In short
AI SEO is the work of making a website visible in AI-generated answers as well as in traditional search results. The foundation is the same as SEO: crawlable pages, real authority, content that matches intent. What changes is the unit of competition, from ranking a page to being quoted inside an answer.
Key points
- AI SEO is SEO plus a citation layer, not a replacement discipline.
- The unit of competition shrinks from the page to the passage. Assistants quote paragraphs, not URLs.
- Rankings are stable and trackable. Citations are probabilistic and vary run to run.
- Most of the new work is structure and evidence, writing claims a model can lift cleanly.
- If your SEO fundamentals are broken, fix those first. No citation tactic compensates for an uncrawlable site.
The short version
AI SEO is the work of making your site visible where AI writes the answer: Google's AI Overviews and AI Mode, ChatGPT, Perplexity, Copilot. It sits on top of traditional SEO rather than replacing it. The crawling, the intent research, the authority building all carry over. What changes is what you are competing for. In classic search you compete for a position on a list. In AI search you compete to be one of the three or four sources an answer is assembled from.
I have done SEO for twelve years, and my honest summary is this: about seventy percent of the work is the same work with the same name. The other thirty percent is genuinely new, and it is mostly about structure, evidence and being a recognisable source. This page walks through both parts.
What stays the same
Assistants do not browse the web freehand. When one needs current information, it queries a search index, Google's, Bing's or its own, retrieves a shortlist of pages, and writes from them. That mechanism has a blunt consequence: everything that made a page retrievable for search makes it retrievable for AI.
So the old checklist survives intact:
- Crawlability and indexation. A page that is blocked, noindexed or buried is out of the game before it starts. AI search adds new crawlers to allow, but the discipline is the same.
- Search intent. Assistants answer questions people actually ask. Content mapped to real questions was good SEO in 2015 and is the entry ticket now.
- Authority. Links, mentions and a consistent identity still decide whether you are a plausible source. Models lean on the same signals the indexes already computed.
- E-E-A-T. Google's quality framework, experience, expertise, authoritativeness and trust, reads like a description of what a cautious model wants in a source it names.
If an agency tells you AI SEO makes any of that obsolete, they are selling novelty, not results.
What actually changes
Four things, and they are specific.
The unit shrinks from page to passage. A ranking is won by a whole page. A citation is won by a paragraph. Assistants lift the passage that answers the question, which is why a page with one crisp, self-contained answer near the top beats a better page whose answer is smeared across twelve hundred words. Structure stops being cosmetic.
Evidence becomes mechanical, not rhetorical. The Princeton GEO study, the first controlled look at this, found that adding quotations, statistics and cited sources to a page raised its visibility in generated answers by up to around forty percent in their benchmark. A claim with a number and a named source is easier for a model to justify quoting than the same claim written as opinion.
Identity starts to gate selection. Models prefer sources they can resolve: a named person, an organisation with consistent facts about it across the web, an entity rather than a domain. Two identical articles are not equal if one has a recognisable author and the other does not.
Measurement changes character. A ranking is stable: position four today, probably position four tomorrow. A citation is probabilistic: the same question asked twice can cite different sources. You stop tracking positions and start sampling, asking a fixed set of questions on a schedule and counting how often you are named. Single results mean nothing. Trends over months do.
The same work, two scoreboards
| Traditional SEO | AI SEO | |
|---|---|---|
| You win | a position on a list | a mention inside an answer |
| Unit of competition | the page | the passage |
| Result stability | stable between crawls | varies run to run |
| Primary metric | rank, clicks | citation rate, share of voice |
| Traffic shape | clicks at every position | fewer clicks, higher intent |
| Core lever | relevance and links | quotable evidence and identity |
| Time horizon | months | months, sometimes faster on fresh topics |
How to start, in order
- Confirm the foundation. Crawlable site, indexed pages, AI crawlers not blocked at the firewall. The Getting found by AI section covers this layer.
- Pick the twenty questions that bring you business, and check what the answer engines say today. Ask each one in ChatGPT, Perplexity and Google. Write down who gets cited. That list is your real competitor set, and it is usually a surprise.
- Restructure your key pages so each opens with a direct, complete answer in the first sixty words, followed by the evidence.
- Add verifiable substance: a real number, a named source, a claim quotable in one sentence. Pages made of paraphrased common knowledge give a model nothing to cite.
- Build the identity layer: named authors, an about page with checkable facts, consistent profiles, schema that ties it together.
- Re-run your twenty questions monthly and track the trend, not the individual runs.
Common mistakes
- Treating it as a separate project. Standing up an "AI SEO workstream" beside an SEO workstream doubles meetings, not results. It is the same site and mostly the same levers.
- Optimising citations while blocking crawlers. I have audited sites paying for GEO advice while their firewall returned 403 to GPTBot. Access first, always.
- Chasing every assistant separately. The platforms share most of their selection logic. Structure, evidence, identity move all of them. Platform-specific tuning is the last five percent, not the first move.
- Judging from one prompt. Citations vary between runs. Deciding anything from a single ChatGPT answer is reading tea leaves.
- Publishing AI-written summaries of other people's pages. A model has no reason to cite a paraphrase of content it can already see at the source. Information gain is the whole game.
When this does not apply
If your business does not depend on being found for questions, most of this page does not either. A restaurant lives on maps and reviews. A brand people search by name already gets cited for its own queries. And if your search fundamentals are genuinely broken, no citation layer will compensate: fix crawling, intent and content quality first, because AI search sits downstream of all three.
FAQ
What is AI SEO? AI SEO is the practice of making a website visible in AI-generated answers, in Google's AI Overviews and AI Mode, ChatGPT, Perplexity and Copilot, as well as in traditional rankings. It keeps the foundations of SEO and adds a layer of work aimed at being quoted as a source inside the answer itself.
How is AI SEO different from traditional SEO? The foundation is shared, but the unit of competition changes. Traditional SEO competes for a ranked position with a whole page. AI SEO competes for a citation with a passage, so structure, quotable evidence and a verifiable identity carry more weight, and results are probabilistic rather than stable.
Is SEO dead? No. Assistants retrieve their sources from search indexes, so pages that cannot rank generally cannot be cited either. Classic search still drives far more visits than AI referrals for almost every site. SEO is the foundation AI search runs on, which is the opposite of dead.
Do I need a separate AI SEO strategy? You need a citation layer inside your existing strategy, not a second strategy. The same pages serve both goals. What changes is how they are structured, what evidence they carry, and how you measure success, which now includes citation sampling alongside rankings.
How do I measure AI SEO? Sample it. Fix a list of questions that matter to your business, ask them across the major assistants on a schedule, and record how often you are mentioned or cited. Track AI referral visits in your analytics alongside it. Judge trends over months, because individual runs vary.
Does schema markup matter for AI SEO? It helps at the margins and costs little. Schema does not make weak content citable, but it makes a good page unambiguous: who wrote it, what entity it belongs to, what question each FAQ answers. Ambiguity is friction, and friction loses close selection decisions.
Can a small site do AI SEO? Yes, and narrowness helps. Assistants reward the page that answers a specific question completely, and a specialist can out-answer a generalist on their own ground. What a small site cannot do is win citations on broad questions against established authorities, so pick the questions where you are genuinely the best source.
Sources
- GEO: Generative Engine Optimization, Aggarwal et al., KDD 2024, the Princeton study on what raises visibility in generated answers
- Google Search Central documentation on AI features in Search
- This site's own monthly prompt sampling, which is where the workflow above comes from
- First published.