What Is GEO? How to Get Your Brand Surfaced in AI Search
The short answer
GEO (Generative Engine Optimization) is the work of making your content findable and — crucially — citable by generative search engines: ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. The difference from SEO: SEO aims for a click on a link, GEO aims for your name inside the answer. It comes down to four steps: (1) explicitly allow the crawlers in robots.txt (GPTBot, ClaudeBot, PerplexityBot, Google-Extended…), (2) add machine-readable structure to every page — JSON-LD (Article, FAQPage, Organization/Person), (3) open every article with a direct, quotable short-answer paragraph, (4) publish an llms.txt at your site root as a plain-text content map. On top of that, the decisive factor is original first-hand information: models cite what only you have, not what everyone has.
Why is everyone suddenly talking about this?
Because search behaviour changed quietly. A customer no longer types "hair transplant clinic Istanbul" and scans ten blue links; they ask "which clinic would you recommend in Turkey and what should I watch out for" and get a single paragraph back. If your name isn't in that paragraph, your ranking doesn't matter — because nobody is looking at the list.
This doesn't mean SEO is over. Most models still build answers by reading search results — if you're invisible on Google, you're invisible in AI too. GEO isn't a replacement for SEO; it's a layer on top of it.
I first took this seriously because of a client's inbound message. On a health-tourism brand we work with, a patient wrote "I asked an AI and it recommended you". Until that moment no dashboard we had measured this — even in Analytics it wasn't clear where that traffic came from. Then I noticed the reverse: when I asked the same question, other brands' names came up, and there was a reason. Their sites had clear answers a model could read and quote; ours had marketing sentences.
The real difference between SEO and GEO
| SEO | GEO | |
|---|---|---|
| Goal | A click on your link | Your name inside the answer |
| Success metric | Rank, CTR, sessions | Citation and brand recall |
| Content format | Long text around a keyword | Direct answer + structured data |
| What wins | Authority, backlinks, technical health | Originality, clarity, machine readability |
| Biggest mistake | Keyword stuffing | Rewriting what everyone already says |
| Measurement | Search Console, Analytics | Manual testing + referral traffic (still immature) |
GEO's weakest point is measurement. There's no reliable answer today to "how often was I cited". So if an agency promises you a clean ranking report for GEO, be careful — a report on something unmeasurable is, by definition, a guess.
The setup: what I did to this site today
Instead of theory, let me use my own site — the page you're reading applies every step below. You can open the source and check.
Explicitly allow AI crawlers in robots.txt
Plenty of sites block these bots without knowing it — some security plugins do it by default. I wrote explicit allow lines for GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, Google-Extended, ClaudeBot, Applebot and CCBot. Note: blocking Google-Extended won't drop you from normal Google search, but it costs you any chance of appearing in AI Overviews.
Add JSON-LD to every page
This page carries three schemas: BlogPosting (headline, author, date, keywords), FAQPage (the questions below) and BreadcrumbList. In the author field I don't just give a name — `sameAs` links my LinkedIn/X/Instagram profiles and `worksFor` names my companies, so a model can connect "who is this".
Put the direct answer at the top of every article
The "short answer" box above isn't decoration. The system requires that field on every article, and the same text goes into JSON-LD as `abstract`. It's exactly what a model looks for when reading a page: one paragraph, unhedged, straight to the point. A paragraph that opens with "this is a broad topic that depends on many factors" never gets quoted.
Publish an llms.txt
A plain markdown file at your site root: who you are, what you do, and a short annotated list of your pages. Mine lists blog articles, guides and site pages in both languages, and it regenerates automatically on every build. It isn't an official standard yet, but it costs nothing and the signs that it gets read are increasing.
Publish in two languages and pair them with hreflang
Every article here exists in Turkish and English, on separate URLs, paired with hreflang. The reason isn't only international traffic: models still lean toward citing English sources. With Turkish content you have a shot at Turkish questions; with English you have a shot everywhere else.
