ChatGPT is wrong about your business. Here's how to fix it.

The call usually goes like this. An owner asked ChatGPT about her own company and it named the wrong owner, or listed a location that closed in 2022, or confidently described a service she's never offered. Then she checked Google's AI Overview and it got something else wrong. Then she clicked "report" and nothing changed. Now she's wondering whether the engines are quietly doing this to every prospect who asks about her.

They might be. Here's the thing to understand first: the engine isn't lying about you on purpose, and it isn't reading from a file you can edit. It's assembling an answer from everything it can find, and somewhere in what it can find, the wrong fact is better supported than the right one. The fix is to change what it finds. This guide is how.

Why the engines get your business wrong

ChatGPT, Perplexity, Gemini and Google's AI Overviews answer a question about your business by pulling from sources (your site, your Business Profile, directories, review platforms, news, old pages, forum threads) and, for some of them, from what the model learned during training. Then they write the answer that best fits the evidence. Three kinds of wrong come out of that process, and they need different fixes.

One more cause sits under all three: your own sources disagree with each other. A firm I audited this month had four different review counts, four different "clients served" figures and four different versions of a key statistic across its own pages. When your own site can't agree on the facts, the engine picks one, or hedges, or quotes a competitor whose facts are stable. Consistency isn't a nice-to-have here. It's the mechanism.

Step 1: capture the wrong answer, with its sources

Before changing anything, write down exactly what each engine says and where it says it got it. Run these ten prompts on ChatGPT (with search enabled), Perplexity, Gemini, and a normal Google search to see whether an AI Overview appears:

1. Who is [business name] and what do they do?
2. Where is [business name] located and what are their hours?
3. Who owns or runs [business name]?
4. How much does [your main service] cost at [business name]?
5. Is [business name] any good? What do reviews say?
6. Does [business name] offer [a service you DON'T offer]?
7. Compare [business name] and [your closest competitor].
8. Who are the best [what you do] in [your city]?
9. I need [your main service] in [your city]. Who should I call?
10. What is [business name]'s phone number and website?

For each answer, log three things in a spreadsheet: the wrong fact (quoted), the sources the engine cited or linked, and which kind of wrong it is from the table above. ChatGPT with search and Perplexity show citations inline; Google's AI Overview shows source links to the side or under the text. The sources column is the diagnosis. Nine times out of ten, the wrong fact is sitting in one of the cited pages, and now you know which page.

Prompt 6 is deliberate. Ask about a service you don't offer and see whether the engine says yes. That's the invented category, and it tells you how thin your evidence is.

Step 2: write the one-page fact sheet

This is the single most useful document in AI-search work and almost nobody has one. One page, owned by one person, listing every fact about the business that should be identical everywhere. Every page on the site, every listing, every bio, every piece of schema gets checked against it. When a fact changes, the sheet changes first and the rest follows.

BUSINESS FACT SHEET  (one page; the version the whole web should agree with)
Last updated: 2026-10-21   Owner of this sheet: ______________

IDENTITY
  Brand name (exact spelling, punctuation):   ______________________________
  Legal name:                                  ______________________________
  Former names / common misspellings:          ______________________________
  Founded (year):                              ______
  Owner / principals (name, title):            ______________________________
  Licence or registration numbers:             ______________________________

WHERE
  Address (each location, exactly as on the Business Profile): ______________
  Phone (one number per location):             ______________________________
  Hours:                                       ______________________________
  Service area (named counties/cities):        ______________________________
  Locations CLOSED or MOVED (and when):        ______________________________

WHAT
  Services, one line each, with price or range and the URL of the page:
    1. ________________________________  $______   /url/
    2. ________________________________  $______   /url/
    3. ________________________________  $______   /url/
  What we do NOT do (the things people wrongly assume):  ____________________

PROOF (dated)
  Review count and rating, by platform, as of ____:   Google ___ / ___  Other ___
  Clients served / jobs completed (one number, one definition): _____________
  Years in business:  ____       Credentials / memberships:  ________________

WHERE THESE FACTS APPEAR (every one should match this sheet)
  [ ] Website: homepage, about, contact, each service page, each bio
  [ ] Schema markup on the site
  [ ] Google Business Profile       [ ] Bing Places
  [ ] Apple Business Connect        [ ] Facebook / Instagram / LinkedIn
  [ ] Industry directories: ___________________________________________
  [ ] Regulator / licence directories
  [ ] Press and old articles (list):  __________________________________

Two rules for filling it in. First, one number, one definition. If "clients served" is 3,500 on one page and 5,000 on another because one counts matters and the other counts people, pick one definition and use it everywhere. Second, date the proof. "915 Google reviews as of October 2026" can be checked; "900+ reviews" can't, and it will be stale by spring.

