Eye-Tracking Analysis of Visual Attention and Preference in Human-Created, AI-Generated, and AI-Refined Human...

Generative artificial intelligence (GenAI) is increasingly used in advertising, yet its effects on visual attention and preference remain insufficiently characterized. This study compared human-created (H), AI-generated (AI), and AI-refined human (HAI) advertisements, the latter obtained by submitting human-created advertisements to a generative model with a standardized refinement prompt, using eye tracking and a forced-choice task. Thirty-nine young adults (18-26 years) with high generative-AI experience completed 858 trials involving 22 advertisement pairs while gaze was recorded at 60 Hz. Fixation duration, fixation count, time to first fixation, saccadic magnitude, and pupil diameter were compared using repeated-measures analyses of participant-level means with Holm correction, and choices were analyzed using binomial models with participant-clustered standard errors. AI-generated advertisements were selected more frequently than human-created advertisements (58.97% of trials; odds ratio = 1.44, p=0.0015), whereas HAI advertisements did not differ from human-created advertisements in choice frequency (49.88%; p=0.978). Relative to human-created advertisements, both AI and HAI stimuli were associated with higher fixation counts and smaller saccadic magnitudes, with these effects surviving Holm correction. Time to first fixation was statistically equivalent across conditions within the smallest effect detectable by the design, pupil diameter did not differ, and differences in fixation duration did not survive correction. The higher fixation counts reflected a larger share of fixation time on AI and HAI advertisements rather than a higher fixation rate. Although the gaze pattern was similar for AI and HAI stimuli while only AI stimuli were preferred, the difference between the two choice effects was not statistically reliable. The observed gaze differences cannot be uniquely attributed to creative origin because AI and HAI stimuli also exhibited higher edge density than human-created stimuli. These findings support the use of eye tracking as a sensor-based method for characterizing visual responses to AI-assisted advertising while highlighting the need for tighter stimulus control and more diverse samples in future studies.