“We want to show up in ChatGPT” is one of the most repeated briefs of the year. Translated, the client is asking for something far bigger than an indexing job: for a system that synthesizes thousands of sources to conclude that their brand is a good answer. Not just to find it, but to choose it — and to have something good to say about it.

That request almost always lands on the digital agency, and it’s almost always met with the same 3-to-6-month package: technical audit, “AI-optimized” content, schema markup, an llms.txt, a visibility dashboard. It gets billed, delivered, reported. And the needle barely moves.

In our honest opinion, GEO is not a performance service on a quarterly cycle. It’s medium- and long-term brand-positioning work, and most of what makes it work doesn’t happen on the client’s site — nor, often, is it the digital agency that executes it.

It’s not one problem; it’s three

Showing up in a generative answer from ChatGPT, Claude or Gemini isn’t a single thing. It’s three chained problems, with completely different economics.

  1. That the LLM finds you. The model has to be able to crawl and retrieve your site. This is technical, contained and cheap.
  2. That it picks you. Out of twenty possible candidates, the system builds a shortlist. Here the winner isn’t whoever wrote the best homepage: it depends heavily on the next point.
  3. That it has something to say in your favor. Once chosen, the model describes your brand with the attributes it finds repeated across different sources.

Almost all of the short-term, simplistic GEO offering focuses on problem 1 and sells it as if it solved problem 3. The reason is obvious: problem 1 is the only one a digital agency can fully control, deliver in four weeks and show in a report. The other two depend on third parties who don’t sign your contract.

The evidence isn’t on your site

Three figures worth keeping handy when someone promises AI visibility by optimizing pages.

1. Classic ranking explains far less than assumed

Across 15,000 long-tail queries — between search engines and four assistants — only 12% of the URLs cited by AI were in Google’s top 10, and 80% didn’t rank anywhere for that query. (Ahrefs)

Assistants don’t pick one page for a query: they fan the question out into many variants and merge results. You’re not competing for a position; you’re competing to appear consistently across dozens of different retrievals.

2. Being cited and being mentioned are not the same

We analyzed 115 prompts across 14 countries and four engines, and found that 62% of citations don’t translate into a brand mention in the text of the answer. (Semrush)

Your site feeds the answer and the user never reads your name. All that content effort can end up as free raw material for the model to recommend someone else.

3. The brand story is built outside

Across 126 million prompts between January and April 2026, ChatGPT cites about 15 sources per answer, leaning heavily on Reddit and Wikipedia; there are cases where visibility holds up thanks to the overlap between reviews, retailers and independent publications — not the brand’s own site. (Semrush AI Visibility Index)

This is what almost no one says in a proposal: GEO is not mainly about optimizing content. It’s about manufacturing public, corroborable, distributed evidence about an entity. And the evidence about you, by definition, isn’t written by you.

In that sense it’s closer to a press or PR agency than to a digital agency that builds creative and code — though, obviously, we’re simplifying.

Why it isn’t just SEO under a new name

The easy version says: on Google you optimize the site; on the models you build reputation. That’s not true.

SEO was never mainly on-page either. Links were always third-party votes, and domain authority was always reputation measured indirectly. Anyone who thought they ranked by writing good title tags wasn’t ranking.

The real difference is another one, and it’s far more interesting.

  • The currency changed. Google counted links: a structural, anonymous, buyable signal. A model reads sentences. A link to you isn’t enough: it needs someone to state, in text, what you’re good at. A sentence claiming you’re great at agro-industrial marketing is not just a backlink.
  • The size of the prize changed. Google gave ten positions per query and you fought for one. A model returns three or five names. Reputation stops being a gradual tie and becomes winner-take-most.
  • Who arbitrates changed. The search engine showed you sources and the user decided. Now the model reads the sources, compares them and issues a verdict. The contradiction between what you say and what everyone else says is no longer resolved by the reader: it’s resolved by the system — and almost always in favor of the source you don’t own.

So yes: GEO is reputation work. But not because SEO was site work — rather because SEO let you simulate reputation with structural signals, and now you have to have it written down.

What others say about you is the key

Let the models’ bots crawl your site; check that the CDN isn’t blocking them; take critical content out from behind JavaScript; add structured data and write pages that literally say what you do, for whom and where. That’s fast, and not doing it invalidates everything else.

But the key isn’t really there, nor even in “optimizing” your content — it’s fundamentally about being a brand that’s recognized in your industry, linked in your core media, on portals and by the main recommenders. There are, let’s say, two layers.

Layer 1: who you are

It’s the spec sheet: name, category, city, size, founders, services, clients. Data, not opinions. No one is evaluating you — it’s a record. And you can fill in almost all of it yourself: the LinkedIn company page, Wikipedia, your website, the chamber-of-commerce listing, the industry directory, the sector’s ranking pages.

Layer 2: how good you are

It’s someone else’s judgment: a party that isn’t you making an evaluative claim about your work, in a place you can’t edit. An article, an award, a ranking, a review, a case study published by your client, review platforms (Trustpilot, G2, Glassdoor), podcasts. Much of Layer 2 a brand or advertising agency can’t guarantee: it’s not usually a core part of its scope, and it was always the expertise of press and PR agencies and of the client’s own corporate-communications teams.

The gray market that has already begun

There’s a predictable consequence to all of this. If Layer 2 is what decides, and the entry ticket is no longer bought in the search engine, it will be bought at the source: premium placement in directories, sponsored rankings, paid lists in niche media, managed reviews. The model reads all of it and doesn’t distinguish editorial content from paid content. It would be a faster way to build reputation — just not as organic or transparent.

Takeaways

Drop the word GEO and this is what’s left: is there public, verifiable, repeated evidence — in sources you don’t control — that you’re good at what you say you’re good at?

It’s the same question a purchasing committee asked in 1998 and a search engine asked in 2010. The evaluator changed, not the criterion. What really changed is that now there’s a system reading every source at once, comparing them, and mercilessly penalizing brands that only speak well of themselves on their own domain.

That’s why GEO isn’t a new discipline to be invented. It’s reputation, press, content generation, social presence, positioning your company’s speakers and figures, and brand consistency — with a two-week technical layer on top and a much higher bar of proof.

The question you should ask yourself as a client isn’t “how much does a GEO project cost?”, but “is my company ready to take on a medium- and long-term brand- and corporate-positioning project, with real effects on how LLMs index and recommend our brand?” A three-month package is unlikely to give you that.

Sources