Step 3: fix the sources, in this order

Work from the sources you control to the ones you influence to the ones you can only ask about. Each level is cheaper and faster than the next, and the engines weight your own site and your Business Profile heavily for identity facts.

Your website

• State the facts in plain sentences on the homepage, about page, contact page and every service page: who, what, where, since when, for how much. Not in a hero image, not in a PDF. In text an engine can lift.
• Add or correct the schema so it matches the sheet exactly: the business block, a Person block for each principal, Service blocks with prices. The copy-paste versions are here.
• Find the stale pages. Search Google for site:yoursite.com "old phone number" and site:yoursite.com "old address". Old landing pages, press-release pages and event pages are the usual culprits. Update them or redirect them to the current page.
• Kill the near-duplicates. If you have four versions of the homepage live (a paid landing page here, a test page there), each with slightly different numbers, consolidate them. Canonical tags and redirects, not noindex-and-hope.
• Add the negatives. What you don't do, where you don't work, which locations closed and when.

Google Business Profile and Bing Places

• Match the sheet field by field: name (exactly as on the site), address, phone, hours, categories, services, description. If a location closed, mark it closed rather than deleting it, so the engine learns the fact rather than losing it.
• Do the same on Bing Places, and register the site in Bing Webmaster Tools. ChatGPT's web search draws on Bing's index among its sources, which makes Bing more important for ChatGPT answers than its share of searches suggests.
• If Google shows a Knowledge Panel for the business, claim it and suggest edits from there; those corrections tend to stick faster than anything else on this list.

Directories, review platforms and the regulator

• Go to the sources your log says the engines cited and correct them first: Yelp, Avvo, Justia, FindLaw, Angi, Houzz, the BBB, the chamber of commerce, industry associations, the state licensing database. The ones the engines actually used matter more than the ones a listings tool says you "should" be on.
• For a regulator or bar directory, the fix is usually a form and a few weeks. Do it anyway; the engines treat those as authoritative.
• Don't build fifty new listings to drown out one wrong one. Google's own guidance warns that manufactured mentions "aren't as helpful as they might seem," and a burst of thin listings can introduce new inconsistencies.

Press, old articles and third-party pages

• For an outdated news article, ask the outlet for a correction or an update note. Many will add "Editor's note: as of 2026, the firm is located at…" if you make it easy.
• Where you can't change the source, outweigh it: publish the current account on your own site (a dated "About our move" or "Our history" page works), make it quotable, and get it corroborated by newer third-party mentions.
• Wikipedia and Wikidata only if the business is notable enough to have an entry on its own merits. Editing your own entry, or creating one that doesn't belong, gets reverted and can draw attention you don't want.

Step 4: tell each engine, with the right expectations

Feedback is worth sending, and it is not the fix. It feeds product improvement, not a corrections file about your business. The sources are the fix. With that said:

A note on timing. Answers grounded in live search (AI Overviews, ChatGPT with search, Perplexity) update when the sources do, typically within days to a few weeks. A fact that's baked into a model's training data can persist until the model is updated, which can be months. You can't control that part, which is exactly why the live-search path is where to spend the effort.

Special cases

A closed location that won't die. Mark it closed on the Business Profile (don't delete it), add a line on the site's locations page saying it closed and when, redirect the old location page to that line, and correct the directories that still list it.

A former partner or employee named as the owner. Update the about page and the Person schema, make sure the regulator's directory shows the current principals, and ask LinkedIn-level sources (the person's own profile, press) to reflect the change. Engines lean on bar and licence directories for this; get those right first.

A name collision. Two "Summit Dental" practices in neighbouring counties will be conflated until one of them is consistently more specific. Add the city or a descriptor to the brand everywhere it appears, use alternateName in the schema, and state the distinction on the about page.

Wrong prices. Almost always because the price isn't on your site, so the engine took it from a directory, a forum or a competitor. Put a current, dated price or range on the service page and in the FAQ. Mark it up as a Service with an Offer.

A negative claim from a review, a lawsuit or an old story. If it's false, that's your lawyer's conversation. If it's old and resolved, publish the accurate, current account on your own site, make it the clearest thing on the web about the matter, and build newer corroboration. Engines prefer recent, well-supported facts over old, isolated ones.

Step 5: keep it fixed

Put a thirty-minute block on the calendar for the first of every month. Re-run the ten prompts on the four engines, log the answers next to last month's, and note anything that drifted. When a fact changes in the business (a hire, a move, a new price), change the sheet first and work down the checklist on the sheet the same week. That's the whole maintenance program, and it's the one I run for clients: a spreadsheet with a row per prompt per engine per month. It's the only AI-visibility report that matters.

The deeper goal, once the facts are right, is to be named at all when a buyer asks who to hire. That's the next guide.

Questions I get about this