The Four AI Engines Do Not Agree on Who to Mention

This study ran the identical set of category prompts against four engines, ChatGPT, Claude, Gemini, and Google AI Overview, for each of the 50 companies in the cohort. If the engines drew on similar underlying information and reasoned about categories the same way, we would expect a company visible on one engine to be visible, or close to it, on the others.

That is not what the data shows. 10 companies, 20% of the cohort, were mentioned on at least one engine but not on all of them. Combined with the 5 companies invisible everywhere, that means a full 10 plus 5 of the 50 companies, fifteen in total, did not achieve uniform visibility across engines. Only the remainder achieved something close to consistent presence.

The raw appearance-rate gap between engines

Part of the disagreement is explained by engines simply mentioning companies at different overall rates. Google AI Overview mentioned cohort companies in 82% of its answers, the highest of the four. ChatGPT and Claude were tied at 80% and 80% respectively. Gemini was lowest, at 74%. An engine that mentions companies less often overall will naturally produce more cases of "visible elsewhere, not here," simply as a function of its stricter threshold, not necessarily because it has a different opinion about any specific company.

But appearance rate alone does not explain all of it, because the engines also read different sources. Claude's top-cited sources are dominated by guideflow.com, gartner.com, and capterra.com. Google AI Overview leans on youtube.com and reddit.com ahead of gartner.com. ChatGPT leans most heavily on learn.microsoft.com and g2.com. An engine trained more heavily on, or more willing to retrieve from, enterprise documentation sites is answering a subtly different underlying question than one drawing on community discussion and review platforms, even when given the same literal prompt text.

Why this matters for anyone buying an AI-visibility score

A vendor selling a single blended visibility score across engines is implicitly averaging over this disagreement. A company that scores, hypothetically, sixty out of a hundred on such a blended score could be the company visible on three engines and absent on the fourth, or the company visible everywhere at a moderate position, or the company visible on one engine at a strong position and absent on three others. These are three different situations requiring three different responses, and a single number collapses all of them into the same reported score.

The honest response to this finding, for a company trying to act on AI-visibility data, is to ask for and read the per-engine breakdown before acting on any blended metric. If a company is absent specifically on Gemini but present elsewhere, and Gemini's own citation pattern showed only google.com as a recorded top source in this study, the likely lever is different than if the company is absent specifically on Claude, where guideflow.com and gartner.com dominate the citation pattern. Treating the four engines as one undifferentiated "AI" channel discards the information that would actually tell a company what to do differently.