompts with responses capped at 100 words), or unrestricted access.

The analysis revealed no statistically significant differences in overall essay quality among the three groups. However, distinct variations emerged regarding writing behavior and perceived authorship. Participants in the limited-access group demonstrated a stronger sense of ownership, with 62.5% indicating they would submit their essays as independent work, compared to only 25% in the unlimited-access group. Additionally, the restricted group reported greater improvements in organization and utilized prompting strategies that were more strategic and focused on revision.

Conversely, students with unlimited access spent more time writing, generated essays that closely resembled LLM-generated text, and noted a decline in creative expression. These results imply that imposing constraints on LLM usage, rather than prohibiting it entirely, may help maintain students’ confidence in their authorship while still leveraging the supportive scaffolding provided by AI tools.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC