One anonymized domain, five channels, five different visibility scores: what the gap between Google rank and AI citation share actually measures, and what closes it.
Pull up a brand's dashboard and the top-line number looks settled: health score 87, organic rank holding at position three, crawl clean. Ask the four answer engines the same question about that domain and the number falls apart. Gemini cites the page on nine runs out of ten. Ask ChatGPT the identical query and the citation never shows up. Perplexity cites a different page entirely, one that has never cracked Google's own top fifty. Claude, the most selective of the four, doesn't cite the domain at all. Same brand, same day, four incompatible answers, and a health score that never moved.
No single crawler is misbehaving here. This is what search looks like now that there isn't one search anymore. A domain used to carry a single visibility number: where it sat among ten blue links. It now carries five, one per channel, and Caposeo tracks all five daily across a 12.4K-keyword set spanning Google, ChatGPT, Perplexity, Claude, and Gemini. The composite health score of 87 is what those five channels roll up into. What the composite can't do is tell you where the 87 came from, or which of the five channels is quietly dragging it down while the other four hold it up.
Google rank is still real and still worth tracking. A page holding position two or three is doing something right: it's crawlable and it loads fast, and it's structured cleanly enough that Core Web Vitals and the rest of the crawl-coverage math add up to a healthy score. A domain can hold that position for months while its backlink profile and its technical health keep improving underneath it. None of those same signals move a citation model that never checks backlinks and treats page speed as, at most, a tiebreaker. Google's ranking is still, at bottom, a contest between pages: whose content and links, backed by a sound technical foundation, earn the strongest position on a results page. An answer engine is judging something else entirely: whether one passage, lifted out of context, settles the question well enough to hand to someone who will never click through to the source. That's a different test, scored by different machinery, and a page can pass one while failing the other completely.
Even Google doesn't fully agree with itself on this anymore. Ahrefs' March 2026 update to its AI Overview citation study, covering 863,000 keywords and four million cited URLs, found only 38% of AI Overview citations came from a page that also ranked in Google's own top ten for the same query, down from 76% in its original July 2025 run. The AI layer sitting directly on top of Google Search has drifted from Google's own organic ranking in eight months. If the search engine can't keep its own two surfaces in agreement, four outside engines lining up with any of them was never realistic.
Plot one domain across all five channels and the shape tells you more than any single number could. Take a mid-market software brand in Caposeo's tracking set, watched across the branded and high-intent queries it should win outright. Citation share here means the percentage of that domain's own tracked queries where a channel surfaces or cites it at all, not an average pulled from someone else's keyword set. It's a narrower cut than the 12.4K-keyword corpus Caposeo tracks overall, but it's the cut that tells you where this specific domain actually stands to lose traffic.
Google, unsurprisingly, is the strongest bar. This domain earned its position-one-through-three organic slots and holds nearly all of them. Gemini comes closest to matching that number, because it grounds a large share of its answers directly in Search results and inherits some of Google's own judgment in the process. Perplexity sits meaningfully lower: it's citing sources rather than agreeing with rank, and it rewards the freshest passage that answers a question cleanly, wherever that passage happens to sit in Google's results. ChatGPT is lower again, citing far more selectively and often only when browsing is switched on for the query. Claude barely registers, the most conservative citer of the four by a wide margin. None of the four gaps below the top bar are equally easy to close, which is exactly what a single blended average hides.
A domain doesn't have one visibility problem. It has as many as there are engines answering for it.
An 87 is a useful number if the goal is a single glance at whether a domain's technical and structural foundation is sound. It's a poor number if the goal is deciding what to fix next, because 87 is an average across five very different stories. A domain can lose ten points of citation share on Claude and gain them back on Gemini, and the 87 won't move at all. The composite is doing its job. It's just answering a different question than the one an operator actually has, which is closer to: why did we lose the ChatGPT citation on our best-performing page. Run the same domain through a per-engine breakdown and that story usually surfaces fast: a schema change that helped Gemini's grounding did nothing for Claude's more conservative citation bar, and the 87 sat still through both.
None of the five channels move for the same reason, so there isn't one lever. On a domain that's ranking fine on Google and showing up nowhere else, here's the order we'd check, roughly cheapest to fix first:
Fix the passage-level answer first. It's the one lever that tends to move Gemini, Perplexity, and ChatGPT together, even when Claude takes longer to notice. Everything else on the list is worth doing, in that order, not all at once, and not before a per-engine breakdown says which channel is actually behind.
Caposeo tracks your Google rank alongside your citation share on ChatGPT, Perplexity, Claude, and Gemini: five numbers, one workspace, updated daily.
