Does telling readers 'AI helped write this' make them trust a report less?

This explores whether labeling a report as AI-involved makes readers trust it less. The short answer is yes, but how much depends heavily on what kind of text it is. For a news article, the penalty is real but tiny. When nearly 2,000 human readers and 2,500 LLM raters scored the same article with and without an AI disclosure line, the disclosed version scored lower by less than 0.15 points on a 7-point scale Does disclosing AI assistance make readers trust articles less?. The penalty gets much bigger when the writing is supposed to come from a person. In a study of 261 readers, disclosing AI authorship lowered perceived trustworthiness, caring, and likability across the board, and the biggest drops came in interpersonal writing, where readers felt that using AI broke a social expectation of real empathy How does revealing AI authorship change reader trust?. So the lesson isn't that AI disclosure hurts trust everywhere. It hurts most where readers expect a human to have cared.

You might not expect this: the penalty doesn't have to last. When people work with an AI partner over and over and can see the results, their first reluctance about a disclosed AI fades and can even reverse into a preference Does revealing AI identity help or hurt user trust?. Seeing outcomes is what makes the difference. A disclosure on its own, with no track record to check it against, leaves the bias in place. For a one-off report, readers get exactly that: a label and no feedback. That suggests the trust penalty on reports is a first-impression effect that track records could wear down.

The author's side of the story matters just as much. Across four experiments with more than 4,400 people, AI users expected others to see them as less competent and less diligent, and as a result they were less willing to tell managers and colleagues they had used AI Do people fear judgment when they use AI at work?. Put that next to the small measured penalty for news articles and you get a gap: people may fear the disclosure penalty more than readers actually impose it, at least for informational writing. That fear pushes AI use out of sight, which works against the transparency disclosure is meant to provide.

There's an awkward flip side, too. The cues that actually drive trust in AI often have nothing to do with whether the content is right. Users in every language studied follow confident-sounding AI outputs even when they're wrong Do users worldwide trust confident AI outputs even when wrong?, and ChatGPT earns trust through conversational back-and-forth rather than accuracy Does conversational style actually make AI more trustworthy?. Meanwhile, training AI to sound warmer can make it measurably less reliable Does empathy training make AI systems less reliable?. Taken together, a small drop in trust after disclosure may be closer to well-calibrated skepticism than to unfair bias, and a warm, fluent, unlabeled report may earn more trust than it deserves.

The corpus has a gap here. None of these studies separate kinds of reports, such as analytical, financial, or scientific writing, or test whether *how* the disclosure is worded changes the penalty (for example, "AI drafted, human verified" versus "AI-generated"). Work on users' sense of ownership hints that how much control a person had over the text matters Does user control over AI text shape feelings of ownership?, but nobody has yet tested that from the reader's side.