When someone asks ChatGPT or Claude for a plumber, a dentist, a contractor in your town — do you come up? We ask, and we send you the answers. Not a score. The actual words.
It answered anyway — about a clinic in another country.
“is Bayview Dental Studio good”
“Do you mean Bayview Dental Studio in
South Surrey/White Rock, BC, run by Dr. [name]?
If so, it appears
generally reputable but not unanimously reviewed …”
Bayview Dental Studio is a name we made up for testing. The assistant found a real practice in British Columbia and answered as if that were the one. If your business has a common name, this is what your customers may be reading.
We've taken the dentist's name out — he's a real person and he didn't ask to be in anyone's marketing. Everything inside the quotation marks is otherwise word for word, unedited and unjoined.
That answer, the three sources it showed, the 71 pages it read to get there, and the timestamp are all archived. You get the file.
curl command, so you or your web person can run it and see the
same thing we saw.This is early. Here is exactly where it stands today — two assistants tested with live calls, three not wired up yet.
| Assistant | Status |
|---|---|
| ChatGPT | live-tested 15 Aug |
| Claude | live-tested 15 Aug |
| Google Gemini | not connected |
| Perplexity | not connected |
| Google AI Overviews | not connected |
The two that are connected have each answered twelve questions a dental patient in Seattle might ask — eight of them with no business name in them at all — for 24 answers, all stored. That is the whole of the evidence so far.
ChatGPT showed 44 links across 25 different sites. Claude showed none at all — though it had read 277 pages to answer.
The two don't mean the same thing by "source": one hands you links, the other reads and doesn't cite. So this is a difference in what they show you, not in what they read. If it holds, "get cited by AI" advice doesn't transfer between them — and we're still checking whether it's Claude's behaviour or the route we called it through. Either way you'd get the raw answers and could judge for yourself.
559 — the tests in the code, which you can run yourself.Ask it. Not your name — ask the question a customer would ask, the one with no business name in it at all, and see who comes up. That is the whole test, and you can run it yourself in ten minutes: the question pack builder writes the questions and a tally sheet, free, with no signup and nothing sent anywhere.
Two things surprise people. The assistant will answer confidently about a business with a name like yours somewhere else, as it did above. And the sites it cites are mostly ordinary business websites — a short list you can go and get onto, once you know which ones they are.
Start free. Send your business name, your town, and your website. You'll get back the actual questions we asked, the actual answers, and where you did or didn't show up. No charge, no call, no signup.
hello@doestheaiknowyou.comOne person reads these. Replies take a day or two.
Why it's $390 and not more: two assistants instead of five, and no case study to show you. Both of those are on us, so the price reflects them. It goes up when they're fixed.
Reply to the free check and say you want the full one, or write straight to us. You get an invoice, and nothing is charged before you've seen the free sample.
hello@doestheaiknowyou.comTurnaround is about a week. Not a subscription — one audit, one invoice. If it turns out we can't tell you anything you didn't know, don't pay.