What is GEO, and how do you actually do it

What actually changed

For twenty years, finding a business worked one way. You typed something into a search box, you got a list of links, and you clicked two or three of them. The list was long. Being tenth on it was not great, but it was not nothing, because the list was right there and people scrolled.

That is not how a growing share of buying decisions start now. Someone opens AI search and types something like “what is a good natural deodorant for sensitive skin”. What comes back is not a list. It is a paragraph, and inside that paragraph are two or three business names.

There is no eleventh result. There is no second page. There is a short answer naming a few companies, and everyone else is not mentioned at all. To that buyer, at that moment, a business that is not named does not exist.

This is not a small-company problem. In our own measurements, a clothing brand doing roughly a billion dollars a year in revenue was not named once in a set of answers to a question about the exact category it sells in. AI search answered confidently. It named other companies. The billion-dollar brand simply was not in the sentence.

What GEO means

GEO stands for generative engine optimization. It helps to take the three words apart, because each one is doing work.

Generative means the program writes something new. It does not find a page and hand it to you. It reads several things and composes an answer in its own words.
Engine is just the software doing that: ChatGPT, Claude, Gemini, Perplexity, the AI answers now sitting at the top of Google.
Optimization means changing things you control so that a result you want becomes more likely. Nothing is guaranteed. You are improving odds.

Put back together: GEO is the work of making it more likely that when AI search writes an answer about the thing you sell, your business is one of the names inside it.

You will see two neighboring terms. They overlap, and people use them loosely, but the distinction is real and it is worth holding onto.

How SEO, AEO and GEO differ in what they are trying to win.
Term	What you are trying to win	What losing looks like
SEO	A high position in a list of links	You are on the list, further down, and fewer people click
AEO	Being the source quoted in a direct answer	Someone else is quoted, and the answer is given without you
GEO	Being named inside an answer that recommends businesses	You are absent entirely, with no list to be further down

The last row is why this matters more than it first sounds. In search, being beaten costs you traffic. In an answer, being beaten costs you existence. There is no consolation position.

How an answer gets built

You cannot influence something you do not understand the shape of, so here is what actually happens between the question and the answer. It takes a few seconds and it has four steps.

Step three is the one worth staring at. AI search does not read the whole internet before answering you. It opens a small handful of pages, often around five, and writes its answer out of those. Every other page on the subject, including yours, might as well not have been published.

There is a second route into an answer, and it is worth knowing about precisely so you stop hoping for it. AI search also carries training data, which is the huge body of text it learned from before release. If your brand is famous enough to have been written about constantly for years, your name may come out of that memory with no page fetched at all. That route is real, it is slow, it is closed to almost everybody, and there is no lever on it. Everything practical happens in retrieval, the live fetch in step three.

Why most sites are invisible

When a site never gets named, the cause is almost always one of five things, and they are not the ones people expect. In order of how often we see them:

The site turns the visitor away at the door. Every website has a file called robots.txt that lists which automated visitors are allowed in. Many sites block AI search by accident, because a plugin or an agency added a rule years ago to keep scrapers out and AI search inherited it.
There is nothing to read without running the code. Some sites are built so the page arrives nearly empty and the text is assembled a moment later by code running in the browser. A person never notices. A machine that does not wait sees a blank page.
The pages never say plainly what the business is. A homepage that opens with a slogan and a photograph, with the actual trade never stated in ordinary words, gives AI search nothing to hold. It cannot name you as the best X if no sentence on your site says you do X, which is what being a clear entity means.
The pages sell instead of answering. The buyer asked a question. If every page you own is written to persuade rather than to answer, there is no passage that fits the answer being written, so nothing gets quoted.
Nobody else mentions you. AI search reads other people’s pages too. If the roundups and the forum threads and the buying guides in your category never say your name, you will lose answers no rewrite of your own site can win back.

Notice that the first two have nothing to do with writing. They are plumbing. A site can have excellent pages and still be invisible because of a single line in a file the owner has never opened.

The work, in order

The order is not arbitrary and it is not a matter of taste. Each step is worthless if the one above it is broken, so doing them out of sequence wastes money. There is no point being quotable on a page nobody is allowed to open.

1. Let them in

Add /robots.txt to the end of your domain and look at what comes back. You are looking for lines that say Disallow under a name you recognize as an AI search crawler: GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended, and a handful of others. If you find them, decide deliberately whether you want to be readable by that AI search engine. Most businesses do, and are blocking by accident.

Saying nothing about a crawler is usually the same as allowing it, so an empty file is rarely the problem. An old, aggressive one usually is.

2. Make the words survive without the code

Test it the crude way. Open one of your important pages, view the source, and search that raw text for a sentence you know is on the page. If the sentence is there, you are fine. If it is not, your text only exists after the code runs, and you have work to do that a developer will understand immediately once you can show them this.

3. Say what you are, in words a stranger would use

Somewhere near the top of your homepage there should be a sentence a twelve-year-old could read that says what you sell and who it is for. Not a slogan. Not a promise. The trade, stated. This one change moves more than anything else on this list, and it costs an afternoon.

The same idea, done for machines, is called structured data: a block of information inside the page, invisible to visitors, that states in a fixed format what the page is. This is a product. This is its price. This is the business behind it.

This is the one point in this guide where the field genuinely does not agree, so here are both positions rather than the one we prefer.

