Someone wakes at 2am with a cracked molar and asks Google for a dentist who takes their insurance. Someone else, hours after a car accident, asks ChatGPT whether a firm offers free consultations. Neither of them scrolls ten blue links anymore. They read one paragraph and act on it. We wanted to know whether that paragraph is true. About one in four times, it is not.
- 27% of Google AI Overviews answers and 20% of ChatGPT answers contained something the business does not publish. Wrong hours, wrong insurance, services that do not exist.
- When a business website is vague, Google states invented specifics as fact 80% of the time. ChatGPT does it 10% of the time, because it usually names the directory it came from.
- The invented details are lifted from third-party listings. Sharecare, WebMD, ADA findadentist and Delta Dental’s own directory all supplied carrier lists that practices never claimed.
- Some of this is not the AI’s fault. Of 59 businesses publishing no hours on their own site, 46 had set their Google listing to “Open 24 hours.” The engines were repeating the business’s own claim.
- Accuracy climbs to 84% and 89% for businesses that publish plainly. Most of this is fixable, and the fix belongs to the business.
One in four AI answers about a local business gets something wrong.
Our last Labs study looked at which law firms Google’s AI Overviews recommends. The obvious next question was whether the engines are right about the businesses they name.
So we took 600 small businesses: 200 dental practices and orthodontists, 200 personal injury firms, and 200 home service companies. We asked both engines the two things every prospective client wants to know. Are you open? Do you take my insurance, handle my kind of case, offer the service I need?
The first thing we noticed was that the two engines kept contradicting each other about the same business on the same afternoon. One reported a plumbing company as open around the clock. The other gave weekday hours. Both cited the company’s own website. They could not both be reading it correctly.
When a website is vague, both engines invent the specifics, and Google rarely says where they came from.
Nearly every dental practice writes some version of “we accept most major insurance plans,” and names none. The sentence is true, reasonable, and legally careful. It also leaves a hole. The engines fill it, usually with a carrier list lifted from a directory, then hand it to the reader as the practice’s own.
When a business’s own website was vague about a fact, how often did each engine fill in the specifics anyway, and how often did it hand them to the reader as the business’s own?
Filled in specifics anyway 88%
80% stated them as fact, naming no sourceFilled in specifics anyway 80%
10% stated them as fact, naming no sourceBoth engines reached past the website at close to the same rate. What separates them is disclosure: ChatGPT usually named the directory it had read, Google usually did not.
Orthodontic Practice A names no carriers anywhere. A blog post on their site asks “Does Cigna, Delta Dental, Humana, or MetLife cover Invisalign?” AI Overviews turned that question into an answer: those four carriers are accepted, plus Aetna, which appears nowhere on the site.
A patient books, arrives, and learns you are out of network. You have burned a chair slot, made a bad first impression, and may be arguing over a bill for a promise you never made.
The same engine invented a free consultation for one firm and denied it for another.
Neither site says what a consultation costs. AI Overviews answered anyway, and answered differently each time.
Law Firm A. The site’s only wording is “Schedule a Consultation.” AI Overviews answered: “Yes. Law Firm A offers a free initial consultation.” People are calling that firm expecting free advice it never advertised.
Law Firm B. The site says nothing about consultation cost. AI Overviews answered: “No, they do not offer free consultations,” inventing a rationale about intensive paid assessments. That is a prospect talked out of picking up the phone. ChatGPT, asked the same question, said the terms are discussed during a screening call, which is accurate.
Home Services Company A posts Mon to Fri, 8am to 4:30pm, and advertises no emergency service. AI Overviews described 24/7 after-hours emergency coverage and a “Planned Service Agreement” priority tier. Neither exists. Someone calls at midnight expecting a truck.
Both engines read a 24/7 phone line as 24-hour opening hours.
In these industries, “available 24/7” nearly always means an answering service or an emergency line. Both engines regularly report it as the business being open around the clock, at almost identical rates: Google 25%, ChatGPT 21%.
We kept finding businesses with no hours anywhere on their website, and wondered where the engines were getting hours at all. So we opened the Google Business Profiles of 59 of them.
had their Google listing set to “Open 24 hours.” All 59 had hours on Google, though none had hours on their own website.
That is usually a lead-capture setting, ticked once during onboarding and never looked at again. But it is the business’s own listing, which means the engine reporting it is not making a mistake. It is accurately repeating something untrue that the business published about itself.
When an AI says a law firm is open at 3am, it is usually repeating what the firm told Google.
Law Firm C and Law Firm D both set primary hours to 24 hours, then list their real hours, 8:30am to 5:30pm, in a secondary field the engines do not read.
