A travel booking brand we reviewed recently showed up in roughly four of every ten Gemini answers about trip planning. On Perplexity, asking the same questions, a single competitor was cited in almost every response. Same category, same query set, two completely different outcomes. That gap is not a tracking error. It is what happens when you measure share of citation on one engine and assume it tells you anything about the rest.
Key Takeaways
- Share of mention and share of citation are different metrics, and a brand can score high on one while scoring near zero on the other, on the same engine.
- Each AI engine builds its own hierarchy of trusted sources from a different retrieval mechanism, so dominance on one platform does not transfer to another.
- Citations compound within a single engine once a model starts citing a source, which means a competitor’s early lead on one platform gets harder to dislodge the longer you track only elsewhere.
The Scenario That Breaks Single-Engine Dashboards
Most teams start AEO tracking with whichever engine their customers mention most in interviews or support tickets. That is a reasonable starting point and a poor place to stop. A brand can be named constantly inside ChatGPT answers, pulled from open-web coverage, while Claude, which leans on high-authority publications and documentation, cites a competitor almost exclusively because that competitor has better structured documentation and fewer ambiguous entity signals.
Neither number is wrong. They are measuring different retrieval systems with different source preferences. The mistake is reporting one of them as “AI visibility” and building a strategy on top of it.
Share of Mention Is Not Share of Citation
Share of voice is the web-wide version of this problem: how often a brand gets talked about anywhere, regardless of whether an AI model ever surfaces that mention as a source. Share of citation is narrower and more useful. It measures how often a model pulls a brand into an answer and names it as the source behind the claim.
A brand can have heavy share of voice from press coverage, reviews and social chatter, and still have a share of citation close to zero on the engine that actually sends a buyer to a decision page. Mentions are ambient. Citations are the ones a model is willing to stake its answer on. Confusing the two is how a marketing team ends up confident about AI visibility while a competitor quietly owns the recommendation layer.
Why Each Engine Builds Its Own Hierarchy
The hierarchy difference is not random. It follows directly from how each engine retrieves. ChatGPT pulls from across the open web, which rewards breadth of coverage and recency. Gemini cross-references Search, YouTube and Scholar, so a brand with strong video presence or academic citation can outperform its web-only share of voice. Claude leans on high-authority publications and documentation, which favors brands with clean technical content and third-party validation over brands with high press volume. Perplexity cites its sources inline, which makes its citation behavior the most visible and auditable of the five, and also the most literal: if your page is not structured to be quoted cleanly, it does not get quoted. Grok reads the real-time social web, so a brand active and credible on X shows up there in ways that do not register anywhere else.
Five different retrieval logics produce five different winners for the same query, asked the same week.
PR topic visibility growth for NewDay USA against a +10% target, tracked weekly across five engines over six months, which is the kind of gap that only shows up when you measure every engine instead of one.See the case study
How a Citation Hierarchy Locks In
The gap between engines does not stay fixed. It widens on its own, because citations compound. Once a model cites a brand for a given query type, that brand’s content becomes part of the pattern the model has already treated as reliable, and the next citation for a related query becomes more likely. A competitor who got cited first on Perplexity for “best VA loan lenders” is not just ahead by one answer. They are ahead by a trend the model keeps reinforcing.
A brand can have high share of voice and near-zero share of citation, and only one of those numbers predicts what the model recommends next.
This is also why single-engine tracking misleads strategy more over time than it does on day one. A team watching only ChatGPT might see flat, healthy numbers for months while a rival’s Claude citation lead compounds unseen, until that rival shows up as the default answer across documentation-style queries that a buyer researching a serious purchase actually asks.
What Single-Engine Tracking Costs a Strategy
Three practical failures follow from tracking one engine.
First, budget gets allocated to the engine you can see, not the engine where the competitive gap is widest. If Gemini looks strong and Claude is never checked, the team keeps investing in the metric that already looks good instead of the one that is losing ground.
Second, content decisions get made for the wrong retrieval logic. Content built for ChatGPT’s open-web breadth, heavy on link-worthy commentary, does not serve Claude’s preference for documentation-grade clarity. Optimizing for one without knowing the other means half the content budget is aimed at a target that was never the problem.
Third, the training-data layer gets ignored. AEO has two layers: training-data presence, earned slowly through wide, consistent coverage, and retrieval presence, earned technically through structured, current content. Single-engine tracking tends to surface retrieval problems, because those show up fast in citation counts. It rarely surfaces the training-data gap, which takes longer to form and longer to fix, and which is usually the real reason one engine’s hierarchy favors a competitor by a wide margin.
What To Do Next
Pull citation data, not just mention data, from at least three engines before drawing any conclusion about where a brand stands. If one engine shows strong presence and another shows a competitor dominating citations almost entirely, treat that as the real finding, not noise to average out.
Map which engine favors which kind of source for your category. If Claude-style documentation citations are going to one competitor, audit their technical content and compare it structurally to yours, not just topically. If Grok-style citations favor a rival, check whether the gap is a content problem or a presence problem on the real-time social web, because the fix is different for each.
Re-run the comparison quarterly, not annually. Citation hierarchies compound, and a three-point gap on one engine today can be a double-digit gap in two quarters if nobody is watching it.
FAQs
What Is the Difference Between Share of Mention and Share of Citation?
Share of mention measures how often a brand is talked about anywhere online, including in content an AI model never surfaces. Share of citation measures how often a model actually names a brand as the source behind an answer. A brand can score high on the first and near zero on the second, even on the same engine.
Why Would a Brand Dominate One AI Engine and Barely Appear on Another?
Each engine retrieves from different sources: ChatGPT pulls from the open web, Gemini cross-references Search, YouTube and Scholar, Claude favors high-authority publications and documentation, Perplexity cites sources inline, and Grok reads the real-time social web. A brand strong in one source type will not automatically be strong in another.
Why Does a Competitor’s Citation Lead Get Harder to Close Over Time?
Citations compound. Once a model cites a brand for a query type, that pattern reinforces future citations for related queries, which widens the gap the longer it goes unaddressed.
How Many AI Engines Should a Brand Track Before Making Strategy Decisions?
At minimum, track citation behavior across three engines with meaningfully different retrieval logic, for example one open-web engine, one documentation-weighted engine and one source-transparent engine like Perplexity, before concluding anything about overall AI visibility.