The common advice is that structured data helps you get cited. The reasoning is that a machine choosing between sources will favour the one it does not have to guess about, and structured data removes the guessing.
The largest test we know of found no effect. Ahrefs tracked 1,885 pages that added structured data between August 2025 and March 2026 against 4,000 comparable pages that did not. Citations moved down 4.6% in Google’s AI Overviews and up around 2% in AI Mode and ChatGPT, small enough to be indistinguishable from noise.

One study is not the end of an argument, and the people who disagree with it are not fools. What we would take from it is a change of expectation rather than a change of plan: add structured data, because it is cheap and it demonstrably helps search engines understand you, and stop treating it as the lever that moves AI visibility. If anyone sells it to you as that lever, ask them what they measured.

4. Write the page that answers the question

Take the question your buyer actually types. Not the phrase your industry uses internally, the one a person with a problem would say out loud. Then write a page whose first paragraph answers it directly, before any persuasion begins.

The reason is mechanical. AI search lifts the smallest passage that answers the question, and shows it as a citation. A page that spends four hundred words warming up has no liftable passage in it.

5. Attach a checkable fact to every claim

“Ethically sourced” is unquotable, because it is a word anybody can type. “Sourced from four farms in Abruzzo, listed by name, audited in March 2026” is quotable, because it can be checked. AI search leans toward specifics for the same reason a careful person does: a claim with a number attached is one somebody could be caught being wrong about.

6. Get named on pages you do not own

This is the slowest step, which is why it is last, and it is the only one with no shortcut. Roundups, comparisons, industry directories, forum threads, trade press. When AI search fetches five pages about your category, most of them are not yours. Being named on the pages it does fetch is the whole game at that point.

Do this last, not first. Every hour spent getting mentioned elsewhere is wasted while your own site is unreadable, because AI search that follows the mention arrives at a page it cannot use.

How to tell if it worked

There is exactly one honest way to know whether any of this changed anything: ask AI search the questions your buyers ask, and count how often your name comes back. Not your rankings. Not your traffic. The count.

One catch makes this harder than it sounds, and almost every tool in this market gets it wrong. AI search does not give the same answer twice. Ask the identical question three times and you can be named twice and missed once. So a single check tells you almost nothing: it can report you as absent when you are named half the time, which is a wrong answer that feels like a fact.

The fix is sampling, which just means asking each question several times and reporting the share. Named in six answers out of nine is a measurement. Named or not named, once, is a coin flip written down as a percentage.

Watch two numbers over time. Your mention rate, the share of your questions where you are named at all, and the rate on the specific questions you decided to chase. The second is the one that responds to work within weeks. The first moves in months.

This is what we built Keigeo to do, and you can read exactly how we measure it, including what the number cannot tell you.

Mistakes that waste months

Repeating your keyword everywhere. This was bad advice for search engines a decade ago and it is worse now. AI search is reading for meaning. Repetition reads as padding and padding is what gets skipped.
Writing for the machine instead of the reader. Pages written to be eaten by AI search read like they were, and they lose the human who arrives after the answer. You need both, and the same clear writing serves both.
Chasing questions you cannot win. Some questions are answered entirely out of huge published roundups, and no amount of work on your own site will change that answer. Finding out which of your questions are like this, before you spend the quarter, is most of the value of measuring at all.
Measuring once. A single run tells you what happened once. Two runs a month apart, asked the same way, tell you whether you are moving.
Starting with outreach. It is the most expensive step and the one that fails hardest when the plumbing underneath it is broken.

Glossary

Every term used above, defined without using another term you would have to go and look up. Each one has its own page if you want more than a sentence.

AI search: AI search is a program you ask questions in plain language, like ChatGPT, Claude, Gemini or Perplexity, which writes you an answer instead of handing you a list of links.
Generative engine: A generative engine is software that writes a new answer to your question rather than retrieving an existing page for you to read.
GEO: GEO, or generative engine optimization, is the work of making it more likely that AI search names your business when someone asks it a question about what you sell.
SEO: SEO, or search engine optimization, is the work of getting a page of your website to appear high in a list of search results.
AEO: AEO, or answer engine optimization, is the work of getting your page used as the source of a direct answer, such as the box at the top of a results page, rather than as one link among ten.
Crawler: A crawler is a program that visits web pages on its own and copies what it finds, so a search engine or AI search can use the contents later.
robots.txt: robots.txt is a plain text file at the top level of a website that lists which automated visitors are allowed in and which are not.
llms.txt: llms.txt is an optional plain text file at the top level of a website that tells AI search, in ordinary language, what the site is and which pages matter most.
Retrieval: Retrieval is the step where AI search fetches live web pages in the middle of answering you, instead of relying only on what it already knows.
Training data: Training data is the enormous body of text AI search learned from before it was released, which is fixed, often months or years old, and cannot be edited by you.
Citation: A citation is a link or source name AI search shows next to its answer to indicate where a piece of information came from.
Structured data: Structured data is information written into a page in a fixed format that states plainly what the page is: that this is a product, that this is its price, that this is the business selling it.
Rendering: Rendering is the step where a browser runs a page's code to build what you finally see on screen.
Sampling: Sampling means asking the same question several times instead of once, because AI search does not give an identical answer every time.
Mention rate: Mention rate is the share of answers that named your business, out of all the times a question was asked.
Entity: An entity is a specific thing in the world that a machine has decided it knows about: a company, a person, a place.

See where your own site stands

Everything above is measurable. The scan reads your site the way AI search does and tells you which of these it already passes.