Paying ChatGPT Does Not Make It Recommend You

Now that ads exist inside ChatGPT, a specific misunderstanding has started arriving in enquiries, usually phrased as a budget question. If we spend enough, does it start recommending us?

No. And the reason is worth understanding, because the same confusion has cost people money in every channel this has ever come up in.

Two systems sharing a screen

When somebody asks ChatGPT to recommend a business, two different things can put your name on that screen, and they have almost nothing to do with each other.

The answer is generated from what the model has read and, when it browses, from what it can find and cite in the moment. Your business appears there because the facts about you are published somewhere it can reach, stated consistently, and phrased so a clean sentence can be lifted out. Nobody sells access to that. There is no rate card.

The ad is a sponsored placement below the answer, bought through OpenAI's Ads Manager, matched to the context of the conversation. You can turn it on this afternoon with a card and a budget. We've written separately about what that product actually is.

One is earned and can't be bought. The other is bought and can't be earned. They render about two centimetres apart, which is the whole problem.

Why the confusion is expensive

Two failure modes, and we've seen both.

The first is spending the visibility budget on ads and assuming the recommendation follows. It doesn't. You get a month of placements, you stop paying, and you are exactly as invisible in the answer as you were before. Nothing compounded.

The second is quieter and worse: believing the ad is the recommendation, and reporting it that way internally. A sponsored card and a sentence saying "you might try GrowMint" carry completely different weight with a reader. People know what an ad is. The whole reason the recommendation is valuable is that it isn't one.

What actually earns the mention

The unglamorous answer is that it's mostly the same work that has always made a business legible to a machine, done properly.

Your facts have to be published where they can be read, which means in text on a page rather than assembled by JavaScript after the fact. They have to agree with each other everywhere they appear, because a business whose address or service list contradicts itself across three sources is one an assistant will quietly skip rather than get wrong. They need structured data doing the declaring so nothing important is left to inference. And the writing itself has to answer questions directly, because a paragraph that gets to the point in its first sentence is the one that gets quoted.

None of that is exotic and none of it is fast. It's also the reason it holds: the same work makes you legible to Claude, to Perplexity, to Copilot and to Google's AI Overviews, none of which sell you anything. What each of them is actually weighing is a longer subject, but the inputs you control are the same across all of them.

The order that works

For most local businesses the honest sequence is not the one that sounds most exciting.

1. Find out where you stand. Ask ChatGPT the question your customer would ask, in your city, and write down who it names. If that's a competitor, that's your baseline and it costs nothing to establish.
2. Fix what's readable. Facts on the page, consistent everywhere, structured, quotable. This is the part that keeps paying after you stop working on it, and it's what our AI SEO and analysis service is for.
3. Then test the ad, with a capped budget and a way to identify the enquiries it produces.

Doing step three first isn't a disaster. It's just paying rent on attention while the thing you could own sits untouched.

Telling the two apart in your own numbers

This matters more than it sounds, because if you run both at once and can't separate them, you'll credit whichever one you were more excited about.

Keep them distinguishable on arrival. Ads go to a landing page built for the ad, with its own URL and its own tracking, so anything hitting it came from spend. Earned mentions arrive on your ordinary pages and, more often than not, arrive with no referrer at all, which is why chatbot traffic mostly lands in Direct and why a rising Direct line landing on deep service pages is a signal rather than noise.

Then read them separately every month. Citation share is a slow number that should trend up and stay up. Cost per enquiry is a fast number that stops the day you stop paying. Averaging them together produces a figure that describes nothing.

Both halves are worth having, and we run them as one programme precisely because the reporting only makes sense side by side. What you can't do is buy the half that matters most.