AI SEO optimization: using AI to do SEO, and being named by AI

"AI SEO optimization" means two different jobs, and people searching for it almost always mean the first one. The first is using AI tools to do your SEO work faster: keyword research, briefs, drafts, internal links, technical audits. The second is optimising so that an AI assistant names your business when a buyer asks it a question. The first makes your existing work cheaper. The second decides whether you exist in the place a growing share of buyers now start.

They need different work and different measurement. Doing the first very well does not produce the second. This page covers both and says plainly where each one stops.

The two jobs, side by side

	Using AI to do SEO	Optimising to be named by AI
The goal	A ranked link in a list of results	Your name inside the answer
The reader	A person who will choose from ten options	An assistant that will name three
What you change	Volume and speed of output, technical hygiene	What your pages can prove, and where else your name appears
How you measure	Impressions, clicks, average position	How often you are named on a fixed question set, and which sources were cited
Feedback speed	Daily rank data	Varies between runs, so you need a range rather than one reading
What goes wrong	More pages, none of them specific	Specific pages nothing else references

Using AI to do SEO: where it genuinely helps

This part is real and worth doing. The honest version is that it compresses work you were already doing rather than changing what wins.

Research and clustering. Turning a long keyword export into intent groups is tedious and a model is good at it. The judgement about which cluster you can realistically win is still yours.

Briefs and outlines. Converting one buyer question into a page title and a set of sub headings is mechanical, and a model does it in seconds.

Technical review. Reading a crawl export, spotting pages with no internal links, finding duplicated titles, checking which pages render only after JavaScript. Fast and low risk, because the output is checkable.

Editing rather than writing. Asking a model to find every sentence on a page that carries more than one claim, or every claim with no number behind it, is one of its most useful jobs.

And one place it reliably hurts. A model asked to write about your business fills the gaps it has with invented specifics: coverage figures, client counts, years in business, competitor prices. Hand it your real numbers and it writes from fact. Those are exactly the claims an assistant will later try to check. Let a model shape a page and never let it supply a fact.

Optimising to be named by AI: what the work actually is

If you want to be named in an answer, three things carry most of the weight.

Pages that answer one buyer question each, with evidence. Not a services page. A page whose title is the question somebody types, whose first paragraph is the answer, and whose body is the proof. We go through the structure in how to write a page an assistant will quote.

Facts that can be checked at the moment of reading. Coverage as a count rather than "comprehensive". Prices with a date. Competitor figures with a source and the day you checked. A correct number with no source is treated as advocacy, which we wrote about in how to get recommended by Claude.

Your name on sites you do not control. Category directories, review platforms, trade press, comparison pages, public documentation. An assistant that has seen your name only on your own site has one source for you and several for your competitor.

The third one is the part people skip, and it is usually the reason a well written site still is not named.

Why the first job does not produce the second

Three reasons, and each one is a place where ordinary SEO advice points the wrong way.

Position ten is gone. Ranking tenth still puts your page on the results page. Being the tenth thing an assistant might have mentioned means it mentioned nothing about you.

Keyword matching is not the mechanism. An assistant is not looking for a page containing a phrase. It is looking for a page that lets it answer confidently. More pages at the same level of vagueness adds nothing.

Withholding the answer backfires. Conventional advice is to hold attention and reveal the answer slowly. An assistant wants the answer immediately and will pass over a page that delays it.

What does carry over is most of the technical work. If a crawler cannot read your page, nothing else matters. We set out the overlap in GEO versus SEO: what actually changes.

What to measure for each

Keep them in separate columns, because they can move in opposite directions. It is possible to lose informational clicks while being named more often, and possible to hold traffic while quietly disappearing from every recommendation. The second is worse, because nothing in a standard report flags it.

• For search: impressions, clicks and position from Search Console, as before.
• For assistants: how often you are named across a fixed set of buyer questions, which engines named you, and which sources they cited.
• For referrals: visits arriving from assistant domains. Usually small, often better qualified, and almost never looked at.

One rule on the second column. Never put your own company name in the test question. Asking "is our product good for X" primes the answer and you will see yourself everywhere. Ask the question your buyer would ask and see who comes back.

What this page does not cover

This page does not rank or review AI SEO tools, and it does not give a technical SEO checklist. It also makes no claim about how any particular engine weighs sources. The companies behind ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews do not publish that, the behaviour changes, and everything here is drawn from what is observable in the answers themselves. Treat it as a pattern rather than a rule.

What gets asked about AI SEO optimization

Which of the two should I start with?

Start with the measurement, because it is cheap and it tells you which problem you have. If you are already named in most of your buyers' questions, your work is to hold that. If you are named in none of them, more output in the old style is unlikely to change it.

Does using AI to write pages hurt my rankings?

Google's published guidance is about whether a page is useful, not about how it was produced. The practical risk is that generated pages tend to be vague, and vague pages give an assistant nothing to quote.

Do I need new tools for the second job?

You need some way to ask the same questions repeatedly and record the answers. That can be a spreadsheet and a disciplined hour each month. A tool is for when the question set and the number of engines make that impractical.

Is this just SEO with a new name?

The technical foundation is shared and the target is not. Ranking puts you on a list somebody chooses from. Being named puts you inside the answer they were given. The content work that produces each is different enough to brief differently.

What does AI Knows Us do here?

The second meaning is the one with no dashboard behind it. Using AI to draft faster leaves a trail in your own tools; being named by an assistant leaves nothing anywhere, unless somebody asks and records the answer. AI Knows Us does the asking on a set of buyer questions and keeps what came back, which turns the second meaning into something you can report on.

What to do first

• Write down the ten questions your buyers ask before they choose. In their words, not your product vocabulary.
• Ask those ten questions in two or three assistants, without naming your company, and record who gets named and which sources are cited.
• Check whether you have a page that answers each question with evidence. Most companies find they have a services page instead.
• Check where your name appears outside your own site. If the answer is nowhere, that is the first gap to close.