What gets a brand cited by AI: what we have measured.
Three pieces of research, each labelled by engine, collection route and date, and none of them gated. Where a figure is quoted, the page it comes from states the denominator and the method. Where we only have a pattern from product data, we say so and attach no number.
Correction, 5 September 2026: this page previously claimed an analysis of 2 million AI answers and offered a downloadable report behind an email form. No such dataset or report existed. Both have been removed.
The UK AI Shortlist Study, September 2026
Does AI name the suppliers whose own websites rank first on Google for the same buying question? 1,000 questions across 25 commercial categories, put to Google and six consumer AI services three times each: 17,537 completed AI answers and 3,000 Google captures. Method fixed before collection; raw answers kept; the data is published with the findings.
Collected 6 to 7 September 2026. ChatGPT did not name the business at Google's first organic position in 72.5% of eligible answers. The full results, methodology, analysis code and worked examples are free to read and download.
Read the study →The AI Visibility Index, July 2026
40 well-known brands, plus UK fintech and B2B SaaS editions of 20 each, scored on Claude with live web search through the Anthropic API. Scores are comparable within an edition only, and two methodology changes since mean a check run today is a different quantity.
Published, with the method and its limits stated on the page.
Read the index →How often AI gets brands wrong, July 2026
30 brands described by one assistant (Claude, via the Anthropic API) from memory and then with live web search, each claim checked against the brand's own published facts: 27 of 30 misdescribed from memory, 13 of 30 with search on.
Published. One assistant, API route, not a consumer-interface measurement; the raw answers were not retained in a publishable form, so the figures stand as recorded at the time.
Read the study →Patterns from product data, not a study
These come from watching thousands of customer checks. They are consistent enough to act on and not measured to a standard we would publish as findings, which is why they carry no percentages.
Third-party pages carry most category answers
When AI recommends suppliers for a category question, the pages it cites are mostly review sites, community threads and roundups rather than the suppliers' own sites. The UK AI Shortlist Study measures this directly.
Community discussion is cited often
Reddit and forum threads appear regularly among cited sources for buyer questions, though their share moves month to month and differs by engine.
Answer-first pages are easier to lift
Pages that state a direct answer in their first sentence are the ones we see quoted in answers; pages that bury the answer tend to be retrieved without being cited.
Concrete numbers get quoted
Content carrying specific figures, dates and named sources shows up in answers more than general prose does.
See whether AI mentions your business.
Run a free check in under a minute. No card, no login to see your score.