How Does ChatGPT Decide Which Businesses to Recommend?

Suppose someone asks:

Who are some good management consultants?

That’s very broad. Now change it to:

Who are some management consultants who work with small healthcare practices?

We’ve already narrowed the field considerably. And if the question becomes:

Who could help an owner-led healthcare practice improve its management systems without bringing in a large consultancy?

we’re asking for something much more specific again.

The job has become much more specific: find businesses that fit this person, this type of organisation and this particular problem. A universally agreed list of the best management consultants, even if such a wonderfully cooperative thing existed, wouldn’t answer it.

There’s another layer here because ChatGPT Search doesn’t always pass the exact words you typed straight to a search provider. OpenAI says it usually rewrites the question into one or more targeted searches, and it can run further searches after seeing the first results. If memory is enabled, relevant saved memories can affect those rewritten searches too.[1]

So when the recommendations change after you alter the question slightly, you haven’t necessarily nudged your business up or down a ranking. You’ve changed what ChatGPT is looking for and, in some cases, the searches it runs to find the information.

People often assume ChatGPT sends the question to Google and works from whatever comes back. Its search process is broader than that.

ChatGPT can answer from information already available to the model, and it can search the web when current information would help. When it searches, OpenAI says it can rewrite the query, send searches to third-party search providers and use the sources it finds to build the answer.[1]

Location can come into it as well. ChatGPT can use an approximate location based on the user’s IP address, and users can choose to share a more precise device location.[1]

That means recommending an occupational therapist and recommending an occupational therapist near me who works with workplace rehabilitation are doing quite different jobs. The second request includes a location and a specialist need, either of which can change which businesses fit.

And ChatGPT isn’t restricted to what your own website says about you. When it searches, public web pages, directories, professional profiles, articles and other sources can all form part of the information it finds.

That information isn’t guaranteed to be right. OpenAI warns that search results and citations can be incomplete, outdated or incorrect.[1] Knowing where the information came from is only part of the job; you also need to check whether it describes the business accurately enough to connect it with the request.

Imagine 2 management consultants. One works mainly with large corporate teams on restructuring and leadership. The other works with owner-led healthcare practices that have grown quickly and now need better management systems.

If somebody asks for a highly experienced management consultant, either could be a reasonable suggestion.

But ask for someone who helps small healthcare practices put better management systems in place, and the second consultant fits far more of what the person has actually asked for.

OpenAI hasn’t published the internal weighting used to evaluate those businesses, so we can’t tell you that industry counts for so much, location adds a few more points and client size finishes the calculation. What we can see is that the question has changed the job the recommended business needs to do.

Industry, location, business size, specialist need, budget, type of client and the problem being solved can all narrow the field. The earlier conversation can affect it too, and relevant saved memories can influence the searches ChatGPT runs.[1]

This is why “Does my business appear in ChatGPT?” is much harder to answer than it sounds.

Does it appear when somebody searches for the business by name? What happens when they search for the service, add their industry, narrow the location or describe the exact problem they need solved?

Your business doesn’t need to appear for every possible variation. The public information needs to describe it accurately enough for ChatGPT to connect it with the kinds of requests it genuinely fits.

This is usually the point where somebody produces a neat list called The ChatGPT Ranking Factors, which would be tremendously convenient if OpenAI had published one. But it hasn’t, not yet anyway.

There’s no known weighting that says service descriptions count for X, third-party mentions count for Y, and schema adds another 6 points. A better question is whether the information available about the business gives ChatGPT enough to make the right connections.

If your website says you provide “strategic solutions that help organisations reach their potential”, a determined visitor can click around until they eventually work out what that means. ChatGPT has far less to connect with when the request names a specific service.

If you provide management consulting to small healthcare practices, say that accurately. If you specialise in operational systems rather than HR consulting, make that distinction. And if a service is only available to a particular kind of client, that belongs in the description too.

This is where I think accuracy needs at least as much attention as clarity. Clear writing makes the information easier to read. Accurate writing tells ChatGPT what the business does, where it fits and where it doesn’t.

A very broad description can make a business sound as though it could help almost anyone. That becomes less convincing as the request gets more specific.

If you regularly work with small healthcare practices, professional associations, manufacturers or owner-led businesses, saying so gives ChatGPT something concrete to connect with when those details appear in a search.

That doesn’t mean adding every customer type you could conceivably work with. Before long, the description starts to sound like the business equivalent of a restaurant serving pizza, sushi and Sunday lunch. Describe the work you actually do and the situations in which you genuinely do it.

Businesses change all the time. Services are renamed, offers disappear, prices change, people leave, and addresses move. Old pages, meanwhile, can stay online long after everybody involved has forgotten about them.

Directories can be even more determined.

If the current website says one thing and an old directory profile says another, ChatGPT can find itself with a mixed picture. And when the disagreement affects whether the business fits the question, such as the service it provides, where it works or the customers it serves, that inconsistency gives you something specific to correct.

ChatGPT isn’t limited to what you’ve written about yourself on your own website. Other public sources can add information, support what you’re saying or contradict it completely.

• An industry article might connect your company with a particular specialism.

• . A professional profile could describe your work more accurately than an old website page.

• A directory could still list a service you stopped offering 4 years ago.

• And a third-party mention could connect the business with a category or type of work that isn’t explained particularly well anywhere else.

None of these is a published ChatGPT ranking factor. They’re places where the public picture of your business can become more accurate and specific, or considerably more muddled.

Once you let go of the idea of one fixed ranking, the variation is easier to explain.

• The wording can change.

• The conversation before the question can change.

• The user’s location can change.

• ChatGPT can generate different searches, and the information available online can change.

• The model and product configuration being tested can make a difference too.

If you want to know what customers see in ChatGPT, test inside ChatGPT. A tracker can still help you follow a consistent set of prompts, but you need to know what it is actually measuring before treating its score as a picture of the consumer product.

And one search still doesn’t establish a permanent position. If your business appears once, you haven’t suddenly “ranked number 3 in ChatGPT”. If it disappears once, we still don’t know whether there’s a real visibility problem. What keeps happening tells you far more.

ChatGPT could know exactly who your business is when you ask for it by name but leave it out when you ask for the service. You could appear for broad searches and disappear when the customer type becomes more specific. Or the same competitor could keep turning up across several versions of the same need. That gives you a pattern to investigate instead of one screenshot and a rapidly expanding collection of theories.

You can’t control whether ChatGPT recommends you, but you can improve the information it has available to work with.

That includes describing your services accurately, being specific about who they’re for, correcting old information and checking that important public pages can be accessed. It also means looking beyond the current website because that isn’t always the only source contributing to the public picture of the business.

OpenAI says OAI-SearchBot is used to surface websites in ChatGPT’s search features. A site that blocks it won’t appear in ChatGPT search answers, although it can still appear as a navigational link.[2] Allowing the bot makes the site eligible for search; it doesn’t guarantee that a page will be used, cited or that the business will be recommended.

And this is where a tidy checklist of “ChatGPT ranking factors” starts causing trouble. You can work through 14 recommended fixes and still have no idea whether any of them explains why your business was missing in the first place.

The question I’d rather ask is:

What would ChatGPT need to know for this business to fit this specific request, and does the public information give it accurate enough evidence to make that connection?

If your business keeps appearing for some questions and disappearing for others, compare those differences. Look at the questions where you appear, the ones where you don’t, which businesses are being recommended instead and what public information supports each apparent match.

That gives you a much better starting point than trying to reverse-engineer a ranking formula nobody has published.