AI SEO is the work of getting a business found and named when people search using AI, rather than only ranked in a list of blue links. In practice it covers two jobs that often get sold as one: making your pages eligible and quotable inside AI answers such as Google AI Overviews and ChatGPT search, and using AI tools to produce ordinary search work faster. The first job changes what buyers see. The second only changes your production speed.

Do I need special schema or an AI file to appear in AI Overviews?

No. Google states plainly that there are no additional technical requirements to appear in AI Overviews or AI Mode, no special schema.org structured data, and no machine readable AI text file to create. A page needs to be indexed and eligible to be shown with a snippet. Anyone selling a proprietary AI markup as the unlock is selling something Google says does not exist.

Is AI SEO different from normal search engine optimisation?

The foundations are the same and the scoreboard is not. Normal search engine optimisation competes for a position in a list of links, and rank is a fair proxy for whether it worked. AI search often answers the question outright, so a business can hold its rank and still lose the visit. That makes citations and mentions inside AI answers the thing to measure, alongside rank rather than instead of it.

How much traffic do AI answers actually take?

The best public measurement is a Pew Research Center study of 68,879 Google searches by 900 United States adults in March 2025. Around 18 percent of those searches produced an AI summary. When a summary appeared, people clicked a traditional search result on 8 percent of visits, against 15 percent when no summary appeared, and clicked a link inside the summary itself on only 1 percent of visits.

Can I block AI engines and still rank normally?

Partly, and the controls are separate. Google treats AI Overviews and AI Mode as Search surfaces, so nosnippet, max-snippet and noindex reduce or remove AI appearance and cut normal search visibility with it. Google-Extended is a distinct control covering wider model training rather than Search. OpenAI splits its crawlers by purpose, so OAI-SearchBot handles ChatGPT search while GPTBot handles training, and you can allow one and block the other.

How do I know if AI SEO is working?

Fix a set of buying questions your customers actually ask, then run them against the engines on a schedule and record whether you were named, which page was cited, and who was named instead. That gives a citation rate to trend. Without a repeated probe set you are relying on the occasional lucky screenshot, which tells you nothing about direction.

What is AI SEO, and what actually works?

AI SEO without the sales pitch: what Google says you actually need, what the click data shows, and the four things that move visibility in AI answers.

Ask five agencies what AI SEO is and you will get five answers, at least two of which involve a proprietary schema they will install for a monthly fee. The useful version of this question is narrower: what has actually changed about being found, and which of the work on offer moves that.

The short answer

AI SEO is getting a business found and named when people search using AI, rather than only ranked in a list of links. It matters because the answer now often arrives before the click.

The term covers two different jobs that are frequently sold as one. The first is optimising so that AI answer surfaces, meaning Google AI Overviews and AI Mode, ChatGPT search, Perplexity and Gemini, can find your pages, understand who you are, and quote you cleanly. The second is using AI tools to do ordinary search work faster, such as drafting, clustering keywords or auditing at scale. Only the first changes what a buyer sees. The second changes your production cost, which is worth having but is not the same purchase. When a proposal blurs the two, ask which of the two it is measuring.

What Google actually says you need

This is the part most sales decks skip, and it is a single paragraph on Google’s own documentation. To appear in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet. Google states there are no additional technical requirements, no special schema.org structured data to add, and no machine readable AI text file to create.

Read that again before you buy anything described as AI markup. There is no secret tag. Structured data still earns its place for the things it has always done, such as describing your organisation, products and reviews so machines read them without guessing, and Google’s structured data guidelines still apply. But it is not a hidden door into the answer box, and anyone pitching it as one is either behind or hoping you are.

The controls work the same way. Because Google treats these as Search surfaces, the preview controls you already know apply: nosnippet, data-nosnippet, max-snippet and noindex all reduce or remove AI appearance, and they cut your normal search visibility at the same time. Google-Extended is a separate control covering wider model training rather than Search. OpenAI splits its crawlers by purpose, so OAI-SearchBot serves ChatGPT’s search features while GPTBot crawls for model training, which means you can be visible in ChatGPT search while declining to be training data. Those are two different decisions and they deserve two different conversations.

What genuinely changed is the click

The measurable shift is not in how pages are ranked. It is in what happens after the results load.

The clearest public number comes from the Pew Research Center, which analysed 68,879 Google searches made by 900 United States adults in March 2025. Around 18 percent of those searches produced an AI summary. When a summary was present, people clicked a traditional search result on 8 percent of visits. When no summary was present, they clicked on 15 percent. Clicks on links inside the summary itself ran at 1 percent of visits.

That is the whole problem in three numbers. Roughly a fifth of searches, and on those searches the traditional click roughly halves. It is United States data on Google specifically, so treat it as direction rather than a forecast for your category, and note that a search that ends without a click has not necessarily ended without an outcome. Someone can read your name in the answer, skip your site and call you anyway. Which is exactly why rank alone stopped being an honest scoreboard.

The four things that actually move it

Once the mythology is out of the way, the work is unglamorous and mostly familiar.

Be unambiguous about who you are. AI answers name entities, not URLs. That means one consistent business name, address and description across your site, your Google Business Profile, your directory listings and your social profiles, plus organisation markup that says the same thing. Contradict yourself across those surfaces and you are asking a machine to guess, and it will often guess a competitor with cleaner signals.

Answer the question in the first two sentences. Engines lift passages, not pages. A page that buries its answer under six paragraphs of preamble is hard to quote, and a page that answers directly then supports the answer is easy to quote. This is the least technical change on the list and usually the one with the most immediate effect.

Be corroborated somewhere other than your own website. This is the uncomfortable one. Saying you are the best AI consultancy in Sydney on your own homepage is worth close to nothing. Being named in a directory, an industry listing, a customer’s case study or a piece of press is worth a great deal more, because it is a claim you did not make about yourself. On a new domain this is usually the binding constraint, and no amount of on-page work substitutes for it.

Be accurate, and be checkable. Google’s guidance on helpful content is explicit about experience and trustworthiness, and the Australian Government’s Voluntary AI Safety Standard sets the same expectation from the other direction for businesses deploying AI. Dated claims, named sources and visible corrections are the cheapest trust signals available, and they are the ones most sites skip.

How to measure it without fooling yourself

Pick fifteen to twenty questions a real buyer would type or say, in their words rather than yours. Run them against each engine on a fixed schedule. Record three things every time: were you named, which page was cited, and who was named instead of you.

That last column is the one people leave out and the one that tells you the most. A competitor cited every time for a question you should own is a specific, fixable gap. A Wikipedia page cited every time is a signal that the question is too general to be winnable and you should be asking a narrower one.

Trend the citation rate month to month alongside your rankings. Two numbers, both moving, is a scoreboard. One screenshot of a good answer is an anecdote. Our AI visibility checker runs this kind of probe set, and our AI SEO services page sets out how audit, implementation and monitoring fit together for businesses that want it run continuously.

Where to start

If you are choosing between an AI SEO agency and doing this in-house, or building the capability internally with an AI agency alongside you, the sequence is the same either way. Fix your entity consistency first, because it is cheap and everything else compounds on it. Rewrite your five most commercially important pages so the answer comes first. Establish a baseline citation measurement before you change anything else, so you can tell a real improvement from a good week. Then go after external corroboration, which is slow, and which is the part that no software product can do for you.

If you would rather see where you currently stand before deciding any of that, our free AI use-case audit will tell you which questions your category is losing and who is being named instead. It comes back in plain English and there is no obligation attached to it.