Questions about this study

It measured what three AI assistants, ChatGPT, Gemini, and Claude, say when asked buyer-style questions, the kind a person types when they are trying to pick a vendor or service. Over August 4 to August 31, 2026, the study ran 14,951 distinct prompts against a cohort of 100,966 companies and collected 1,013,625 usable answers. For each answer, the study recorded whether a given company was mentioned at all, and if so, where in the list of recommendations it landed. It is a snapshot of recommendation behavior during that window, not a measure of brand reputation or site traffic.

When an AI assistant answers a buyer-style prompt, it typically names several companies in some order, first, second, third, and so on. Mention position is where a given company fell in that list on a given answer. Averaged across 363,777 mentions, the overall figure is 4.95. Gemini tended to mention companies earlier on average (4.72) than Claude (5) or ChatGPT (5.08). Lower numbers mean closer to the top of the list, which is generally the more valuable spot since readers skim top to bottom.

Yes, by this study's construction. Zero companies in the cohort of 100,966 failed to appear on all three engines (0%). But this should be read carefully: the cohort itself was built from companies that already had some plausible claim to visibility in these categories, so 'nobody was invisible' partly reflects how the study selected its companies, not a universal law that every business gets mentioned by AI search. A company outside this cohort could easily be invisible.

Across 337,500 company-prompt pairings where at least one engine named a company, all three engines agreed on only 2.8% of them, while 88.7% were named by just one engine alone. This is consistent with each assistant drawing on different training data, different retrieval or browsing mechanisms, and different internal ranking logic. The study cannot see inside those systems, so it cannot say which engine is 'more correct', only that they are not interchangeable. A company optimizing for one engine should not assume the same effort moves the others.

Not much, within what this study measured. Appearance rates across the four prompt types tracked, informational, navigational, commercial-investigation, and transactional, all land in a similar high-thirties percent band (38.8%, 37.4%, 38.6%, and 37.9% respectively). So whether someone asks a research-style question or a ready-to-buy question did not, by itself, predict visibility in this run. It is worth testing your own category, since broad averages can mask narrower effects.

Yes, and it is worth being explicit about it. The study ran one model version per engine (1 for ChatGPT, 1 for Gemini, 1 for Claude), over a single window from August 4 to August 31, 2026. AI assistants are updated frequently, and a different version, or the same version a month later, could surface different companies or different orderings entirely. The findings describe this window and these versions, not a permanent state of affairs. Treat it as a dated measurement, like a poll, not a constant.

The cohort covers 100,966 companies across dozens of categories, but the categories are far from evenly sized. Some, like Mobile App Development (6,048 companies) or Education IT Services (4,049 companies), have thousands of companies. Others are tiny: Tourism Digital Marketing has only 3 companies, and Legal Digital Marketing has 10. Every category shows visible_pct at or near full visibility, but a finding from a three-company category and one from a four-thousand-company category do not carry the same statistical weight, and the report does not pretend they do.

Start by checking where you stand on each engine separately rather than assuming one score. If you can only afford to test one engine first, note that ChatGPT showed the largest share of companies in this run (43.9%), so it is a reasonable first check, but Gemini (32.6%) and Claude (40.9%) behaved differently enough that a presence on one does not imply presence on another. Beyond appearing at all, look at where you land in the list, since position averaged 4.95 across engines and earlier is better. Then watch your own numbers over time rather than treating this one window as fixed.

The five most visible companies across engines this run were Visit Fort Wayne, Ecobee, TurboTax, GasBuddy, and Shutterstock. The study did not analyze why these five rank highest, it only measured that they do, so treat them as a benchmark set for comparison rather than a template to imitate. These are different kinds of organizations (a tourism board, a thermostat maker, tax software, a gas-price app, a stock photo site) which itself suggests no single tactic explains their visibility. A more useful exercise is tracking your own trajectory against these names over successive months.

Appearing just means a company's name showed up anywhere in an AI assistant's answer to a buyer-style prompt. Ranking well means it showed up near the top of that answer, since assistants usually list multiple options in some order. A company can appear frequently but consistently near the bottom, which is a weaker outcome than appearing less often but near the top. This study tracked both: the visibility percentages per engine measure appearance, while the average mention position (overall 4.95) measures where in the list those appearances tend to land.

All 1,013,625 of 1,013,625 rows returned a valid, parseable result in this run, so there is no visible backlog of failed or malformed responses to account for. That said, 'status ok' only confirms the response was readable and scorable, it does not confirm the assistant's underlying answer was accurate or current. A clean parse rate tells you the measurement pipeline worked, not that every recommendation was a good one.