AI Says Wrong Things About My Business: How to Fix It

If ChatGPT says your business is closed, or gets your address, phone, or services wrong, the cause is almost always a stale source page — not the AI itself. The fix is to find every page carrying the wrong fact, correct them at the source starting with Google Business Profile, and wait days to weeks for the answers to catch up.

That’s the whole method. The rest of this page is how to do each step properly, why you can’t just “report it to ChatGPT,” and what a realistic timeline looks like.

Why AI gets your business wrong

Chat models answer local questions from two places. First, training data: a static snapshot of the web that is months or years old by the time you’re talking to the model. Second, live retrieval: when search is enabled, the system pulls current web pages and composes an answer from them, usually with citations.

Both paths can carry wrong facts, but for different reasons:

• Training-data lag. If you moved, rebranded, or changed your phone number since the model’s snapshot, an answer without search enabled can confidently recite the old facts. This isn’t a bug you can report away; it’s how static training data works.
• Stale directory pages. Old listings on directories and aggregators — created years ago, never claimed, never updated — keep publishing your old address or hours. Retrieval-backed answers read those pages and repeat them.
• Mistaken “permanently closed” flags. A closed flag on your Google Business Profile, or on a directory that scraped one, is the single most damaging wrong fact. It doesn’t just misinform — it removes you from recommendations entirely, because no system recommends a business it believes is closed.
• Conflicting data. If three pages say Suite 200 and two say Suite 210, the answer engine picks one. Consistency across sources is not a nice-to-have; it’s what determines which version of you gets repeated.

Note what’s not on this list: hallucination. AI systems do fabricate things — that’s real. But when the wrong fact about your business is specific and repeatable (same wrong address every time), it’s almost certainly sitting on a real page somewhere. That’s good news. Pages can be fixed.

First, find every wrong source

Don’t start correcting until you know the full extent. A scattershot fix leaves the worst page standing.

Ask ChatGPT — with search enabled — and Perplexity, and Google’s AI answers, the direct question: “Tell me about [Business Name] in [City].” Then the variants a customer would use: “Is [Business Name] open?”, “[Business Name] phone number”, “[Business Name] address.” Ask each more than once; answers vary between askings, and a wrong fact that shows up intermittently is still a wrong fact.

Now the important part: read the citations, not just the answer. Retrieval-backed answers cite the pages they drew from. Every citation attached to a wrong fact is a page on your correction list. Click through each one and note exactly what it says about you.

Log it all in a simple spreadsheet: date, system, prompt, the wrong fact verbatim, and the cited source URL. This is the same discipline we use in audits — every response stored raw with a UTC timestamp, quoted verbatim, never paraphrased — because “I think it said something about being closed” is not actionable and “on July 6 it cited yellowpages listing X saying permanently closed” is.

If you haven’t checked what AI says about you at all yet, start with does my business show up in ChatGPT — the same session that measures your visibility will surface wrong facts as a side effect.

The correction workflow, source by source

Work in this order. It’s sequenced by how much each source influences AI answers, based on the pattern we see across the local-business audits we run: presence and consistency on the sources AI answers already cite predicts visibility better than anything else.

1. Google Business Profile. Fix this first, always. It feeds Google’s AI answers directly and is scraped by half the directory ecosystem, so an error here re-infects sources you’ve already cleaned. Verify name, address, phone, hours, categories, and website URL. If you’re flagged permanently closed, use the “mark as open” flow and be prepared to verify ownership — Google may ask for evidence. This one flag outweighs everything else on the list.

2. The cited directories and review sites. Go down your spreadsheet’s source list. Claim each listing if you haven’t, correct the facts, and delete true duplicates (two listings with conflicting data are worse than one wrong one). Prioritize the pages that actually appeared in citations — those are the ones the answer engines demonstrably read in your market. A directory nobody cites can wait.

3. Your own website. Make the correct facts unambiguous and machine-readable: current name, address, phone, and hours in the footer or contact page, matching everywhere character for character, plus LocalBusiness schema markup so crawlers get structured confirmation. If you moved, keep a plain-language line like “As of March 2026, we’re located at…” — retrieval systems quote that kind of sentence directly.

4. Everything else, opportunistically. Local press mentions, chamber-of-commerce pages, social profiles. Lower priority unless one of them showed up in your citations.

Why you can’t edit the model directly

There is no dashboard where you correct what ChatGPT knows. No AI company — OpenAI, Google, Anthropic, Perplexity — offers businesses an interface to edit model outputs about themselves, and nobody can pay for corrections or placement in organic AI answers. Anyone selling “we’ll get it fixed inside ChatGPT” is describing the upstream source work above, or guessing.

The feedback buttons exist and you’re welcome to use them, but they train aggregate behavior; they don’t create a fact-override for your business. The model is downstream of the web. You edit the web.

This is also why the fix is durable when done right: correct the sources and every AI system reading them — including ones that launch next year — inherits the correction.

Honest propagation timelines

Corrections do not show up the next day, and you should distrust anyone who promises otherwise.

• Google Business Profile edits typically reflect in Google’s own surfaces within days; a contested “permanently closed” reversal can take longer if verification is required.
• Directory and review-site corrections propagate as those pages get re-crawled — commonly one to three weeks before retrieval-backed answers (ChatGPT search, Perplexity, Google AI answers) reflect them.
• Training-data errors — wrong answers that appear even with search off — persist until the model is retrained. You can’t schedule that. The practical mitigation is that most local queries now trigger retrieval, so clean live sources usually outvote the stale snapshot.

Plan on days to weeks, verify monthly, and expect the systems to update at different speeds. It’s normal for Perplexity to show the correction while a ChatGPT answer still lags, or vice versa.

What to do this week

1. Run the name-check battery. Ask ChatGPT (search on), Perplexity, and Google: “Tell me about [Business Name] in [City],” “Is [Business Name] open?”, and your address/phone variants. Twice each.
2. Build the wrong-facts spreadsheet. One row per wrong answer: date, system, prompt, wrong fact verbatim, cited source URL.
3. Fix Google Business Profile today. Especially any closed flag. This is the highest-leverage 30 minutes on the list.
4. Correct the cited sources. Claim, fix, and de-duplicate every listing that appeared in a citation next to a wrong fact.
5. Update your own site. Consistent name/address/phone, current hours, LocalBusiness schema, and a plain sentence stating any recent change.
6. Set a re-check date one month out. Same prompts, same systems. Log what changed. If facts are still wrong, the citations will tell you which source you missed. Our monitoring guide covers turning this into a repeatable monthly protocol.

If you’d rather see in one pass what the AI systems currently say about you — and exactly which sources they’re citing in your market — run the free AI visibility scan.