Ai Search

What to do when ChatGPT has the wrong information about your business

When ChatGPT states the wrong hours, the wrong address, an old phone number, or services you no longer offer, it is not inventing those details out of nothing. It is repeating information it pulled from somewhere else: a stale directory, an outdated version of your site, a third-party profile, or a page that mentions you in passing.

You cannot log into ChatGPT and edit your record, because there is no record to edit. The fix is to find the sources the model actually reads, correct them at the origin, and give it a cleaner, more authoritative version of the facts to pull from next time. Here is how that works and what to change first.

Why ChatGPT gets your details wrong

ChatGPT answers in one of two ways, and the difference decides how you fix the error. When it answers from memory, it reconstructs your details from patterns in its training data, a frozen snapshot taken months before you asked. When it browses the live web, it reads pages at that moment and summarizes what it finds. Most wrong answers trace back to one of these two paths.

The model has no built-in way to know which version of a fact is current. It weights information by how often it appears and how authoritative the source looks. If three old directory listings agree on a phone number you dropped last year, that wrong number can outvote the correct one on your own site. Consistency, not truth, is what the model rewards.

Training data errors versus live retrieval errors

If the mistake is baked into training data, you cannot delete it directly. You can only change what the model finds when it retrieves fresh information, so that new, correct sources start to outweigh the old ones. If the mistake comes from a page the assistant browsed, the fix is faster: correct that page, or get the outdated one removed from the index.

Find where the wrong answer comes from

Before you fix anything, trace the error to its source. Ask ChatGPT the exact question that produces the wrong detail, then ask it to show the sources it used. When it browses, it will often cite URLs, and those pages are your first targets. Repeat the same question in Perplexity and Google AI Overviews, which browse by default and usually reveal their citations plainly.

Then audit the places that feed business data to every assistant:

Correct the sources, then give the model one clean record

Fix each wrong source at its origin, not with a note asking the model to ignore it. Update your Google Business Profile and every directory so the name, address, and phone number match exactly, character for character. Inconsistent formatting reads as two different businesses to a machine, and that ambiguity is often the whole problem.

Then make your own website the single most authoritative statement of the facts. State your name, address, hours, services, and service area in plain text, not locked inside an image or a script. Add structured data so those facts are machine readable and unambiguous, which removes the guesswork a crawler would otherwise do. An llms.txt file can reinforce this by pointing assistants to your canonical pages and stating key facts in a form crawlers read easily. Each line in that file is a plain declaration of a fact or a link to the page that proves it.

Why one correction is not enough

Changing a source does not update ChatGPT instantly. Browsing assistants can reflect fixes within days, because they read your corrected pages live. Memory-based answers can lag much longer, until the model retrains or until live retrieval overrides the old pattern. You are not editing the chatbot. You are changing the evidence it reads, then waiting for the weight of correct, consistent sources to win.

That is why this is maintenance, not a one-time task. Keep every listing consistent, keep your site the cleanest source, and re-check what the assistants say on a schedule so a new wrong mention does not quietly take hold.

How Ninetyfifth Marketing fixes it

We handle this as part of SEO and AI search optimization, which we perform only on websites we build. The work is concrete: we audit what the major assistants currently say about you, trace each wrong detail to its source, correct the listings and third-party records, and rebuild your site as the authoritative source with clean structure, structured data, and an llms.txt file. Then we monitor what the assistants say over time.

We will not promise to force ChatGPT to say a specific thing, because no honest practitioner can control a model's output. What we control is the evidence, and we make yours the clearest and most consistent version available. We take one SEO client per industry per city, on a published retainer with a 90 day initial term and the custom website included. See /services/seo/ for the method and /pricing/seo/ for the model.