AI visibility November 2026
What AI tells your customers, checked against your own website.
296 of 566 had at least one answer that was wrong or half right.
For 225, at least one was flatly wrong.
All 3,265 answers
- 2,328 right
- 171 half right
- 354 wrong
- 412 no answer
Asked in November 2026 about 566 businesses with no chatbot of their own (312 dental practices, 254 others); every answer checked against the business's own website.
Half right: the right fact with a condition that matters left out, or right only after a wrong first line.
Three answers, word for word
What AI told a customer, beside what the website says.
A question that assumed the practice does NHS dentistry: can I book an NHS check-up there?
Perplexity said“Not as a new NHS patient for a routine check-up: the NHS listing says [the practice] does not currently accept new NHS patients for routine care… The practice does offer NHS check-ups for patients it can see…”
The website says“A fully private service.”
Where the answer came fromAn NHS listing, a 2017 inspection report and NHS-dentist directories that disagreed with each other. The practice's own website wasn't among them.
A UK dental practice · 28 Oct 2026
How much is an NHS Band 2 course of treatment there?
Perplexity said“A Band 2 NHS course of treatment at [the practice] is £75.30, according to the practice’s fee page.”
The website saysBand 2: £76.60.
Where the answer came fromIt credited the practice's own page, which says £76.60. £75.30 is the old national charge, on a money-advice page lower in its sources.
A UK dental practice · 27 Oct 2026
What time does it open and close this Thursday?
What Perplexity said, in shortThursday hours from the afternoon into the evening, ending hours after the practice shuts.
The website saysOpen from nine in the morning until five on Thursdays.
Where the answer came fromThe regulator's inspection report, which differs from the practice's homepage. The homepage was in its sources too.
A UK dental practice · 26 Oct 2026
What was in its sources
With your own website in its sources, it was right far more often.
A pattern, not proof of the cause: most slips, 208 of 280, still came with the page listed.
Your website among Perplexity's sources 1,407 answers
- 76% right (1,075)
- 15% wrong or half right (208)
- 9% no answer (124)
Not among them 322 answers
- 27% right (87)
- 22% wrong or half right (72)
- 51% no answer (163)
Which facts slip
Where the slips were.
- 1 Other facts and policies 15%122 of 791
- 2 Opening times 15%80 of 551
- 3 Cancellation and notice 14%23 of 163
- 4 Prices and fees 14%153 of 1,125
- 5 Deposits 13%7 of 54
Answers wrong or half right, of each kind.
By sector
Every sector we checked had slips.
- 1 business 82%9 of 11
- 2 Care providers 67%12 of 18
- 3 state-school 64%14 of 22
- 4 Private clinics 60%36 of 60
- 5 Dental practices 50%157 of 312
- 6 Solicitors 50%8 of 16
- 7 Schools 44%26 of 59
- 8 Accountants 40%4 of 10
- 9 Nurseries 38%5 of 13
- Others 56%25 of 45
Businesses with at least one answer wrong or half right. Others: dental suppliers, estate_agent, opticians, vets, health, deli, beauty, caravan_site, college, pub, furniture, convenience, pharmacy, care-home, golf_course, community_centre, doctors, supermarket, social_facility, indoor_play, clothes, doityourself, cinema, university, restaurant, charities and hotel.
What to do about it
Start by seeing what AI says about you.
Free check
Give me your website. I'll ask your chatbot, or ChatGPT or Perplexity if you haven't got one, three questions your customers ask, and email you what it said beside what your own pages say.
Get a free checkFour things to do yourself
Put each fact customers ask about on a normal page of your website, in plain words, once. A price that only lives in a PDF or a picture is easy to miss.
Make your NHS listing, CQC entry, Google profile and directory pages say the same hours and status as your website.
Ask ChatGPT, Perplexity and Gemini your five most-asked questions in fresh chats, and compare the answers with your own pages.
Ask one with a wrong belief built in, like “since you're open on Sundays…”, because customers ask that way too.
None of this makes AI name you or get you right. It makes the right answer easier to find.
If you have a chatbot of your own
56 of 120 business chatbots got a fact wrong or half right.
A chatbot on your own website is a different test, so we count it separately. 21 of the 120 got every answer right.
How the chatbot check works → Dental practices' own chatbots →The detail
How we checked, and what owners ask.
How we checked
We found the 566 businesses with our own search, so the figures describe them, not every business. Perplexity was asked about 382, ChatGPT about 139 and Gemini about 192, and 130 were asked more than one.
Each got up to five questions per assistant (most got five), of five kinds set before we asked. They were a price with a condition, a fact on a deeper page, today's hours or a current price, a question built on a wrong belief, and a plain fact. We asked from 26 to 5 November 2026, with web search on and a fresh chat each time. Gemini was asked as a visitor would, without signing in. The ranking of kinds leaves out the questions built on a wrong belief, so the traps don't set it.
Two graders marked every answer apart, and each disagreement was settled by re-reading the website. Saying a fact isn't published, when the site has it, counts as no answer. We left out 14 answers where the right answer was unclear and 1 business we had partly checked against a sister site.
Is this LLM SEO?
Partly. LLM SEO and most AI visibility tools are about how often AI mentions you. This checks whether what it says is true, which is the part a customer acts on. For the mention side, see how to check if ChatGPT cites your site.
Does an llms.txt file fix it?
It's a tidy-up. In an early pilot of eight businesses, two of the three wrong answers came from sites that already had one. The assistant has to reach the page with the fact on it.
Which assistant is most accurate about businesses?
No clear winner yet. On our records, 17% of ChatGPT's answers and 16% of Perplexity's were wrong or half right. Gemini's was 15%, but we've only just started asking it, so we don't rank them yet.
Does a listed source mean AI read it?
No. A source in the list is no proof the answer came from it, as the Band 2 example above shows. We checked that on its own.