This week, Peter called me. Owner of a dental clinic in a city of 80,000 inhabitants, website made by a local agency in 2019, super happy with his site until he read an article about GEO. He called to ask: “Daniel, can you make my site appear in ChatGPT when people ask for dentists in my city?”.
I told him no. Not literally (I can do things that help a lot), but I can’t guarantee that his site will show up. And I started a conversation that lasted 45 minutes and that, when he hung up, made me think that there was material for an honest post. What I’m going to tell you here is the free transcription of that conversation, with his permission (anonymised), because Peter’s question is the same one 4-5 people ask me per month. And because the honest answer isn’t the one they want to hear.
What Peter asked me first
“Daniel, I read that ChatGPT has 200 million users and that more and more people ask it ‘recommend a dentist in my city’. If I don’t appear when someone asks that, I’m losing clients, right?”.
I told him probably yes. “Discovery” searches (the ones people make to find something new, not to buy directly) are migrating fast to ChatGPT, Perplexity, Claude and Gemini. The data I gave him: “zero-click searches” (where the user doesn’t click on any Google result, because the answer is in Google itself or in an AI) have gone from 56% to 65% between 2024 and 2025. That means 2 out of 3 searches no longer generate traffic to any website. If your site doesn’t appear in the AI’s answer, you don’t exist for that user.
Peter got serious. “OK, what can I do to appear?”.
What I told him I CAN do
I explained that there are 7 concrete things that ARE under my control and that DO work in GEO (Generative Engine Optimization). They don’t guarantee citation, but they multiply the probabilities. I put them here exactly as I told them to him, with the explanation.
The first is well-done Schema.org. 80% of websites don’t have schema, or have it badly. The AI notices. If your clinic has a LocalBusiness schema with address, phone, schedules, services, price range and responsible doctor, the AI has structured data to extract. If it doesn’t have it, it has to parse your HTML and guess. Which AI is going to recommend the website that has to guess? None.
The second is llms.txt. It’s a file at the root of your website (yourdomain.com/llms.txt) that summarises your business in markdown format so the AIs can read it in seconds. It’s like a robots.txt but for AIs. The specification is from llmstxt.org and in 2026 it’s supported (or recommended) by Anthropic, Perplexity and Cursor. It’s not magic: it’s simply giving the AI the ordered information it needs. Mine has 8 sections, ~3,000 words, and I update it every 3 months.
The third is consistent NAP. Your Name, Address and Phone must be EXACTLY the same on your website, on your Google Business listing, on LinkedIn, on social media and on every directory where you appear. If on your website you put “+34 983 123 456” and on LinkedIn you put “983123456”, the AI treats it as two different entities and splits your “authority” between the two. This is a mistake I see in 90% of the websites I audit, and it gets fixed in 30 minutes.
The fourth is presence in sources that AIs read. AIs don’t only read your website. They also read and cite content from Reddit, LinkedIn, Dev.to, Medium, GitHub, Hacker News. If on some active subreddit someone recommends your clinic, or if on LinkedIn you publish long and useful posts about dentistry with your website linked at the end, that weighs more than 50 backlinks from PBNs. AIs distinguish organic content from bought content, so don’t buy mentions: get them by being useful where your audience is.
The fifth is content that answers real questions. The AI rewards content that answers a specific question, in the first 2-3 sentences, with an example or data. This is called “answer-first content”. If your website has a “Dental veneers” page that starts with “At [Clinic Name] we are experts in dental veneers with years of experience…”, the AI will ignore it. If that same page starts with “A dental veneer in Spain costs between €350 and €800 per piece in 2026, lasts between 10 and 15 years with proper care, and is placed in 2 sessions of 1 hour each. The main types are composite (cheaper, less durable) and porcelain (more natural, more expensive)…”, the AI will cite this because it has the direct answer. See the difference?
The sixth is visible FAQs with FAQPage schema. A block of frequently asked questions in HTML with its corresponding schema. AIs extract these answers verbatim and use them in their responses when they match the query. If on your website you have “How much does a filling cost?”, “Do you accept insurance?”, “Is there parking?”, and they’re well answered with schema, the AI will use them to answer users who ask exactly that.
The seventh and last is third-party mentions (E-E-A-T). Google and AIs reward E-E-A-T: Experience, Expertise, Authoritativeness, Trustworthiness. This is demonstrated with real cases with data, photos of the team, verified reviews, articles in media or specialised blogs that mention you. It isn’t built in an afternoon. It is built over time, being a good professional and letting your work speak.
What I could NOT guarantee
This is where Peter cooled down a bit. I told him: “Of everything I’ve told you, NOTHING guarantees that when someone asks ChatGPT ‘recommend a dentist in [your city]’, your clinic will appear. Because the AI decides in each query which sources to cite, and doesn’t publish its rules. There are days it cites 3 websites, there are days it cites 8. There are days it cites yours, there are days it doesn’t. And what’s worse: the AI doesn’t warn you, doesn’t tell you ‘I’ve cited you’ or ‘I haven’t cited you’. You have no way of knowing in real time”.
