ChatGPT Still Describes the Old Version of Your Business

You added a service in the spring. You dropped the one that never paid. The website says so, the Google Business Profile says so, and a customer still shows up asking for the thing you stopped doing two years ago, because an AI assistant told them you do it.

Nothing broke. The assistant is describing a business that used to exist.

How an AI assistant ends up describing last year's business

A language model learns from text collected up to a cutoff date. Anything you changed after that is invisible unless the assistant also searches the live web for that particular answer, and plenty of answers never trigger a search.

Semrush's write-up adds two causes beyond the cutoff. Your own footprint may disagree with itself: the homepage says one thing, an old directory listing says another, and the model picks a version. And years of old descriptions can simply outnumber a few months of new ones. In the article's words, a new positioning "can be accurate and well documented and still be outnumbered."

Semrush's named examples are Hotjar, Blue Apron, Duolingo and Peep Laja, where ChatGPT kept describing an earlier version. Those are big brands with large footprints. A small business has less text out there, which cuts both ways: less old material to outweigh you, and fewer places for the new version to show up.

Where small businesses get hit hardest

The lag matters most after the changes that small businesses make constantly:

• A new service line, or an old one retired.
• A rename, a merger, or a change of ownership.
• A move to a new service area.
• New hours, a new phone number, a new booking link.
• A shift from serving everyone to serving one kind of customer.

Each one leaves a trail of old text: a directory listing nobody logs into, a blog post from 2023 that still ranks, a "services" page that quietly kept the old list because editing it was never anyone's job. Hours and addresses are the easiest to check. Our post on AI giving the wrong business hours covers the data on that. Positioning is harder, because nobody has a field for it.

Find out which version the assistant believes

Before fixing anything, read what it says. The routine is the same one we describe in what AI says about your business, pointed at the thing you changed:

1. Ask two or three assistants, in plain language, what your business does and who it serves. Use the words a customer would use, not your own.
2. Ask about the specific thing you changed: "Does [your business] offer [old service]?" and "What does [your business] offer?"
3. Write down every answer that is stale or wrong, and note which assistant said it. Answers vary between assistants and between sessions, so one run is a sample, not a verdict.
4. Look for the source. If the wrong detail matches a particular directory or old page, that page is your first fix.

This takes about twenty minutes, and it tells you whether you have a lag problem or a different one.

The fixes, in the order they pay off

Make your own site say one thing. Read the homepage, About page, services page, and page titles in one sitting. Any page that still describes the old offer either gets updated or, if it has no reason to exist, redirected. This is the cheapest fix and the one under your control.

Say the change in plain words, once, in a findable place. A short, dated line on your About or services page ("As of this year we focus on X and no longer offer Y") gives a model something unambiguous to quote. Write it for a stranger who has never heard of you.

Update every listing that names you. Google Business Profile first, since it is the most-referenced source for local businesses, then the directories and social profiles with your description in them. Our piece on NAP consistency and AI visibility walks through why mismatched listings make the model's job harder.

Structure the pages so they quote cleanly. Short sections under headings that match how people ask the question ("Do you offer X?") are easier for an assistant to lift than a paragraph of brand language.

Recheck on a schedule. Run the same questions a few weeks after the changes go live, and again every quarter. Models update on their own schedule, so the new version may take a while to appear. This is part of the broader job described in the second search: your business gets described in places you do not control, and someone has to look.

Fixes that waste your time

• Re-submitting to every directory you can find. If the old description lives on twenty sites, fixing three of them does not move the needle. Fix the ones the wrong answer traces to.
• Stuffing the old name or service into hidden text "to be safe." It does not help a person or a model, and it puts your own site in conflict with itself.
• Paying for a one-time "AI visibility" fix and calling it done. A model's picture of you drifts as the web changes. A one-off fix goes stale like any other.
• Assuming the assistant will find the update because Google did. Search indexing and model knowledge are different systems on different clocks.

FAQ

What is the LLM positioning lag?

It is Semrush's name for the gap between how a business currently presents itself and how large language models still describe it. The model reflects older information until newer, consistent information outweighs it or the assistant looks the answer up live.

How long does it take for ChatGPT to learn that my business changed?

Nobody outside the model providers can give a reliable number, and Semrush does not give one. Assistants that search the web for an answer can pick up changes faster than ones answering from training data alone. Plan on rechecking after a few weeks and again each quarter.

Does updating my Google Business Profile fix it?

It helps, and it is the first listing to fix for a local business. It does not fix descriptions that live on other directories, old blog posts, or your own services page, so treat it as one fix among several.

Should I publish a press release about a small change?

Semrush recommends announcing repositioning publicly, including through a wire service for larger brands. For a small business, a clear dated statement on your own site and updated listings is the proportionate version. A paid wire release for a new service line is rarely worth it.

Can I ask an AI company to correct what it says about me?

Some assistants offer feedback buttons on answers, and it costs nothing to use them. There is no guaranteed correction process, so the durable fix is making the accurate version easy to find and consistent.

How do I know the assistant is wrong and not just inconsistent?

Ask the same question in a fresh chat on a couple of assistants. If the stale detail shows up repeatedly and matches an old page or listing, it is likely sourced from there. If it changes every time, you are looking at variation and the better fix is clearer, more consistent pages.

Is this only a problem for brands that rebranded?

No. Any change that leaves old text behind can cause it: a retired service, a move, new ownership, or a narrower focus. Rebranding is the dramatic case; a quietly outdated services page is the common one.

If keeping every page, listing and AI answer in step with what your business does today sounds like a job you will never get to, that is the kind of ongoing watching our get found on Google and AI work covers.