Someone calls at 9pm expecting a person, gets voicemail, and calls the next firm on the list.
Most businesses publish their hours in more than one place, and the versions disagree.
Only 55% of the businesses we studied publish their hours cleanly. The rest have two or three versions in different places: the contact page says one thing, the footer another, and the structured data in the page’s code a third. Most owners have no idea the third one exists, and it is frequently the one the engines read.
Home Services Company B. The contact page says Mon to Fri, 8 to 5, with weekend emergency service. The footer on that same page says “Open 24 hrs, Mon to Sun.”
Whichever version the engine picks is a coin flip, and it may be the wrong one.
We had a list of answers the engines had gotten wrong, and went back to check each one against the business’s website. Again and again, the fact we were grading against was not on the website at all. The practice did not name its carriers. The firm did not post office hours. We had been treating a directory listing as the truth, and the engine had been doing exactly the same thing.
That changed the question. These systems are not careless readers. A third of these businesses are not telling them anything to read, and something always fills the silence.
One caveat worth knowing. These systems change constantly. Run the same searches next month and you may get different answers, which is exactly why a single check is not enough.
How to fix what AI gets wrong about your business.
The uncomfortable finding is also the encouraging one. If the engines were simply unreliable, there would be nothing to do but wait for Google to improve. They are mostly reflecting what they find, so most of this is within reach. The first three take under an hour.
- Check your Google Business Profile hours right now — If it says “Open 24 hours” and you are not, change it. The edit takes about two minutes. This is the one place where the bad information starts with you rather than with the engine. (For the owner)
- Search your own business in Google and ChatGPT — Ask what insurance you accept and what your hours are. Whatever comes back is what your prospective clients are reading. Most owners have never looked. (For the owner)
- Rewrite “available 24/7” to say what is actually available — “Our phones are answered 24/7. Office hours are Monday to Friday, 8 to 5” gets read correctly. The bare phrase does not. One firm in our sample words it this way, and both engines handled it accurately. (For whoever edits the website)
- Audit the directories before touching your site copy — The invented carrier lists came from Sharecare, WebMD, ADA findadentist and Delta Dental’s find-a-dentist tool. Correcting those listings addresses the real source, and avoids the liability of publishing a definitive carrier list that goes stale as networks change. (For the agency)
- Make your hours say one thing everywhere — Contact page, footer, and the structured data in your site’s code should match. This is usually a theme or plugin setting rather than a content edit, worth a developer’s twenty minutes. (For the agency or developer)
If you are the agency
- This is a billable audit that takes under an hour per client and produces findings a client can immediately verify.
- Multi-location clients are most exposed. We repeatedly found one location’s insurance list applied to another, including plans a location explicitly does not accept.
- Check both engines. They make different mistakes at very different rates, so testing one gives you half the picture.
- Most hours conflicts live in structured data, invisible to anyone reviewing the page visually.
Verdict
AI engines are not inventing your business out of nothing. They are reading what you publish, and when you publish nothing, they read what a directory published about you years ago, then hand it to a prospective client in a confident paragraph with your name on it.
Google does this without signalling any uncertainty in four out of five cases. That is the finding with teeth. Not that the answer is sometimes wrong, but that nothing in it tells the reader to check.
The businesses that publish plainly, with real hours, named carriers, and an explicit line about what “24/7” covers, get accurate answers from both engines the large majority of the time. The gap between those two groups is the entire story, and which group you are in is still your choice.
See exactly what AI is telling people about your business
We run the same audit against your hours, your insurance or practice areas, and the services you actually offer. You get every answer the engines got wrong, side by side with what your site publishes, and the source each engine read before it decided.
Get your accuracy report
AI Accuracy Audit · Business Profile
Dental Practice A
What Google and ChatGPT tell people about this practice
We tested 600 businesses: 200 dental practices and orthodontists, 200 personal-injury-focused law firms, and 200 home service companies. Two questions each, for 1,200 graded answers. Google answers were generated using Gemini 3.5 Flash with Google Search grounding enabled, the model that powers AI Overviews. ChatGPT answers used the OpenAI Responses API with web search enabled. Ground truth was verified by hand against each business’s own website, including FAQ pages, insurance pages, location sub-pages and structured data, and every disputed answer was re-checked by opening the cited source directly. Each business was independently rated on how clearly it publishes each fact, without reference to what either engine answered. An answer counts as wrong if it differs from what the business publishes, regardless of which source the engine cited. Where a business publishes conflicting versions, matching any of them counts as correct. Accuracy figures for clear and vague publishers are estimated from a random sample and carry a margin of a few points. Full per-business results available on request.