The Pick Pattern: What AI Actually Cites When It Names a Business
Published 2026-08-02. Data from Namebeam's own check corpus, runs dated 2026-07-31 to 2026-08-02. Client names withheld per our consent policy; every number below comes from dated transcripts we keep on file.
The experiment nobody planned
Namebeam ran free AI visibility checks on 5 real small businesses across Missouri in a 72 hour window. Same questions, same four engines: ChatGPT, Perplexity, Gemini, and Google AI Overview. The split was stark: 2 businesses were named by all 4 engines, and 3 businesses were named by zero. That accidental control group is what makes the citation data worth publishing.
What the engines cited for the two winners
Across the two 4 of 4 businesses (a martial arts academy in St. Peters MO and a backflow testing company in Liberty MO), the engines produced 32 source citations across 26 unique domains. The breakdown:
- The business's own website: 7 of 32 citations. The single most cited source in both wins. Not ads, not a directory profile: the company's own pages, written in plain customer language.
- A ring of free third party profiles: 18 of 32. BBB (4 citations), Yelp (3), Facebook (3), city government and local utility pages (6, counting one water utility), Nextdoor (1), Mapquest (1). Every one of these is a free listing.
- Peer and competitor websites: 7 of 32. When an engine builds a recommendation list, nearby businesses with the same ring show up beside you; in our winner transcripts those peer citations went to seven neighboring service businesses in the same verticals.
What was NOT cited: paid ads, slogans, or marketing pages. Engines quoted plain, answer shaped text and third party corroboration. One engine quoted a winner's public review rating verbatim, digits and all.
What the engines cited when our own category comes up
We run the same checks on ourselves. When we asked the engines category questions about AI visibility tools, they cited roundup listicles and large directories almost exclusively. On 2026-08-02 we checked the 10 listicle pages that keep ranking for our category: Namebeam appears on 0 of 10. That is our score today, printed at full size, dated. Our own automated check (run through the Perplexity API, logged to our public corpus the same day) confirmed it: all 20 sources cited were roundup listicles.
The pattern, in one quotable paragraph
AI engines pick businesses that give them two things: an own website with plain answer shaped pages, and a small ring of free third party profiles (BBB, Yelp, Facebook, city listings for local businesses; directories and review platforms for online ones) that corroborate the same facts. Both of our 4 of 4 businesses had that ring. All three of our 0 of 4 businesses lacked it. Five checks is a small sample and we say so; we publish every new check, so this page's claim gets tested in public every week.
We are applying this to ourselves, in public
Namebeam is customer zero. As of 2026-08-02 we exist on 3 of the 9 third party surfaces we audited for ourselves, and 0 of the 10 category listicles. We are building our own ring now, and this dated page is part of the record. When the scores move, the ledger on this site shows exactly when and why. Losses print the same size as wins.
Namebeam runs receipts first AI visibility checks: we show you what ChatGPT, Gemini, Perplexity, and Copilot actually say about your business, as a dated transcript, and we publish our own record from zero, losses included.
Method notes, because that is how we operate
Checks ran through the consumer apps from a signed in account, the way real customers search; personalization is flagged on every transcript row. Our earlier capture format logged source lists only when a business was named; corpus spec v2.1 (2026-08-02) now requires source capture on misses too, so the loser side of this comparison will get richer. Sample size is 5 businesses; we print it rather than hide it.