Studies / Sources
When an AI names a dentist or a law firm, it is usually repeating something it read. We recorded more than 70,000 sources behind AI answers, and each AI turned out to read from its own shelf.
What we found
Each AI reads from different websites.
838 vs 0
In Belleville, one review site was cited 838 times by ChatGPT, and not once by Claude.
Finding 01 ↓A new page can become the AI's top source within days.
10 days
The most-cited page in our Toronto law study had been published ten days before we asked.
Finding 05 ↓Being on a page the AI reads doesn't mean it repeats it.
50 → 0
One AI cited an article calling a firm "the #1 choice" in all 50 answers, and named that firm in none.
Finding 06 ↓Dentists in Peterborough and Belleville, and personal injury lawyers in Toronto. ChatGPT, Claude, Gemini, Perplexity and Google's AI answer.
Why it matters
AI answers are built from other websites: ranking guides, "best of" lists, review sites, directories, forums, news, and businesses' own pages. Two firms can be equally good and still be named very differently, because the pages the AI happens to read list one and not the other. A mention rate tells you where you stand. The sources show you what the AI was reading when it decided.
What the work delivers
For a business in a market we've measured, we can lay out the pages behind its answers, platform by platform. Facts only: we show what the AI read and what those pages say, and leave the conclusions to you.
What each AI reads
The pages each platform cites most when it answers in your market.
Whether they list you
For each of those pages: are you on it, and where.
What they say about you
The lines the AI is reading about your business, good or bad, word for word.
Along the way
Patterns from dental and legal markets. No business or website is named.
01
Same question, same city, different reading lists. In Belleville, one review site was cited 838 times by ChatGPT and not once by Claude. In Toronto law, ChatGPT leaned on a lawyer-ranking guide while Claude leaned on two "best of" pages that the other AIs barely touched.
838
ChatGPT citations of one review site, Belleville
0
Claude citations of the same site
02
Leaving out the practices' own sites, five or six organisations made up most of what the AIs cited about a town's dentists: 74% in Peterborough and 64% in Belleville.
5–6 sites
behind 64–74% of outside citations
03
In both dental cities, no practice was named often without the AI also reading its own website. In Toronto law, the website of one of the most-named firms was cited by two platforms and never by the other three. We can't say which comes first: AI may name what it can read, or read what it already names.
04
In Toronto law, many of the most-cited pages were "top 10" style articles on general-interest or out-of-town websites, most without a sponsorship label. One was a list of Toronto's top injury lawyers published by a news site in Florida. Peer-reviewed rankings and lawyer directories were in the mix too, but they were far from the whole picture.
05
The single most-cited page in our Toronto law study had been published just ten days before we asked. Four different AIs were already citing it.
10 days
from publication to most-cited page
06
One article crowned a law firm "the #1 choice". One AI cited that article in all 50 of its answers, and named that firm in none of them. Reading a page doesn't mean repeating what it says.
50 → 0
answers citing the page vs naming its "#1"
07
One page that a single AI read in 47 of its 50 Toronto law answers printed complaints about fees and slow responses at four of the city's most-named firms. If an AI is reading it, so is anyone who asks that AI. Most businesses have no idea which of these pages exist.
Which websites the AIs read about a particular business, and what those pages say, goes to that business. We don't publish it, and we don't sell placement on any of these sites.
We can map the sources behind the answers in your industry and your market.
These are patterns we observed, not causes. We record what the AIs cite and what they say; we don't claim that one causes the other. The dental figures come from two cities measured over the summer; the legal figures come from a smaller Toronto study (50 answers per platform), so treat those as early signals.