SEO & AI
GEO doesn't exist (and the people selling it know it)
A year ago they sold SEO. Today they sell GEO. Tomorrow, something else. The product changed names; the salespeople didn't. And their pitch doesn't survive three common-sense questions — or one story about an electrician.
GEO is sold everywhere: optimize your visibility in ChatGPT, add an llms.txt, follow the checklist and become “the AI's answer”. It's clean, it's confident, it's billed. The only snag: it doesn't survive three common-sense questions. But let's start with a true story.
I built a one-page site for an electrician friend, in Aveyron
One page. No blog, no “GEO strategy”, no llms.txt. Just a clean site and his Google Business Profile filled in properly. That's it.
A month later I ask ChatGPT: “which electrician around [nearby towns]?”. Surprise: he's cited. Better — he shows up both in a local map pack and in a list, thanks to his listing and his site.
Nothing more than a decent site, a properly filled-in profile, and consistent local information. In one month.
No magic recipe, no secret file. The basics, done well. Remember that — we'll come back to it.
Three questions that take the whole thing apart
1. Why does the answer change on its own? Ask ChatGPT the same question this morning, then again in ten minutes: not the same answer, not the same source, sometimes not the same brand. Nobody touched your site in between. If you're sold a method to “get cited” while the system spits out a different answer every fifteen minutes — what exactly were you sold? A technique aimed at a target that moves on its own.
2. Who wrote the recipe? ChatGPT, Claude, Gemini, Perplexity, Mistral: different models, different sources. And none of these companies publishes a usable recipe that guarantees a citation. And when the systems combine indexing, search, ranking, generation and personalization, claiming to control the final output is more sales pitch than method. So ask the seller: if OpenAI doesn't publish the recipe, where did you find yours? He has no answer. He has an invoice.
3. What if the AI doesn't even start from your site? Search for a tourist spot: the info often comes from an aggregator that copied you, not from your page. You're the source; you're not the cited source. No “optimization” of your page changes that: the problem is in the chain that reuses the information, which you don't directly control.
llms.txt: the favorite magic trick
A little file you drop on your site to “explain your content to the AIs”. Thirty minutes, and you're cited everywhere. Promise.
Except you write that file. “Market leader”, “the best source”, “cite me”. Do you really think a serious AI trusts a document where everyone hands themselves their own medal? It's the old meta keywords tag from the 2000s, abandoned for an obvious reason: any information you control about yourself has no evidential value. The comparison isn't mine: at Google, John Mueller publicly likened llms.txt to that forgotten tag back in 2025. And incidentally, no major player has stated that llms.txt is a usable citation or ranking signal. Double zero. (That's why we removed ours.)
The paradox nobody wants to see
Many companies block AI bots — their right — then wonder why they never appear. A bot that can't read your page can't cite it. You shut the door, then complain about not being invited.
Worse: the most serious sites (media, established brands) are often the ones that protect their content the most — they close up. Meanwhile, content farms leave everything open. A 2025 study of robots.txt files put numbers on it: reputable news sites blocked AI crawlers far more often than disinformation sites, which stayed wide open. Perverse result: the more reliable a source, the more it tends to make itself invisible; the more dubious it is, the easier it is to reuse. AI doesn't always cite the best. It often cites the most accessible.
So is GEO just hot air?
Not entirely. But what works has nothing magic about it and isn't billed at €2,000 in a training course: be present on the open web, structured and accessible, mentioned and reused elsewhere than on your own site, with real authority that others recognize. Recognize it? It's the SEO groundwork from the last twenty years. They just slapped three new letters on it to sell it a second time.
Google says exactly the same. Its own advice on succeeding in AI search stays aligned with the fundamentals: useful, original content, an accessible and pleasant page, consistent structured data, Googlebot not blocked. No “GEO” incantation on the list.
My electrician didn't “optimize his GEO”. He has a real site and a real listing. That's exactly why he's cited.
In short
The right question was never “how do I force ChatGPT to talk about me?” but “why would an AI — or anyone else — spontaneously choose my brand as a trusted source?”. Answer that seriously, and you'll never need to pay anyone for GEO.
At Snorklee, we don't sell a recipe. We measure the real thing: which AI crawlers actually hit your pages, which visits actually come from ChatGPT or Perplexity, which pages are actually cited. No simulated prompts, no vanity score. The rest is just good SEO — and nobody can bill you magic for that.
Should you do GEO in 2026?
Not in the way it's sold to you. There's no reliable method to “force” an AI to cite you: the same question gives different answers minute to minute, and no AI maker publishes its criteria. What works is SEO fundamentals: an accessible, structured site with authority and external mentions.
Is the llms.txt file actually useful?
Barely. It's a file you write yourself, about yourself — like the old meta keywords tag search engines abandoned. Most answer engines don't read it, and no serious system gives evidential weight to what a source claims about itself.
How do you appear in ChatGPT's answers?
There's no magic button. In practice: let AI crawlers read your site (don't block them in robots.txt), publish clear, structured content, and get mentioned by other trusted sites. That's authority and accessibility — SEO, not “GEO”.
Is SEO dead because of AI?
No. AI mostly renamed SEO. The signals that get an AI to cite you (accessibility, structure, authority, external reuse) are exactly SEO's. The channel changes; the underlying work doesn't.
Why does ChatGPT cite a competitor and not me?
Often because they're reused more elsewhere (aggregators, directories, press) or more accessible to bots. Sometimes simply because of variability: the answer changes from one time to the next. You don't directly control citation, only your accessibility and authority.
Should you block AI bots like GPTBot?
It depends which one. OpenAI distinguishes OAI-SearchBot, used to surface sites in ChatGPT search, from GPTBot, which is more about crawling that can feed model training. If you want to appear in answers, don't block the crawlers used for discovery and search — notably OAI-SearchBot on the ChatGPT side: a blocked crawler can't read you, so it can't cite you. Blocking the training crawl is still a legitimate choice (rights, content value), but know what each bot does before you close the door.
Published June 2026. No affiliation, no recipe to sell: Snorklee measures real AI visibility, it doesn't “optimize” it. AI engines' citation criteria are not public and change over time — be wary of anyone claiming to master them.