How Often Do ChatGPT, Gemini, and Claude Agree on Which Companies to Mention?

If you want to know whether your company shows up in AI search results, the first question to settle is: compared to what baseline? This study, which ran 8,188 distinct prompts against a cohort of 58,803 companies across three engines (ChatGPT, Gemini, and Claude) over a run window from September 4 to September 25, 2026, collected 536,301 usable answers. Each one was checked for whether a given company was named, and if so, where.

One of the clearest results has nothing to do with any single company's performance. It's about how much the three engines agree with each other in the first place.

Across the study, 178,767 company-prompt pairs were named by at least one engine. Of those, the breakdown by how many engines named the same pair looks like this: 157,472 pairs, or 88.1%, were named by only one engine. 15,865 pairs, 8.9%, were named by two of the three. And just 5,430 pairs, 3%, were named by all three engines. Put another way, 88.1% of the time a company got named at all, it was named by exactly one engine and ignored by the other two. Full three-way agreement, which is the closest thing to a consensus pick, happened in only 3% of cases.

Why would this be true? The mechanical answer is that each engine draws on a different retrieval and ranking process behind the scenes, trained on different data, updated on different schedules, and in this study each engine was tested on a single model version (a count of 1 version each for ChatGPT, Gemini, and Claude, so this isn't an artifact of mixing old and new model snapshots within an engine). The three engines aren't consulting the same index and ranking it slightly differently. They are, in effect, running three separate research processes that happen to be given the same question.

This matters for anyone trying to interpret a single favorable or unfavorable mention. If your company appears in a ChatGPT answer but not a Gemini one for the same prompt, the low agreement rate found here says that's the normal case, not a sign that something is wrong with your Gemini presence specifically. The modal outcome in this dataset is disagreement. A company appearing on exactly one engine is not an edge case, it is most of what happens.

It also means that any claim of the form "we rank well in AI search" needs a named engine attached to it. A result that holds on one engine cannot be assumed to generalize to the others. This study's own visibility figures illustrate the spread directly: 44.6% of the cohort appeared somewhere on ChatGPT, versus 40.5% on Claude and 32.4% on Gemini. Those are not close enough to treat as noise around a shared number.

What would make this finding wrong, or at least less important? If the disagreement were concentrated in a long tail of obscure prompts or companies, where any of several plausible answers is defensible and the engines are simply sampling from a wide pool of acceptable names, that would be a less alarming story than if the engines disagreed even on the companies with the strongest, most unambiguous claim to relevance. This study's design, checking named mentions across 8,188 distinct prompts without separately scoring prompt ambiguity, can't distinguish those two cases on its own. A follow-up that stratifies agreement by how concentrated the plausible answer set is for a given prompt would tell you whether low agreement is a sign of genuine engine divergence or just a sign that many buying questions have many reasonable answers.

What a company does with this on a Monday morning: stop asking "are we visible in AI search" as a single yes-or-no question, and start tracking visibility per engine. A monitoring setup that only checks ChatGPT, because it's the most familiar, will systematically miss what's happening on the other two engines, and given how low the three-way agreement rate is, that gap is not something you can estimate from the ChatGPT number alone.