Do reader judgments reflect actual authorship or just their beliefs?

Akpinar et al. report that "participants' beliefs about authorship shaped their evaluations even when inaccurate." The study separates two conditions. In the guess condition, readers estimate the degree of LLM involvement with no authorship information. In the information condition, the authorship of each abstract is disclosed. Disclosure is the second finding: "Across all abstract types, disclosure consistently elevated ratings of trust and quality." Readers "expressed neutrality when guessing but reported positive trust once authorship was revealed; even for fully LLM-generated texts."

The excerpt's account of the belief effect runs through prior assumptions. Readers start with "a baseline suspicion that LLMs were involved across all abstracts," and the authors set their results beside a human preference bias that Porter and Machery report, in which readers "judge LLM-generated works as human more often than the reverse." The ratings are Likert scales on information quality and trust, plus free-text reasons. The excerpt does not explain why disclosure raised ratings: the authors say the phenomenon is "discussed further in Section 5.3," which is not in the excerpt. Their stated reading is that disclosure "may have a more substantial impact on how readers judge trustworthiness and information quality" than abstract type does.

The disclosure result is a stated contrast. The authors write that it "contrasts the findings of prior work demonstrating decreases in credibility and quality judgments after AI use has been disclosed." That places it against the question in Does banning LLM use in peer review change review outcomes?: there, rules about LLM use barely moved reviewer scores, while here, information about authorship moved reader ratings. The two studies differ in population, outcome and manipulation, so they measure different things; the useful contrast is which lever moved judgment. The reader-side view is also consistent with Do writers want to see each other's AI prompts in shared editors?: if disclosure moves readers' ratings, writers' wish for visibility into AI use has a reason beyond their own workflow. The excerpt does not test writers.

The excerpt does not establish why disclosure raises ratings, since the explanation sits in Section 5.3, which is absent here. It gives no sample size, and its measures are self-reported ratings and stated preferences, not behavior. It therefore shows how readers rate abstracts under one disclosure label, not how they would act on that label in a journal or a review. At this strength, the implication is that disclosure is not a neutral act for readers: in this design, the label changed ratings. Disclosure rules should be tested with readers rather than taken from the prior-work penalty, which this study does not reproduce.