What AI SEO is, and what actually moves it

What AI SEO actually means

AI SEO is the work of making sure AI search names your business when someone asks it what to buy.

It is not a different discipline with a new rulebook. It is the same objective search optimization always had — be the answer, not the tenth link — moved to a place where there is no list of ten and no second page. AI search answers with a handful of businesses. You are named, or you are not there.

The phrase is loose on purpose. People use “AI SEO” for three different things, and it is worth knowing which one somebody is selling you:

Being named in AI search. Getting ChatGPT, Claude and Gemini to recommend you. This is the one that matters and the one this guide is about.
Using AI to do old SEO faster. Generating meta descriptions and outlines with a model. Useful, unrelated, and often what an “AI SEO tool” turns out to mean.
Ranking inside AI features on Google. AI Overviews and AI Mode. Closest to classic search, because Google still picks from its own index.

AI SEO, GEO and AEO: three names, mostly one job

These words are used interchangeably by people selling services and precisely by people writing about them. The distinctions are real but small, and none of them changes what you do on Monday.

SEO wants a page ranked in a list of links.
AEO, answer engine optimization, wants a page used as the source of a written answer.
GEO, generative engine optimization, wants your business named inside that answer, whether or not your page is the source.
AI SEO is the umbrella people actually type. In practice it means GEO with the search work still attached.

The distinction that does matter is the last one. A page of yours can be quoted without your name being said, and your name can be said from somebody else’s page. Those are different outcomes with different fixes, and only one of them puts a customer in front of you.

What actually changed

Three things, and only the third is new work.

The list became a shortlist. Ten blue links rewarded being roughly good at a lot of queries. An answer names three to five businesses. Position six no longer exists, so “we rank on page one” can be true while you are absent from every answer.

The click became optional. A zero-click answer resolves the question in place. Your analytics will not show a visit that never happened, so the first sign of trouble is usually a flat month you cannot explain.

The source mix widened. A search engine ranks pages. AI search builds an answer out of whatever it retrieved — your site, a round-up article, a forum thread, a retailer listing. That is called retrieval, and the practice of writing the answer from what came back is grounding. It is why a business with an immaculate website can still be missing: nothing it controls was in the pile.

The work, in the order that pays

Each step is worthless until the one above it is done. Sites skip to step four constantly and wonder why nothing moves.

1. Let them in. AI search sends an automated reader, a crawler, and a lot of sites turn it away without knowing. Check robots.txt for rules blocking GPTBot, ClaudeBot, PerplexityBot and Google-Extended, and check whether your CDN’s bot protection is refusing them at the edge. This is the single most common invisible failure, and it takes one line to fix.

2. Say plainly what you sell. Somewhere near the top of your homepage there should be a sentence a stranger could read aloud to describe your business. Not a slogan. If no sentence on your site does that job, a model has to infer your category from your navigation, and it will get it wrong in ways you never see.

3. Make the facts machine-readable. Structured data in JSON-LD using the schema.org vocabulary: what you are, where you are, what you sell, what it costs. This is not a ranking trick; it is the difference between a fact being stated and a fact being implied by your layout.

4. Give it something worth quoting. AI search lifts specifics: a price, a range, a policy, a number, a constraint. Pages written entirely in positioning language have nothing liftable in them. “Ships to 45 states” gets quoted. “Committed to excellence” does not.

5. Be a resolvable entity. An entity is a thing a model can recognise as the same across sources. Consistent name, address and description everywhere you appear, and enough third-party mentions that you exist outside your own domain. This is where E-E-A-T lives now.

6. Get into the pages the answers are built from. When you lose an answer, look at what it cited. Usually round-ups, comparison posts, forums and retailers. Getting named there moves the number in a way no change to your own site can.

What an AI SEO tool can and cannot do

Tools in this category do one of two jobs, and the names do not tell you which.

Audit tools read your site the way AI search would and tell you what is in the way. Cheap, fast, and the findings are checkable against your own pages.
Tracking tools ask AI search questions repeatedly and count how often you are named. This is the only way to know where you stand, and it is where the cost is.

The question to ask a tracking tool is how many times it asks. Model answers vary between identical runs — that is variance, and it is large. Ask once and every result is 0% or 100%: a coin toss wearing a percentage sign. A tool that asks once can tell you that you are absent on a question where you are named two times in three.

So the honest form of this measurement is sampling: the same question asked several times per AI search engine, reported as a mention rate with a confidence interval around it. If a tool shows you a number with no interval, it is not telling you how much of that number is noise.

What no tool can do is make the decision for you. The output is a work list; somebody still has to write the page.

What you can check for free, in ten minutes

Before buying anything, three checks cost nothing and rule out the most expensive failures.

Open your robots.txt. Visit yoursite.com/robots.txt and search it for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. A Disallow against any of them is a door you closed.
Ask AI search your own category question. Not your brand name — that guarantees a mention and measures nothing. Ask what a buyer would ask before they had heard of you, and see who gets named. Ask it three times: the answers will differ, which is the whole lesson.
Read your homepage as a stranger. Is there one sentence that says what you sell, in words a customer would use?

If you would rather have that done for you, a free scan reads your site the way AI search does and names every check that is failing. The fixes are the paid half; the diagnosis is not.

How to tell whether any of it worked

Pick your questions first, before you change anything, and write them down. They must be questions a buyer would type before they have heard of you, and none of them may contain your brand name.

Record where you stand: ask each question several times, of more than one AI search engine, and count the answers that name you. That is your baseline. Then do the work, wait — the web those answers are built from has to be re-crawled before anything can change — and measure the same questions the same way.

The trap is reading noise as progress. At three samples a question that moves from 33% to 67% has not necessarily moved at all. A change that does not clear the confidence interval is not a change, and any honest report will tell you so rather than let you celebrate.

What to ignore

Anyone promising a ranking. There is no rank in an answer, and nobody controls what a model writes.
Stuffing your pages with AI-sounding phrasing. AI search retrieves specifics, not vocabulary.
Publishing volume for its own sake. Forty thin pages give a model forty chances to misread you.
Chasing the broadest terms. The category question is owned by round-ups you will not displace. The winnable questions are the specific ones, and they are also the ones with a buyer attached.
llms.txt as a silver bullet. It is cheap and harmless to add, and no major AI search engine has committed to reading it. Do it last, expect nothing.

Glossary

Every term used above, defined once. The same definitions are used across the site.

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.
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.
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.
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.
Answer engine: An answer engine is any search product that responds with a direct answer instead of a list of links.
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.
Grounding: Grounding is tying an answer to real fetched sources rather than letting the model answer from memory.
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.
JSON-LD: JSON-LD is the format structured data is usually written in: a block of machine-readable facts placed in the page's code.
Schema.org: Schema.org is the shared vocabulary that structured data uses, defining what counts as a product, an organisation, an article and so on.
Entity: An entity is a specific thing in the world that a machine has decided it knows about: a company, a person, a place.
E-E-A-T: E-E-A-T stands for experience, expertise, authoritativeness and trustworthiness, the qualities Google says its raters look for.
AI Overviews: AI Overviews are the AI-written answers Google places above its normal list of links.
AI Mode: AI Mode is Google's fully conversational search experience, where the whole result is an answer you can ask follow-up questions of.
Zero-click search: A zero-click search is one where the person gets what they needed without visiting any website.
Mention rate: Mention rate is the share of answers that named your business, out of all the times a question was asked.
Sampling: Sampling means asking the same question several times instead of once, because AI search does not give an identical answer every time.
Confidence interval: A confidence interval is the range a measured number is probably within, given how many times you measured.
Variance: Variance is how much a result moves around when you repeat the same test.

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.