Every week someone forwards us the same email: the classic GEO agency ChatGPT pitch. An agency promises a garage in Sants, or a gestoría (accountancy and admin firm) in Poblenou, "guaranteed first place on ChatGPT" for a monthly fee. The email design is flawless and the tone supremely confident.

And it's a lie. No GEO agency — marketing for showing up in AI, as they call it now — can guarantee you number one on ChatGPT. The reason is uncomfortable for a good part of the industry: nobody controls the model. Not them, not us. We'd rather start there, even if a truth sells worse than a nice round promise.

Why "guaranteed number one" is impossible

ChatGPT doesn't work like Google. On Google there's a list of positions. In a language model, there isn't.

It's a probabilistic system: the same question, asked twice, can give you two different answers. Two customers typing "best hair salon near me" on two different phones read different names. The answer changes depending on how the question is worded, the language, the time and the history of whoever is asking.

There's no stable "position 1" anyone can sell you, because there's no single ranking to conquer. Whoever guarantees you a fixed position is promising something the system, by design, cannot deliver. It's not that the agency is clumsy: the promise is impossible.

The only thing you can really buy is ads

Some assistants have started showing ad formats, labelled and kept separate from the body of the answer. That you can pay for — to whoever runs the assistant, not to an agency — and it's still an ad, not a recommendation.

When the AI writes "I recommend this garage" inside its answer, that slot isn't for sale. It's exactly what no agency can reserve for you.

So the rule is simple: if they promise you'll "come out first" and they're not talking about buying a space labelled as advertising, they're selling you smoke.

The catalogue of lies we've already seen

After months looking at this for Barcelona businesses, the lines repeat themselves:

"We have a special agreement with OpenAI." There's no programme for placing businesses inside the answers. There isn't one.
"One tweak and you're set for good." Models get retrained and answers move on their own. What works today falls over in a few months.
"Results in 7 days." The sources AI relies on take weeks or months to reflect a change. Seven days isn't even enough to get started.
"SEO is dead; we do AI SEO with a trick." There's no trick. There's slow work on the right signals, and whoever sells you a shortcut is selling you their hurry, not your result.
"We'll get you into all three AIs with the same action." They share almost no sources: out of 680 million citations analysed, only 11% of the domains cited by ChatGPT were also cited by Perplexity. What works in one can be irrelevant in another. We explain it in ChatGPT, Perplexity and Gemini don't recommend you the same way.

Why lying is so cheap here

The underlying problem is that almost nobody can check.

Most small business owners don't know how to verify whether they really show up, and the AI itself proves the liars right by accident: the owner asks "would you recommend me?" from their own phone, with their history and previous conversations, and sometimes their business comes up. It looks like it works.

It means nothing. A different customer, in Gràcia or in Sant Antoni, sees something else. Asking about yourself as yourself is the least reliable test there is, and it's exactly the one the agency will show you as a "result".

Four questions that take anyone apart

This doesn't require you to hire anyone. Next time they call you, ask this and listen:

1. "Exactly which questions will you measure me with?"

They have to be the ones a customer who doesn't know you would write — a service and a neighbourhood — not your business name. If they measure you by your own name, they're measuring what you already had.

2. "Which assistants, and how often?"

If the answer is just "ChatGPT", that's half the picture. And if it's "once at the start", it isn't a measurement: it's a screenshot.

3. "Show me the same report from last month, for another client."

What can be checked is the only thing that counts. If all they have is a screenshot taken on their own phone, you know.

4. "What happens if it doesn't work?"

Nobody honest promises you a position. They can tell you what they'll do, how often, and how you'll know whether it worked. If the answer to this question is a guarantee, go back to the first line of this article.

So is it worth the effort?

Yes, and it's worth separating that from the false promise.

The behaviour has already changed. In Spain, use of generative AI went from 51% to 69% of the population in one year, and frequent use — daily or several times a week — rose from 27.8% to 39.1%, according to the second Observatorio Anual IAÓN (an annual Spanish AI survey). And for finding local businesses specifically, the share of consumers using AI went from 6% to 45% in one year, although that figure comes from a survey of 1,002 people in the United States, and it's worth saying so.

What none of those numbers tells you is when you're going to show up. Nobody knows. What you can do is the boring part: have a website AI can read and quote, consistent details everywhere, living reviews, and measure again every so often because the answers change on their own.

Doing it yourself: the honest sum

Free, and done once. Asking about your business in the three assistants with a customer's questions, noting down who shows up, claiming your Google profile and making your name consistent everywhere. One afternoon. Do it even if you never hire anyone for anything.

Slow. The models picking you up: weeks or months, depending on when your website gets crawled. There's no button. There's no date.

The part that never ends. Asking again. The answers move without you touching anything, so a check in September says nothing about October. It's twenty minutes every few weeks, forever, and it's the first thing to be dropped.

What we do, without the hot air

We don't control the model and we won't promise you a position. What we do is the rest: your wiweb page is born written so that an AI can quote it — services with prices, questions answered, details consistent with your Google profile — and with the robots.txt deliberately open to ChatGPT, Perplexity and Google, instead of blocked by a rule inherited from a template.

And the measuring-again part stops being yours. On Grow (€9/month) we check every month whether ChatGPT and Perplexity mention you, for which questions, and which competitors show up where you don't. On Pro (€24/month), every week and across all four assistants. Always the same questions, so that the comparison between months means something.

Creating and publishing your page is free, no card needed: create it in two minutes. Exactly what we check is in SEO and AI, what each plan includes is on pricing, and the source of every number in this article is on sources.

What you won't find in any of our plans is a guaranteed number one. If you find it somewhere else, you know what it is.

FAQ

So is there nothing you can do to show up on ChatGPT?

You can influence it; you can't guarantee it. You work on what AI can read and verify about you, and you measure the result over time. What doesn't exist is a button that puts you first.

How do I know if my agency is lying to me?

Ask them the four questions above. If all they show you is a screenshot taken on their own phone, or they promise you a specific position, it's smoke. What can be checked is the only thing that counts.

And the assistants' ads, are they any use to me?

They're something else: a sponsored, labelled space, separate from the recommendation. They may fit in some cases, but they don't replace the AI naming you on your own merit, which is what really moves someone to call you.

How often does all this change?

Constantly. Models get updated, sources change and the answers with them. That's why "a fix for good" is false: this is more like looking after a reputation than installing something once.

GEO agency ChatGPT promises: "guaranteed number one" and other AI marketing lies