Questions about this study

It measured whether three AI assistants, ChatGPT, Gemini, and Claude, mention a given company when asked buying-type questions. Across a cohort of 58,803 companies, the study ran 8,188 distinct prompts over a run window from 2026-09-04 to 2026-09-25 and collected 536,301 usable answers. Each answer was checked for whether the company was named at all, and if so, where in the answer it appeared. The result is a picture of presence (does the engine name you) and position (where in the list), broken out by engine and by the type of prompt asked.

Almost certainly yes, in the sense of being named at least once. 56,194 companies, 95.6% of the cohort, were named by at least one of the three engines somewhere across the run. Only 0 companies never appeared on any engine. That makes the binary question, does any AI assistant know we exist, answered yes for nearly the entire cohort. It does not mean you are named consistently, or by the engine your buyers actually use, which is a separate and more revealing question.

The gap is between engines, not between visible and invisible companies. ChatGPT named 44.6% of the cohort, Claude named 40.5%, and Gemini named 32.4%. A company can be solidly covered on one assistant and largely absent on another. Since a buyer only ever opens one assistant at a time, a company's real-world visibility depends heavily on which one that buyer happens to pick, even though the company would score as visible in an aggregate count.

Rarely. Looking at pairs of engines that named the same company at least once, 88.1% of those pairs involved only one engine doing the naming, and all three engines converged on the same company in just 3% of cases. That low overlap suggests each engine is drawing on its own retrieval and ranking process rather than consulting a shared underlying source of truth about who the relevant companies are. It is a reminder that 'AI search' is not one system, it is three systems that happen to get asked the same question.

Position varies only slightly by engine once a company is mentioned at all. Average position was 5.08 on ChatGPT, 4.97 on Claude, and 4.72 on Gemini, against an overall average of 4.94 across 197,893 ranked mentions. The practical reading is that the competitive battle here is mostly about getting mentioned at all, not about climbing from third place to first once you are already in the answer. A company chasing AI visibility should spend more effort on inclusion than on reordering.

Only modestly. Appearance rate ranged from 37.6% on navigational prompts up to 39.1% on commercial-investigation prompts, with informational and transactional prompts falling in between. That is a narrow band. It means prompt type alone is not a strong lever for getting a company named. Whatever drives inclusion in an engine's answer seems to sit upstream of the specific phrasing of the question, in how the engine builds its candidate set in the first place.

It is a real limit and worth stating plainly. The study covers 3 engines, each represented by a single model version during the run window, so this is a snapshot of one moment and one build of each assistant, not a trend line across model versions or a guarantee of what a different model generation would say. It is also only three engines; other AI search surfaces exist and were not tested. The 536,301 answer count and 8,188 distinct prompts are large enough to give stable percentages within this run, but a reader should treat the findings as describing September 2026's specific model builds, not AI search in perpetuity.

No, and this is worth being direct about. Every category cut in this run shows visible_pct at the maximum, but that is a property of how the categories were built, not a finding about equal visibility. Companies had to already be matched to a category to be counted in that cut, so the category-level numbers describe only companies that cleared that matching step. They should not be read as evidence that a company in a small, obscure niche is as easy to find as one in a large, well-covered category. The category cuts answer 'among matched companies, how visible were they,' not 'how easy is it to get matched in the first place.'

It is the typical rank at which a company appears within an engine's answer, when the answer lists more than one company, averaged across every instance the company was named. A lower number means the company tends to appear earlier in the list, which likely gets more attention from a reader skimming the answer. Across this study it averaged 4.94 out of 197,893 ranked mentions, and did not vary much between engines, which is why the study treats inclusion, not position, as the more consequential variable to influence.

In this run, no company in the cohort fell into that category, 0 of 58,803, so a true zero-visibility result did not occur this month. If you are seeing zero mentions in your own monitoring, the first thing to check is whether you are comparing against the same prompt set and engine versions this study used, since visibility rates here differ meaningfully by engine. It is also possible your company sits in a category or query space this study's cohort did not cover, which this data cannot speak to either way.

Three things follow directly from the data. First, check visibility per engine rather than in aggregate, since the gap between ChatGPT's 44.6% and Gemini's 32.4% means a single blended number can hide a real weak spot. Second, treat inclusion as the priority over position, since once named, average rank barely moves between engines. Third, do not expect to fix low visibility by rewording prompts into a different type, since appearance rate only moved from 37.6% to 39.1% across prompt types, meaning the lever is probably upstream, in what each engine's retrieval step treats as a credible source about you, not in how the question gets phrased.