SEO put you on the blue links. GEO puts you inside the answer.

We asked ChatGPT, in Spanish, for the best plastic surgeons in Tijuana. It took four seconds. It came back with three names, two clinics, and a link to a magazine piece. It named one of our client's competitors. It didn't name the client.

That clinic ranks on page one of Google for "cirujano plástico Tijuana." They've been paying for SEO for two years. They have reviews, backlinks, a blog updated every Wednesday. And still, in the answer a real patient is going to read on her phone before booking a consult, the clinic doesn't exist.

That's the shift. And almost nobody in Mexico is measuring it yet.

The board changed

For twenty years, SEO was a fight for a slot. Ten blue links, one page, one click. Whoever won first place took the traffic. Whoever landed in fourth took crumbs. The metric was simple: where do you rank?

That's no longer the main game.

According to the Similarweb and SparkToro zero-click study published in July 2025, 58.5% of Google searches in the US end without a single click. When the search triggers a Google AI Overview, that number climbs to 69%. On AI-Overview-triggered queries, it reaches 83%. Most people read the answer on top and leave. Your blue link, earned with two years of SEO, never gets the chance to be clicked.

And that's just Google. The plastic-surgery query that opens this piece didn't happen on Google. It happened on ChatGPT.

GEO: the discipline that measures the new game

In late 2023, six researchers — five from Princeton, one from IIT Delhi, two independents — published a paper called GEO: Generative Engine Optimization. Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, and Deshpande. They presented it at KDD 2024, one of the serious conferences in the field. This isn't marketing. It's academia.

What they did was measure, with academic rigor, which changes to a page's content raise the probability that a generative engine — ChatGPT, Perplexity, Gemini, Claude — will cite it inside an answer. They built a benchmark with 10,000 real queries and tested nine tactics. Three won:

1. Cite sources. Pages that link to verifiable primary sources get cited more often.
2. Add statistics. When a page contains concrete numbers with attribution, the model prefers it.
3. Quote experts. Direct quotes from real practitioners raise the chance of appearing.

The number that stood out was this: for a site ranking fifth on Google — not even on the podium — applying the cite-sources tactic raised its visibility inside generative answers by 115.1%. The paper's aggregate gain went as high as 40% in visibility. For the worst-ranked sites, the lift was much larger.

Read that again. A site at fifth on Google, practically dead in the old model, can more than double its presence in the new one. Not because it climbed the ranking. Because it stopped fighting for the ranking and started fighting for the citation.

The crawlers already knocking on your door

OpenAI doesn't send a single bot to read your site. It sends three different ones, and each does something different:

• GPTBot crawls to train the models. What it learns today shows up in future versions of ChatGPT.
• OAI-SearchBot is ChatGPT's internal search index. When someone asks a question and ChatGPT browses live, this is the bot that decides which results are citable.
• ChatGPT-User fires when a live user — a potential patient, a potential buyer — asks a question and ChatGPT goes out to search right then. It's the strongest signal that your site can appear inside the answer.

Each is controlled separately in your robots.txt. Blocking GPTBot and letting OAI-SearchBot through is a real decision: I don't want them training on my content, I do want to appear when someone searches for me. It's a conversation we're not yet having with clients in Mexico, and one that in six months is going to be obvious.

What we're seeing in Mexican clinics right now

In the last five GEO audits we ran on clinics in Tijuana and Guadalajara, we found the same thing in all five. None had Schema.MedicalClinic declared on their home page. Schema.org has published these types for years. MedicalClinic for clinics, RealEstateAgent for real-estate agencies. They're structured tags that tell the model, in a language it understands, "this site is a medical clinic, not a health blog, not a pharmacy." Without that markup, the model guesses. Sometimes it guesses right. Sometimes it hands your consult to a competitor who did tag themselves.

All five shared another pattern. Blogs written for Google, not for a human and certainly not for a model. Lists of fifteen causes, thirty signs, twelve myths. Filler text between subheadings. Zero concrete numbers. Zero quotes from real surgeons. Zero outbound links to medical sources. It's the content Google rewarded from 2015 to 2022 and that generative models now ignore because there's nothing inside to cite.

One of those clinics had 3,400 words in its rhinoplasty article. Zero statistics. Zero mentions of a study. Zero quotes from the surgeon. When we asked ChatGPT for a good source on rhinoplasty recovery in Tijuana, it didn't cite them. It cited the Cleveland Clinic Journal of Medicine and a clinic in Beverly Hills.

What didn't work for us

When we started doing these audits nine months ago, the first thing we tried was rewriting the content. Rewrite it with numbers, with quotes, with citations. It worked, halfway. There was one case — a real estate firm in Rosarito — where we rewrote six long articles and visibility on Perplexity barely moved. It took us two weeks to understand the problem wasn't the content. It was that their robots.txt blocked every non-Google bot because of an old anti-scraping rule a previous developer had left in place. We were writing for models that couldn't get in to read.

The lesson, written big so we don't forget: GEO starts with letting the right bots through. Then with structuring the site so the model understands what it is. And only then with the content. In that order.

What a GEO-ready site looks like in practice

It's not complicated. It's disciplined. It's the opposite of most Mexican professional-services sites we see.

• Schema.org of the right type on the home page and on every service page.
• robots.txt that explicitly allows GPTBot, ChatGPT-User, OAI-SearchBot, ClaudeBot, and PerplexityBot, on separate lines, with comments so the next developer doesn't delete them by accident.
• One page per service, with a real number up top — average recovery time, starting price, number of procedures per year — and a direct quote from the professional who performs it.
• Outbound links to medical, regulatory, or academic sources. Sounds counterintuitive: link out to win inside. The Princeton paper confirms it with data.
• An llms.txt file at the root — the emerging standard proposed in 2024 — with a markdown summary of the site for models that respect that format. Not universal yet. In six months it might be.

None of these steps is expensive. Almost all can be done in an afternoon by a developer who knows what they're doing. What's expensive is inertia: paying an agency for two years that only knows how to fight for blue links while the patient's question never reaches your link.

The patient already changed channels

Two years ago, a woman in Tijuana thinking about a breast lift would open Google, type "mastopexia Tijuana opiniones," scan four or five links, call a clinic. Today, that same woman opens ChatGPT on her phone and writes, in natural Spanish, "I'm thinking about a breast lift in Tijuana, what do I need to know and who do you recommend?" The answer comes back in four seconds, in one paragraph, with two or three names inside.

If your clinic isn't in that paragraph, you don't exist for that patient. It doesn't matter how high you rank on Google.

This doesn't mean abandoning SEO. It means SEO became half the work. The other half — the one almost nobody in Mexico is doing yet — is GEO. Unfiltered timelines: these changes show up in real metrics in six to twelve weeks, not in thirty days. Any agency promising otherwise is selling you the old game with a new name.

The blue link took you to traffic. The citation takes you to the patient.