Track the same 12.4K keywords on Google and four AI engines at once, and rank stops predicting who gets cited. Here's the divergence data.
Pull the same keyword up on five screens and you'd expect some agreement between them. A page that holds its ground on Google should show up somewhere across the others. We watched that assumption fail for a specialty retail client on a single high-value term: position two on Google, a page shoppers found reliably. Ask ChatGPT, Perplexity, or Claude the same question and the page never came up. Only Gemini cited it, and only inconsistently. The gap wasn't a fluke of one query, either. It showed up again the following week, and the week after that.
That's one keyword. Caposeo runs 12.4K of them through the same five surfaces for the workspace's customer base, tracking Google alongside ChatGPT, Perplexity, Claude, and Gemini every day. Once you're watching all five surfaces at once, the pattern above sits closer to the median than the exception.
Google rank and AI citation share look like they should move together. Both are, in theory, measuring the same thing: does this page answer the query well. In practice, they're scored by different machinery. Google's ranking still leans on crawl depth, the backlink graph, and relevance across the full page. An answer engine skips the ranking step entirely. It retrieves a passage, decides whether that passage settles the question on its own, and cites it (or not) inside a generated answer. The unit of competition shrinks from the page to the sentence that survives extraction.
That shift matters more than most SEO tooling admits, because attention is moving into the summary. Pew Research Center's July 2025 analysis of real Google browsing data found people clicked a traditional result in just 8% of searches that returned an AI summary, compared with 15% on searches without one, and only 1% of visits went to a link cited inside the summary itself. Fewer clicks doesn't mean less consequence. It means the handful of citation slots inside that summary now carry more of the referral value than the ten blue links below it used to.
Plot Google position on one axis and citation share across the four engines on the other, and four clusters show up.
The quadrant that gets an operator's attention first is the one their existing SEO dashboard swears doesn't exist: pages holding position two or three on Google, sending steady organic traffic, sitting at zero citation share across all four engines. A domain's health score, the composite Caposeo builds from crawl coverage, Core Web Vitals, and structured data, can sit at 87 and read clean while the citation column stays flat. Rank tells you the page is fine. It doesn't tell you whether any of the four engines that increasingly answer for your customers have ever picked it up.
Ghost visibility runs the other direction, and it's just as easy to miss. A page barely holding position twenty-five on Google can still turn up as the cited source on Perplexity or ChatGPT, usually because it happens to contain the one paragraph that states an answer cleanly, with no scrolling required. Nobody on the SEO side is watching that page. Nothing in a traditional rank tracker would flag it as worth watching. It can be earning citations from three answer engines a day, long before it ever earns a first-page ranking on Google.
Ask two answer engines the same question and you're often comparing two different snapshots of the internet, not two opinions about the same one.
Caposeo runs the same 12.4K keywords through ChatGPT, Perplexity, Claude, and Gemini every day, and citation share on any single keyword can look completely different from one engine to the next. The gap comes from architecture: each engine draws from a different corpus on a different refresh cycle and rewards a different kind of page for the citation.
Only Gemini's citation behavior tracks Google rank with any consistency, which makes sense given it grounds directly in Search. The other three treat a page as raw material for an answer instead of a ranked result on a page of results. That's exactly where a single rank tracker runs out of story: it can show a domain looking fine across three of the four surfaces that matter most now, when only one of those three readings is telling the truth.
Three things worth pulling up before treating rank as a proxy for AI visibility:
Google rank still matters. It's one read among five now, where it used to be treated as the only one that counted.
Caposeo tracks your Google rank alongside your citation share on ChatGPT, Perplexity, Claude, and Gemini, all in one workspace, updated daily.
