AI Assistants Are Getting Your Business Details Wrong
Someone asked ChatGPT when you close. It answered with confidence.
The Number That Should Worry You
Researchers queried ChatGPT, Perplexity and Gemini more than 13,000 times about a set of real companies and checked the answers against the facts. For 93% of those businesses, at least one basic fact came back either invented or missing.
The gap between company sizes is the part that matters if you run a shop. Around half of the small and mid-sized businesses in the study had at least one fabricated fact returned about them, against roughly a third of large brands. Brand name confusion, where the model attributes your work, your address or your reviews to a different company, showed up about five times more often for smaller businesses.
These are the same assistants your customers now open before they call you. An answer that sounds certain and is wrong costs you the appointment, and you never see it happen.
Why Smaller Businesses Get It Worse
A language model answers from what it has absorbed about you. A national chain has thousands of pages, news mentions, directory entries and structured data behind its name, all of it repeating the same facts. The model has enough material to be confident and correct at the same time.
A single-location business has a website, a Google Business Profile, a handful of directory listings and some reviews. When the model is asked something those sources do not cover, it fills the gap by pattern matching against businesses that resemble yours. That is how a dental practice with Saturday hours gets described as closed on weekends, and how a shop that has never done commercial work gets listed as a commercial contractor.
Sharing a name with another business makes it worse. So does a recent move, a rebrand, or a services page that describes what you do in language nobody else uses.
What Gets Invented Most Often
Hours are the most common. Weekend and holiday hours are the worst, because they change and most sources record them poorly.
Phone numbers come next, usually pulled from an old listing or a call tracking number that was retired two years ago. Service area is a close third, and the failure runs both ways. Models tell people you cover towns you have never worked in, and tell people in your own town to look elsewhere.
Then prices. Almost no local business publishes a full price list, so the model estimates one from industry averages. A customer who arrives quoting a number you never gave is starting the relationship in a hole.
Last, and hardest to catch, is whether you are open at all. Businesses that closed a location, changed names, or went quiet online for a stretch get described in the past tense.
How to Check What AI Says About You
Set aside forty minutes and do this properly, because a casual look will miss the pattern.
Open a fresh session in each assistant your customers use. A logged-out or temporary chat matters, since your own history will steer the answer toward what you already told it.
Ask the same set of questions in each one. What are the hours for {your business} in {your city}. What is the phone number. What services does it offer. Does it serve {a town at the edge of your area}. How much does it charge for {your most common job}. Is it still open. Who owns it. What do reviews say about it.
Record every answer verbatim in a document with the date. You want the record, because you will run this again in ninety days and the answers will have moved.
Mark each answer correct, incomplete or invented. The invented ones are your work list, ordered by what would cost you a customer.
How to Get It Corrected
There is no correction form. For most assistants you cannot file a fix and have it applied, so the work is to change what they read.
Start with your Google Business Profile, since it is the most heavily weighted source for local facts. Hours, holiday hours, phone, service area, categories and attributes all need to be current and complete rather than mostly right.
Make your own site state the facts in plain sentences on the page, in text. A model can read "We are open Saturdays from 9am to 2pm" and struggles with the same information rendered inside an image or a styled hours widget that loads after the page.
Add LocalBusiness schema carrying hours, phone, address and service area, and keep it matched to what the visible page says. Where the two disagree, you have given the model a coin to flip.
Then fix the third-party listings that repeat the wrong version. Apple Maps, Bing Places, Yelp and your main trade directories feed the same pool.
Finally, write the answer to each invented fact somewhere on your site as its own passage. If models keep telling people you do not serve a town you have covered for a decade, publish a page that says you serve it and names the neighborhoods.
Why This Needs a Person Watching It
Corrections do not hold on their own. Models refresh, sources get re-crawled, a directory reverts to an old record, and an answer you fixed in March is wrong again by August. The study above is a snapshot of a moving target.
A quarterly re-run of the audit catches most drift. Put it on the calendar with the same seriousness as filing your sales tax, because it protects the same revenue.
This is the part software cannot close by itself. A tool can flag that an answer changed. Deciding whether the new answer is wrong, working out which source caused it, and rewriting the page that fixes it takes somebody who knows what is true about your business.
What to Do This Week
Run the eight questions in three assistants and write down what comes back.
Fix your Google Business Profile hours and service area first, since they cause the most expensive errors.
Put your hours, phone and service area in readable text on your homepage and your contact page, and match your schema to it.
Diary the same audit for ninety days out.
If the answers come back badly wrong and you would rather hand this to someone, that is what we do. We run the audit, fix the sources, and re-check on a schedule so the corrections stay corrected.