Why doesn’t ChatGPT recommend my business?

Open ChatGPT. Turn memory off — this matters, because if it remembers who you are it will tell you what you want to hear. Then ask it the question a customer would ask. Not “tell me about my business”, which is a completely different question, but something like I need a roofer in Sunderland, who should I call?

Ask three times, in three separate chats. The answers will differ, and the differences are the point. Write down every business named.

If your own name is not in any of the three, that is not the test going wrong. That is the answer. Run it and see.

The first instinct is to look for the place you sign up. There isn’t one. You cannot submit your business, and there is nobody to contact about it.

You can buy advertising in some of these products — ChatGPT and Perplexity both carry ads now, and Google has always sold them. That is real, and it is labelled. But it sits beside the answer, not inside it. The sentence that names businesses is not for sale, and that is genuinely different from anything that came before. It is also why so much of the advice being sold about this is nonsense.

Broadly two things, and the difference between them explains a great deal.

What the model learned during training. A snapshot of a very large amount of text from the public internet, taken at a point in time. If a model is answering from this alone, it is not looking at your website as you read this — it is drawing on what it read months or years ago. Which is why a business that changed its name last year can still be named under the old one.

What it looks up while answering. Most of these tools can now search the web mid-answer and read pages before replying. When that happens, current information matters and changes can show up relatively quickly.

You can usually tell which one you got: when an assistant goes and looks something up it generally says so and shows the pages it used, and writing those down is part of how a check is run. What you cannot control is which one you get. The same question, asked twice an hour apart, can be answered from training the first time and from a live search the second. That is why honest work here means measuring repeatedly rather than checking once, and why a single screenshot proves almost nothing either way.

These systems sometimes state details that are simply incorrect — an old phone number, a business confused with a similarly named one, an address that has not been right for years, occasionally a business that does not exist at all.

This is not a rare edge case. It is a known characteristic of how the technology works, and anybody telling you otherwise either does not understand it or is hoping you don’t.

It also means being named is not the whole job. Being named correctly is. A recommendation with the wrong phone number in it is worse than no recommendation.

Because the six say what the customer asked about and the eighty do not.

If somebody asks for help with a specific problem in a specific place, the business whose public text contains those exact words is easier to surface than a bigger competitor whose text is generic. Size loses to specificity more often than people expect.

That is good news if you are small. It is the one part of this where being a two-person firm is not a disadvantage.

Free, this week. Fill in every empty field on your Google Business Profile. Put your service area in properly. Write out what you actually do in the words customers use. Ask three recent customers for a review and mention the job when you ask — people write more useful reviews when reminded what happened. Check that your phone number is identical everywhere it appears. None of this costs anything, and most businesses never do it, which is exactly why it works.

Slower. Getting your details corrected on directories you did not create. Building the kind of third-party mentions that give a system corroboration. Rewriting a website built to look good rather than to be understood. And waiting for models to pick changes up at all, which for training-based answers can take a very long time and may never happen for some questions.

We find out what ChatGPT, Google’s AI answers, Gemini, Perplexity and Claude actually say when somebody asks for a business like yours in your area. You see what came back, including the answers that never mention you. Then we go to work on the six sources above, because that is where those answers come from.

Being known, we put in writing. Once the setup is done and the sources are corrected, these systems describe your business correctly — and if they do not, we keep working at no extra charge until they do. That is a promise about our own work, and it is in writing on the pricing page.

Being chosen is the second job, and we work on all of it — your reviews, your published work, what other people write about you, and how you compare to whoever else is in the running. What we will not do is sell you a promise about the inside of somebody else’s model.

There is no switch to buy, no listing to join and no back channel. If a supplier tells you otherwise, ask them to name the system, the exact question, and the date they will have you appearing in it. Then ask what happens if the model changes its mind the following week. Nobody has ever answered that second question, because the honest answer ends the sale.