The short answer
AI search in 2026 is no longer a measurement vacuum. Multiple large-scale citation studies published over the past year — Muck Rack's 25-million-link analysis, Semrush's 150,000-citation comparison, and Profound's conversation-level research — now describe how generative engines actually select sources. The pattern they converge on: earned editorial coverage dominates the citation pool, traditional search rankings still predict citations on some engines far more than others, each engine leans on a visibly different mix of domains, and the first question a user asks is where most citations happen. Meanwhile, the tooling market is consolidating into large marketing suites, and Google has begun shipping first-party AI-visibility reporting. Here is what the data shows and what it means for the next quarter.
Earned media dominates the citation pool
Muck Rack's May 2026 analysis of more than 25 million links surfaced in ChatGPT, Claude, and Gemini across 17 industries found that earned media accounted for 84 percent of links, while paid or advertorial content contributed just 0.3 percent. Journalism alone supplied 27 percent of links overall. The study also found intent effects: trend-style questions drew journalism at more than twice the rate of how-to questions, and press releases appeared 3.5 times more often for trend prompts than for best-of prompts. The practical reading is blunt. If your AI-visibility plan is built on sponsored placements, the citation data says you are buying inventory the engines barely read. Coverage that a journalist chose to write is, by a wide margin, the raw material generative engines cite — which pushes PR and original, newsworthy research back to the center of the strategy.
Each engine reads a different web
Muck Rack's cross-engine data shows the top cited domain differs by engine: Wikipedia for ChatGPT, PubMed Central for Claude, and Reddit for Gemini. Profound's source-category analysis of 27 million citations adds another axis: social sources made up 19 percent of Perplexity citations but only 5 percent on ChatGPT and 7 percent on Google AI Overviews. The same study found social citations rose from 5.4 percent on category-level prompts to 15 percent on brand prompts — when users ask about you by name, community discussion carries more weight. The implication is that there is no single 'AI search' channel to optimize. A distribution plan tuned for Gemini's community-heavy diet will underweight the reference-style sources ChatGPT favors, and vice versa. Map which engines your buyers actually use, then weight your source mix accordingly.
Rankings still matter — unevenly
Semrush's study of 5,000 queries and 150,000 citations found that Perplexity cited a domain from Google's top ten results in 91 percent of cases, and the exact URL in 82 percent. For Google AI Overviews the figures were 86 and 67 percent. That is a strong argument that classic ranking work still feeds AI citations on those engines. But the picture is not uniform: a 2026 preprint that ran 55,393 queries across 19 categories over 40 days reported that roughly 30 percent of cited domains did not appear on the first page of search results. That study is still under review, so treat its estimates as provisional — but the direction matters. A meaningful share of citations comes from outside the top rankings, so page-one positions are neither sufficient nor strictly necessary. Semrush also found commercial answers ran roughly twice as long as informational ones, which rewards pages carrying structured comparisons, current prices, and trade-offs rather than thin summaries.
Citations concentrate on the first question
Profound analyzed roughly 730,000 cited US-English conversations from October through December 2025 and found citation incidence fell from 12.6 percent at the first turn to 4.5 percent by turn ten and 3 percent by turn twenty; a cited conversation averaged about six citations. In plain terms: if an engine is going to cite anyone, it usually does so in response to the opening question. Pages built to fully resolve the first thing a buyer asks — what is X, how much does X cost, X versus Y — are competing for most of the available citation surface. Profound's source-category data also found owned sources supplied only 4.3 percent of citations on category-level prompts. Your own site matters, but the bulk of citation volume flows through media, institutional, review, and community sources you do not control. You earn your way into that pool; you cannot publish your way into it alone.
Distribution lifts citations — with two big caveats
Stacker's March 2026 study tracked 87 distributed stories for 30 brands across eight AI platforms and reported a 239 percent median lift in citations after distributed news coverage, with median cross-platform coverage rising from 5.4 to 17.9 percent and 64 percent of observed citations pointing to publisher coverage rather than the brands' own sites. Two caveats are required. First, the study is vendor-authored and explicitly observational — Stacker itself says the analysis cannot establish causation, so the 239 percent figure is a reported median, not an expected result. Second, citations are not automatically accurate: the same 2026 preprint mentioned above found that 11 percent of 98,020 atomic claims in AI answers were unsupported by the citation attached to them — again, a provisional estimate pending review. Counting citations without auditing whether they actually support what the engine said will overstate your real visibility.
Consolidation, and Google enters measurement
Two acquisitions in 2026 reframed the vendor landscape. Adobe completed its acquisition of Semrush on April 28, 2026, and Sitecore acquired Scrunch on June 3, 2026, folding its AI-visibility and agent-experience capabilities into SitecoreAI. AI-search measurement is being absorbed into large marketing platforms rather than remaining a standalone category — a signal worth weighing when you sign a multi-year contract with an independent tracker. At the same time, Google entered measurement directly: on June 3, 2026 it announced generative-AI performance reports in Search Console for a subset of sites, covering AI Overviews and AI Mode by page, country, device, and date. Google's own AI-optimization guidance is also worth taking at face value. It states that the same foundational SEO practices apply to AI features, that no special AI files or schema are required, warns against inauthentic mentions, and notes that third-party tools do not have access to Google's internal search data. Treat vendor tactics that contradict this guidance as hypotheses, not requirements.
What to do in the next quarter
The data supports a specific 90-day plan rather than a general resolution to 'do GEO.' Each move below maps to a finding above.
- Rebuild your highest-value pages to fully resolve the opening question — definition, price, and comparison up top — because Profound's data shows citation incidence peaks at turn one.
- Shift budget from sponsored placements toward earned coverage: pitch original, newsworthy data to journalists, since earned media supplied 84 percent of links in Muck Rack's analysis and paid content just 0.3 percent.
- Build a per-engine source map: reference-style coverage for ChatGPT, community presence for Gemini and Perplexity, and continued ranking work for Perplexity and AI Overviews, where Semrush found the highest overlap with Google's top ten.
- Enroll in Search Console's generative-AI performance reports if your site has access, and treat them as your free first-party baseline before paying for multi-engine tracking.
- Add a citation-fidelity check to your reporting: sample AI answers that cite you and verify each claim matches your page, given the preprint's provisional 11 percent unsupported-citation estimate.
- If you commission distributed research, set expectations as a measured range, not a promise — Stacker's 239 percent lift is a vendor-reported, observational median, not a guaranteed outcome.
