The question
AI Overviews, ChatGPT search, Perplexity, and Claude have been accused of destroying publisher traffic. The data is mixed, the panic is loud, and most reporting is anecdotal.
We tracked 60 client domains across SaaS, DTC, media, and local services for 18 months to see what AI search actually did to traffic — and which strategies are working for domains gaining share despite the shift.
Aggregate findings
-18%Median click-through rate change on informational queries
+2.3%Median click-through rate change on transactional queries
+37%Share of citations in AI Overviews for top performers
31%Of domains grew traffic through the window
Who's winning and who's losing
Losing: Low-differentiation informational content
Domains that primarily published general informational content ("What is X?", "How to Y") saw significant traffic declines. Median: -34% on affected pages over 12 months. AI Overviews are good at summarizing well-covered topics without clicks required.
Winning: Original research + utility content
Three content types showed stable or growing traffic:
- Original data/research content. AI assistants cite sources; domains with proprietary data got cited and attracted direct traffic from the citation.
- Interactive tools and calculators. Can't be summarized by AI. Traffic held steady or grew.
- Deep product/service content. Specific product pages, comparison content, and buyer-stage content stayed resilient — AI assistants send users to transactional pages rather than summarizing them.
Growing: Branded search capture
Branded search grew across the dataset. The theory: AI assistants expose users to specific brand names in their summaries, which drives branded searches to the site. Clients investing in brand marketing saw this effect disproportionately.
What's working for AI citation
Of the 60 domains, 11 significantly grew their share of citations in AI Overviews. Common characteristics:
- Clear, extractable facts. Content with specific numbers, dates, and definitions got cited more than content with abstract claims.
- Original data. Research reports and surveys got cited at ~3x the rate of synthesized content.
- Clean structure. Headers, lists, tables. AI models extract structured content more reliably.
- Entity consistency. Domains with consistent entity signals (same facts repeated across site + external sources) got cited more.
Three strategic shifts worth considering
- Reduce generic informational content investment. If you were publishing "What is X" guides for SEO, that's a declining asset. Reallocate to original research and bottom-funnel content.
- Invest in branded demand. AI search rewards brands users specifically ask about. YouTube, PR, podcast sponsorships, exec thought leadership — these feed the branded search channel that AI search amplifies.
- Add utility, not just text. Calculators, directories, comparisons, interactive tools. Content that resists summarization.
Methodology note: We used Google Search Console click/impression data plus third-party AI Overview visibility tracking (Semrush AIO coverage). Causality is partially inferred — we can observe that CTR dropped on queries where AI Overviews now appear, but can't fully isolate AI Overviews as the cause from other SERP changes.