Year: 2026
Type: article
Abstract: This study investigates how explicit authorship labels — particularly those involving artificial intelligence (AI) — affect readers’ comprehension, trust, and intention to use scientific information. In a between-subjects online experiment (N = 183), university students read the same AI-generated science text about superconductors, with one of three authorship cues: (1) no label (control), (2) “AI-generated,” or (3) “AI-generated, human-edited.” All participants completed a five-item multiple-choice comprehension quiz (MCQ), a semantic differential scale assessing trust, and a self-report measure of behavioral intention to use or share the text. Results showed no statistically significant differences in comprehension scores across conditions, confirming that the inclusion of authorship labels did not impact factual understanding. However, the “AI-generated” label sharply reduced both trust and intended use compared to the control group (Hedges’ g > 1.1), indicating an aversion effect. Importantly, appending the phrase “human-edited” to the AI label eliminated this trust penalty: responses in the “AI+Human” condition were statistically indistinguishable from the control group. These findings align with prior cognitive and communication research, which suggests that while source cues rarely alter comprehension, they substantially influence user trust and perceived credibility. The study highlights the importance of carefully framing AI involvement in educational or scientific content. It suggests that emphasizing human oversight – through transparent labeling – can mitigate skepticism and foster appropriate trust, even when the content is AI-generated. Implications for ethical AI disclosure policies in higher education are discussed.
Authors: Mikhail Matveev, Elena Matveeva
Language: English
Subfield: Developmental and Educational Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: diamond
OpenAlex ID: W7166536037