When ChatGPT quotes hours you changed two years ago, an address you moved out of, or a service you stopped offering, it is not inventing things. It is repeating what it found. AI engines build answers from your website, your Google Business Profile, and the directory listings that mention you — and when those sources disagree, the engine quietly picks a winner for you. You cannot edit the chatbot. You edit what the chatbot reads.
Why AI gets your business details wrong
An AI engine answers a question about your business from two different places, and neither one is a database you control.
- What it learned during training. A snapshot of the web frozen into the model months or years ago. Your old hours can survive in there long after you changed them.
- What it retrieves right now. A live look at pages it can reach the moment someone asks — your site, your Google Business Profile, directories, review platforms, local news.
Both are reflections of what the web currently says about you. When the two disagree, or when the live sources disagree with each other, the engine resolves the conflict on its own and delivers the result in a calm, confident sentence.
That confidence is the real problem. A page of Google results shows a customer ten options and lets them judge. An AI answer shows them one, with no visible seams and no hint that anything was uncertain. If the fact it picked is three years stale, it still reads as today’s truth.
Why this matters more than it used to
According to BrightLocal's 2026 Local Consumer Review Survey, use of AI assistants to find local businesses jumped from 6% to 45% in a single year, putting AI third behind only Google and Facebook. Nearly a quarter of those people act on the AI summary without ever opening a review profile. A wrong phone number is no longer a small back-office error.
Where the wrong answer comes from
A wrong fact is always living somewhere specific. In practice there are only five places it can be, and knowing which one you are dealing with tells you both how to fix it and how long the fix will take.
- Your own website. A stale service page, an old footer, a location page you forgot existed, or schema markup that still carries last year’s hours.
- Your Google Business Profile. The single heaviest local signal, and the one most likely to be quoted verbatim.
- Directory listings. Yelp, Yellow Pages, Apple Maps, Bing Places, and the long tail of industry directories that copied your details once and never checked again.
- Third-party mentions. Press coverage, partner sites, and aggregators that scraped you years ago.
- The model’s training data. Not editable by anyone outside the AI company, and the only category you cannot directly touch.
| Where the error lives | How you fix it | Typical time to update |
|---|---|---|
| Your website | Update the page and its schema, then request a recrawl | 1–3 weeks |
| Google Business Profile | Edit the fields directly and wait for the change to be approved | Days |
| Directory listings | Claim the listing, correct the NAP, remove duplicates | 2–6 weeks |
| Third-party mentions | Request a correction, or publish a newer, more authoritative page | 4–12 weeks |
| Model training data | Cannot be edited — you outweigh it with fresh, consistent sources | Next model refresh |
Why you cannot just correct the chatbot
The instinct is to argue with it. You open ChatGPT, tell it the address is wrong, and it apologises and uses the correct one. Problem solved.
It is not. That correction applies to your conversation and nothing else. The next customer who asks the same question starts from a blank slate and gets the same wrong answer. You have fixed the symptom for exactly one person: yourself.
Reporting the answer is still worth doing — thumbs-down the response and use the report option with a short factual note. But treat it as filing a complaint, not making an edit. It is a signal to the AI company, not a switch that changes what the model says tomorrow.
You cannot log into ChatGPT and correct it. You correct what it reads, and then you wait for it to read again.
The five-step correction playbook
1. Audit before you fix anything. Ask ChatGPT, Gemini, Perplexity, and Google’s AI results the questions a real customer would ask — your hours, your address, your prices, whether you offer a particular service. Record the exact prompt, the exact wrong claim, and any sources the answer cited. Without this you are guessing at what is broken.
2. Trace the fact back to its source. Follow the citations if the engine gives them. If it does not, search the wrong fact itself — the old phone number, the former address — and see which pages still carry it. The result is usually a short, specific list.
3. Fix the highest-authority source first. Your own website and your Google Business Profile carry the most weight and update the fastest. Correct those before you touch anything else, and make sure your schema markup matches the visible page. There is no point cleaning up a directory while your own site still contradicts you.
4. Make every source agree. This is the step people skip, and it is the one that actually holds. If your name, address, phone, hours, and service area are identical everywhere, the engine has nothing to resolve and stops guessing. If they disagree, you have handed it a vote — and one confident stale listing can outvote your own website. This is the same consistency problem behind citations in AI search, and it is why citation cleanup fixes accuracy problems people did not realise were citation problems.
5. Re-test on a schedule. Run the same prompts from step one every month. Corrections land at different speeds across different engines, and a listing you fixed can be overwritten by an automated data feed later. Accuracy is a state you maintain, not a task you complete.
How long the fix takes to show up
Retrieved facts update quickly. Once your site and Google Business Profile agree and have been recrawled, engines that check live sources — Perplexity and Google’s AI results especially — often reflect the change within days to a few weeks.
Trained facts are slower and less predictable. If a wrong detail is baked into a model’s training data rather than pulled live, no amount of directory cleanup flips it immediately. What you are doing instead is stacking the live evidence so heavily in your favour that the retrieved answer beats the remembered one, and so the next model refresh learns the corrected version.
This is why accuracy work and general AEO timelines look similar: first movement in weeks, dependable consistency across engines over a few months.
Frequently asked
Sources
- BrightLocal, Local Consumer Review Survey 2026 — 1,002 US adult consumers surveyed via SurveyMonkey, of whom 455 had used AI for local business recommendations. Verified Aug 2026.
- BrightLocal, consumer trust in AI local business recommendations (2026 Local Consumer Review Survey companion analysis). Verified Aug 2026.
Be the business AI recommends
Start with a free call. We'll audit what AI engines currently say about your business and show you exactly where the wrong information is coming from.
Book a free call