We Asked ChatGPT to Recommend B2B Suppliers. Here's How It Chose

We wanted to see the machine's logic directly, so we did the obvious thing: asked ChatGPT to recommend suppliers across a range of B2B categories, in the words a real buyer would use, then studied what the named companies had in common. You can run the same experiment in your category. Here is the method and what it reveals.

How to run the experiment

Pick a category you know. Write several supplier-search prompts the way a buyer phrases them, across the buying journey: "who are the best suppliers of X for Y," "compare the top providers of X," "which X company is most trusted for Y." Run them in a logged-out ChatGPT session so your history does not skew the result. Record every company named, and where it lands in the list.

Then do the detective work. For each named company, look at its web presence: directory listings, trade-press mentions, how consistent its data is, and what its own pages actually state. Do the same for a few strong companies that did not get named. The contrast is where the lesson lives.

What decided the answer every time?

Across categories, four signals separated the named from the invisible.

A clear identity. Named companies had consistent names, locations, and descriptions everywhere the model could look. The model could say what they do in one confident sentence.

Third-party corroboration. Named companies showed up in directories, trade publications, or forums. Something other than their own website vouched for them.

Quotable specifics. Named companies stated concrete facts on their pages: capabilities, certifications, numbers. The model had something to lift with their name attached.

Consistency tying it together. The named companies did not necessarily do all three brilliantly. They did them consistently, so nothing contradicted anything else. Contradiction is what makes the model hedge, and hedging means omission.

What surprised us

Size did not predict inclusion. In several categories, a focused mid-size company beat a larger, better-funded competitor, because the smaller one was more legible: tight niche, consistent data, specific pages. The larger company had scattered, generic web presence the model could not resolve into a confident recommendation. AI recommendations reward clarity, not budget, and that is genuinely good news for a well-run smaller shop.

What to do with what you find

Your experiment gives you two lists: who ChatGPT names, and what those companies do that you do not. That second list is your work order. If your data is inconsistent, fix it. If you are absent from the third-party sources the named companies appear in, that is your outreach target. If your pages state no specifics, rewrite them.

Then re-run the prompts monthly to watch the answer move. If turning that work order into results is more than your team can take on, it is the kind of program a fractional AI visibility engagement exists to run. The experiment costs you an afternoon. Not running it costs you every buyer who asked the machine and never heard your name.

Frequently asked questions