The decisive factor: do you have anything worth citing?
Everything above is technical plumbing and can be built in a day. Yet hundreds of sites have that plumbing and get cited nowhere. The reason is simple: models don't cite information everyone has, because they already have it.
The content that does get cited shares these traits:
- It contains numbers. "AI saves time" gets ignored; "we saved 14,200 hours across 120 projects" gets cited. A number is a verifiable claim.
- It carries first-hand experience. A process only you went through, a cost only you know, a failure only you saw.
- It takes a position. "It depends" never makes it into an answer. "Don't do this, do that, because…" does.
- It has structure. Tables, numbered steps, FAQs. A model can break a list apart and slot it into an answer far more easily than a paragraph.
- Authorship is clear. A named author with a title, an organisation and verifiable profiles is a far stronger signal than an anonymous "editorial team".
On this site I took that one step further: there's a digital twin speaking in my name on the page. Visitors can ask it questions and it answers from my own knowledge base. I didn't build it as a demo — I wanted the information about me to exist somewhere in the form I wrote it. Models and people end up looking at the same source. I've written up how I built it step by step in a separate article.
How to test it, and how long it takes
- Write the ten questions a customer would ask. Not "my brand name" but "the problem I solve": e.g. "who manages social media for small businesses".
- Ask each of those in ChatGPT, Gemini, Perplexity and Google AI Overviews. Note whose names come up. That's your baseline.
- Visit the sites that were named and look for it: which paragraph could that answer have come from? Almost every time you'll find a clear, directly written passage.
- Build the setup, publish the content, then wait 4–8 weeks. That's a realistic window for pages to be crawled and indexed.
- Ask the same ten questions again and compare. You have no measurement tool, but this manual test is an honest indicator.
Realistic expectation: for a small brand, appearing in broad "best X" questions takes months and is never guaranteed. But in niche, specific questions you can surface within weeks — because there the constraint isn't competition, it's the absence of information.
Key takeaways
- GEO doesn't replace SEO, it layers on top. If you're absent from Google, you're absent from AI.
- Four technical steps: crawler permissions, JSON-LD, a direct answer at the top, llms.txt.
- A paragraph starting with "it depends" never gets quoted; one with numbers and a position does.
- Models cite what only you have, not what everyone has.
- Measurement is immature — be sceptical of any agency promising clean GEO ranking reports.
Frequently asked
Does llms.txt actually get read?
It isn't an official standard yet, and no provider guarantees they read it. I add it anyway because the cost is near zero: it's a text file, generated automatically at build time in my case. No downside, possible upside. But it isn't a strategy on its own — adding llms.txt while leaving the content weak achieves nothing.
Should I block AI bots? Isn't my content being taken?
It's a real tension and the answer depends on your model. If content *is* your product (a course, a database, a paid publication), blocking is defensible. If your content exists to sell a service, blocking shoots you in the foot: nothing gets taken if nobody finds you. I sell a service, so I opened every door.
How much should I pay an agency for GEO?
The technical setup (robots.txt, JSON-LD, llms.txt, page structure) is a day's work and should carry a one-off price — it isn't a retainer. The only thing that earns a monthly fee is content production, because that's where the difference is made. Steer clear of any agency selling you a "GEO subscription" that amounts to technical tweaks.
Is AI-written content a disadvantage for GEO?
What matters is what it contains, not how it was written. But in practice: content that comes purely out of a model repeats what the model already knows — so there's no reason to cite it. Writing *with* AI isn't the problem; writing what AI already knows is. I prepare these articles with AI assistance too; the experiences, numbers and mistakes in them are mine.
When will I see results?
The technical setup starts showing 4–8 weeks after crawling. On the content side the rule is: weeks for niche, specific questions; months for broad, competitive ones. For questions about your own brand ("is company X reliable") the effect is much faster, because there you're the primary source.