Peter looked at me and said: “Then, what’s the point of doing all this if it doesn’t guarantee anything?”.
And here comes the part that no GEO consultant tells you in the first call, but which is the truth: GEO is like SEO in 2010, not like SEO in 2024. That is, there are things that work (the 7 I’ve told you), there’s a signal you send to the world, and over time the AI will cite you more and more. But nobody can tell you “if you do X, in Y months you’ll be in the top 3 of ChatGPT”. That’s fluff. And I don’t sell fluff.
I gave him an example he understood: “Think of classic SEO, the one from Google. When you rank your clinic for ‘dentist in [your city]’, are you in the top 3? Do you know exactly when you got there? Can you predict when you’ll get there?”. Peter nodded. “Well, with GEO it’s the same, but newer, with less data, and with an actor (the AI) that changes its algorithm every few weeks without warning. The strategy is the same: do things right, measure what can be measured, and have patience”.
What I DID measure for Peter (with honesty)
So that Peter didn’t leave empty-handed, I offered him something I CAN measure: the technical and citability signals that ARE under our control. They are 5 metrics, not the 47 that paid dashboards sell you.
The first is Schema.org complete and validated. I run it through the Google Rich Results Test and through Schema Markup Validator. If there are errors, I fix them. This is measured as “is or isn’t it good”, binary. Peter had it badly (3 critical errors in his LocalBusiness schema), I fixed it in an afternoon.
The second is llms.txt correct and complete. It’s a file, exists or doesn’t exist, is right or is wrong. Peter didn’t have it. I created one with his real information, not made up, and with the 8 sections I recommend.
The third is consistent NAP in 12 sites. This is measured as “is consistent or isn’t it”. I audited 12 directories where his clinic should be (Google Business, Yelp, Apple Maps, Bing Places, TomTom, Doctoralia, etc.) and found 4 inconsistencies. They were fixed in 1 hour.
The fourth is structured FAQ on the homepage and on 2 service pages. This is measured as “is there or isn’t there, is the schema right or not”. Peter had a FAQ on the homepage but without schema. I added the schema. Now the AIs can extract the answers verbatim.
The fifth and last is presence in 2-3 sources that AIs read. This is measured as “is there or isn’t it”. I recommended he start with LinkedIn (where he already had a profile) publishing 2 posts per month with his expertise in implantology, and participate in 1-2 local subreddits when he had time. No obligation, no rush.
The 5 signals, in 2 weeks, had them all green. And I told him: “this doesn’t guarantee that you show up in ChatGPT. But it guarantees that when the AI considers citing clinics in your city, yours has the best possible technical signal. And that, in 6-12 months, is noticeable.”
What happened 4 months later
Peter called me in November (4 months after our first conversation). “Daniel, I don’t know if it’s a coincidence, but a new client told me last week that she asked ChatGPT for dentists in [city] and I came up. Do you think it’s because of what we did?”.
I told him what I thought: probably yes, but I couldn’t prove it with data. There are 3 possibilities: (1) the AI was already citing me before and the client hadn’t told me, (2) the GEO improvements we made helped the AI start citing me, or (3) it’s a coincidence and the client already knew me before and that’s why the AI suggested me. I have no way of knowing for sure. What I do know is that the technical signals we sent to the world are the correct ones, and that’s the only thing we can control.
I also told him: “keep publishing on LinkedIn, keep the FAQ updated, and in 6 months we’ll talk again. If in a year you’ve published 10 useful posts and the AI hasn’t cited you once, then we review the strategy. But right now you’re on the right track.”
What I won’t promise you
If you’ve gotten this far, you’ll have noticed that this post is not a “7 tricks to appear in ChatGPT” or a “secret method to make the AI cite you”. It’s an open letter with an uncomfortable truth: nobody can guarantee you that the AI will cite you, not me, not a €200,000/year agency, not the most expensive consultant in the market.
What I CAN promise you is this: if you do the 7 things I explained to Peter (Schema, llms.txt, NAP, presence, answer-first content, FAQ, E-E-A-T), you have the best possible technical signal. If you don’t do them, it’s certain that they won’t cite you. If you do them, it’s likely that they’ll cite you. And that, in 2026, is the best that anyone can offer.
If you want me to review your website (or your client’s)
If you want me to do this same exercise: audit the 7 GEO signals, fix what fails, and leave your website with the best possible citability, write to me at landinowebs@gmail.com or via WhatsApp. I do it for €40 in a 30-60 minute session with a written report.
And if you want to see a real case of GEO applied, here it is: Restaurante Los Pinochos — case study, where “Los Pinochos” appears cited in ChatGPT for “Valencian cuisine restaurant in Valladolid”, manually verified.