A habit changed while almost nobody was measuring it. Before searching Google, a lot of people now ask a model. And the model does not hand back ten links: it hands back an answer, with two or three names inside it.
If you are not one of those names, you do not exist for the person who asked. You are not on page two. You are simply absent.
What is the difference between GEO and SEO?
The two look more alike than they are, and the point where they diverge is the one that matters.
| SEO | GEO | |
|---|---|---|
| Goal | be the first result | be the sentence quoted |
| Winning unit | the page | the paragraph |
| Who decides | the ranking algorithm | the model doing the summarising |
| What it rewards | domain authority, links | clarity, data, verifiability |
| How you measure | position, clicks | how often it names you, and how |
| Time to respond | months | weeks |
The row that changes everything is the second. Google indexes pages, a model extracts passages. Which means an article can be cited for a single paragraph, the one written well, and the rest of the page does not matter in that moment.
That has a consequence which, on our own domain, is the reason this guide exists: GEO pays off before SEO does when a site is new. Ranking first on Google needs authority and links, and those take months to build. Getting cited needs you to be the clearest source on that specific question, and that you can do today. The paper that named GEO, presented at KDD 2024, measured it: when sources were cited, pages in fifth position gained 115.1% visibility, while the page in first lost 30.3%.
Being there is not enough. You have to be quotable.
What do you change on the site to get cited?
None of it is exotic. Just dull work done properly.
- NOIThe direct answer up top — three or four lines answering the question in the title, standing on their own without the rest of the article
- NOIReal FAQs — the questions people actually type, with answers of 90 characters and up. These are the units models quote most often
- AISchema — `FAQPage`, `Article`, `Organization` with `sameAs`. It tells the machine what it is reading
- NOIData with a source — a number without a source does not get quoted, and rightly so
- AICrawler access — `robots.txt` and `llms.txt` that say explicitly who is allowed to read
The first point is worth more than the other four combined. An article that opens with two hundred words of throat-clearing before answering does not get cited: the model cannot find the sentence to extract and goes elsewhere. Putting the answer at the top feels like a narrative sacrifice. It is not. That is what the reader wanted anyway.
The fourth point has numbers behind it. In the same GEO paper, the three edits that worked best were adding quotations, statistics and cited sources: up to 40% more visibility in generative answers. Keyword stuffing did worse than leaving the text alone.
On the fifth, the file that actually controls access is `robots.txt`. OpenAI splits OAI-SearchBot, which decides whether you appear in ChatGPT search answers, from GPTBot, which is about training: you can allow one and block the other. Anthropic does the same with Claude-SearchBot and ClaudeBot, and says its bots honour `robots.txt`.
The llms.txt file, and why it is not magic
You see `llms.txt` around now: a text file at the site root describing who you are and what is inside, written for models. Jeremy Howard proposed it in September 2024.
Honestly: it is a proposed convention, not a standard models are obliged to respect. Nobody can guarantee ChatGPT reads it. For its own AI Overviews, Google says you do not need AI text files or special markup. Ten minutes to add and it does no harm, so we put it there, but anyone selling it to you as the key to GEO is selling you a text file.
What actually counts is on the page: the answer at the top, the FAQs, the verifiable data.
How do you measure whether AI cites you?
This is where ninety per cent of GEO conversations fall apart. People talk about it as a strategy, then nobody can say whether it is working, because AI citations have no Search Console.
The method we use is mechanical and it works.
You write a fixed set of thirty questions, the ones a real customer would ask a model before choosing. Not keywords: questions, in their words. Then every month you ask ChatGPT, Perplexity and Gemini those same thirty, always the same, and you record three things: whether the brand appears, where in the answer, and whether the linked source is yours or somebody else's.
Three numbers, thirty questions, three models, once a month. Twenty minutes of work, and the only thing that turns "we are doing well on AI" from a feeling into a fact.
One warning about the results: models are not deterministic. The same question, on the same day, can give different answers. So the set has to stay fixed, and you read the trend over three or four months, never a single reading.
What to expect, honestly
Nobody can guarantee a citation, just as nobody can guarantee first place on Google. Anyone guaranteeing it is selling something else.
What can be said is where GEO pays off most: on specific, sector questions, where clear sources are scarce. On a generic question the model cites the big publications and there is no contest. On a niche question, where you know something others have not written down, you are a plausible source even with a new domain.
Which is the oldest SEO advice in the book, arriving through a different door: write about what you actually know, and write it better than anyone else. The engine changed, the craft did not. For the wider picture of how we use AI without producing slop, start with what vibe marketing actually is; for how this plays out on a real network, see franchise development.
Frequently asked questions
What is the difference between GEO, AEO and SEO?
SEO aims to rank a page in search results. AEO, Answer Engine Optimization, aims to be picked as the direct answer, for example in Google's featured snippet. GEO aims to be cited by generative models like ChatGPT or Perplexity. The techniques overlap heavily, and in practice a well-made piece works on all three at once.
Do I need a paid tool to do GEO?
No, and there is still no reliable tool for measuring AI citations at scale. The thirty-question method, run by hand once a month, costs twenty minutes and gives you sounder data than any dashboard promising automatic mention tracking.
Should I open my site to AI crawlers or block them?
That depends on the business model. If you sell services and want to be found, opening up makes sense: blocking OAI-SearchBot or PerplexityBot means choosing not to exist in those answers. GPTBot is about model training: you can block it and still appear in ChatGPT search. If you live on advertising traffic the calculation is the opposite, because a citation can replace a visit. The one option that makes no sense is not deciding and leaving the default.
How long before the first citations appear?
First mentions on niche questions can arrive within weeks of publishing, because models with live search read the current web. On broad and competitive questions it takes far longer and depends on overall domain authority, exactly as in SEO.
Does llms.txt actually work?
It is a proposed convention, not a standard models are required to honour, so nobody can guarantee it gets read. Ten minutes to add, no harm done, so we add it. But it is not where you win: you win with the direct answer at the top of the article and FAQs written as real questions.
Will GEO replace SEO?
No, and anyone saying so is oversimplifying. Google searches still happen and organic traffic is still the largest source for most sites. What changes is that a share of questions, especially the exploratory ones early in the journey, moves to the models. Both need covering.
Want to know whether AI names you?
We will tell you which questions you show up on and which you do not, and what it takes to change that. Seven questions, founders answer within 48 hours.
ApplyKeep reading
Founder of As Growth, an AI-first growth partner. He runs growth and strategy across partner businesses: OroEtic, Giojello, Equity Plus and others. Apply if you want to work with us.