RtR-AI: Reading the Reader’s Mind through Eye Tracking — Can AI Generated Texts Match Human Authors?
Published in 2025 Symposium on Eye Tracking Research and Applications (ETRA 2025), ACM, 2025
Abstract
While Generative AI models like Large Language Models (LLMs) are capable of generating extensive text, their efficacy in producing readable content for human participants in experimental settings remains to be evaluated. Eye-tracking technology is increasingly utilised to study cognition and behaviour, yet its application to readers’ cognitive processes when exposed to AI-generated versus human-authored texts remains unexplored.
This study investigates how text generated by LLMs influences reading by analysing gaze patterns. Gaze data were collected from 13 participants as they read AI-generated and human-authored passages. A comparative within-subjects analysis assessed gaze patterns between authors and between text types using the robust I2MC (Identification by 2-Means Clustering) algorithm to identify fixations. Pupil dilation and reading speed were also examined.
Findings reveal significant differences in fixation characteristics not only between individual authors but also between AI-generated and human-authored text conditions — providing empirical evidence that the cognitive processes engaged during reading differ depending on the origin of the text.
Key Contributions
- First eye-tracking investigation directly comparing cognitive reading behaviour between AI-generated and human-authored texts
- Within-subjects gaze analysis using the validated I2MC fixation detection algorithm
- Significant fixation characteristic differences identified between text types and between individual authors
- Pupil dilation and reading speed as additional converging measures of cognitive load during reading
- Contributes to the growing evidence base on human perception of AI-generated content
Methodology
| Component | Detail |
|---|---|
| Participants | 13 readers, within-subjects design |
| Eye-tracker | Tobii (gaze + pupil data) |
| Fixation algorithm | I2MC (Identification by 2-Means Clustering) |
| Measures | Fixation count, fixation duration, pupil dilation, reading speed |
| Text conditions | AI-generated (LLM) vs. human-authored passages |
| Analysis | Comparative gaze pattern analysis across authors and text types |
Research Context
This work is conducted within the “Reading the Reader” (RtR) project, funded by the Novo Nordisk Foundation and hosted at DTU Compute, Cognitive Systems Section. The project investigates how typographic and textual features influence perceptual and cognitive processes during reading using psychophysics and advanced eye-tracking analysis.
Venue
Presented at the 2025 Symposium on Eye Tracking Research and Applications (ETRA ‘25), Tokyo, Japan, 26–29 May 2025. Published open access by the Association for Computing Machinery (ACM). Licence: CC BY 4.0.