Does SEO still matter for AI? We measured it

Does SEO still matter for AI?

Two claims are competing for your budget right now, and they cannot both be right. One says SEO is finished — the future is "generative engine optimisation", a new discipline of narrative and brand, and your rankings are yesterday's game. The other says nothing has really changed — the AI just reads Google, so keep ranking and the citations will follow. I have spent more than a decade being paid to tell which marketing story survives contact with a spreadsheet, so we did the obvious thing: we measured a live market, end to end, and checked.

The short answer is that SEO matters enormously — as the foundation — and is nowhere near sufficient. Rank is still the strongest single predictor of whether an AI answer cites you. And the site that ranks first can still lose the answer to a forum thread it has never heard of. Both of those are true at once, and the space between them is the most useful new number in this whole conversation.

Rank still predicts the citation — the coupling is real

Start with the half the "SEO is dead" crowd gets wrong.

In a UK health market we measured across thousands of questions, where a page ranked on Google's organic results predicted its AI citation almost perfectly in aggregate:

The first source an answer names sits, on average, at organic rank 2.1 — the AI reaches for the top of the page before anything else.

This is not a Theia quirk. Our own share of AI citations going to top-ten organic results is 57%, landing right on the independent industry benchmark of around 54% (BrightEdge's published overlap figure) — different tools, different markets, same coupling. Ranking is the shortlist the model builds its answer from, so classic search visibility is the foundation of everything that follows. To anyone telling you SEO no longer matters for AI: the data says otherwise, and it says it loudly.

But rank is not the answer — so we built an answer-layer share

Now the half the "nothing has changed" crowd gets wrong.

Ranking gets you retrieved. It does not decide who the answer actually leans on — and the answer layer does not reward position the way ten blue links do. In search, position one is a monarch: it takes roughly twice the clicks of position two. In an AI answer, the sources are more equal — being cited second, or fourth, is not the cliff-edge demotion it is in the blue links.

So we modelled it honestly. Instead of search's steep click curve, we weight each position in the answer on a much flatter one — position two keeps about 70% of position one's weight, against barely half in organic search — and from that we build a domain's share of the AI answer's attention, the direct analogue of its share of search clicks. Then we put the two side by side. The difference between them — your answer share minus your search share — is the gap. It is where "does SEO still matter" stops being a debate and becomes a measurement.

The gap: who wins the answer relative to search

On Google's AI Overview, in this market, the gap is brutal and it has a clear shape.

The sites that dominate search are under-cited by the answer. The single biggest organic player — the national health service's own site, with about a 5% share of the clicks — takes under 1% of the AI Overview's attention. The clinical databases, the review aggregators, the big pharmacy chains, the international hospital brands: all of them rank strongly and all of them are cited by the Overview far less than their rank would predict.

And the community is over-cited, massively. Reddit is pulled into the answer on 63% of the questions where the Overview appears — most of them without ranking for the question at all. Facebook, Quora, Mumsnet, YouTube and the patient forums follow it in. Google's AI Overview, asked a health question, reaches past the institutions and builds its answer out of what people said to each other.

The lesson for a marketing leader is uncomfortable and precise: being first on Google buys you the click, not the answer. If your entire AI strategy is "rank higher", you can win search and still watch the Overview recommend a conversation you are not part of.

The gap flips sign depending on the engine

Here is the finding that should change how the budget is split — and the one no single-number "AI visibility score" can ever show you.

We ran the identical measurement on the chatbots — ChatGPT and Gemini — in the same UK health market, the same week, instead of Google's AI Overview. The gap reversed. On the chatbot surface, the authority and review sites that Google's Overview under-cites are the ones that over-index: the sector regulator, the clinical-guidelines body, Trustpilot — all cited well above their search weight — while the community that dominates the Overview falls back. Same market, same questions, same week; the relationship between search and the AI answer flips its sign depending on which engine you ask.

That is why we never report a pooled "AI visibility" number, and why anyone who hands you one is averaging two opposite truths into a meaningless middle. "Where do I win the AI answer" has no single answer. It has one answer per surface: on Google's Overview you win with community and lived experience; in the chatbots you win with authority and third-party validation. Spend against the wrong one and you are optimising for a surface your buyers are not asking.

So — does SEO still matter for AI?

Yes, and here is the reconciliation the two shouting camps never reach.

SEO is the foundation. You cannot be selected if you were never retrieved, and rank is how you get retrieved — the coupling is real and the "SEO is dead" story is a headline chasing clicks of its own.

SEO is not the finish line. Ranking first and losing the answer to a Reddit thread is not a rare failure in these markets; it is the median one. The work that used to end at "rank number one" now has a second half: win the selection that happens after the ranking — and win it per surface, because the surface decides whether that means community or authority.

The brands that will own the AI answer are not the ones who abandoned SEO for a vague new religion, nor the ones who assumed their rankings would carry them. They are the ones who kept ranking and measured the gap — and then closed it where it was open.

Some take-aways

Stop arguing about whether SEO is dead. Measure two shares instead: your share of the search clicks, and your share of the AI answer. If the second is smaller than the first, you have a gap, and the gap is your brief. Close it on the door the surface actually uses — the community on Google's Overview, the authorities in the chatbots — and never, ever trust a single pooled number that hides which surface you are winning.

For the three ways a source gets into an answer in the first place, see the three doors into an AI answer; for the framework that holds all of this together, the AI-visibility ladder; and for how to measure any of it without fooling yourself, how to track AI visibility. For the whole-market view, what market intelligence actually is, or explore the platform.

---

Method note: figures are Theia measurements across live consumer markets — principally a UK private-healthcare market — captured with Google's asynchronous answer layer requested explicitly and reported as rates over repeated captures. The answer-layer influence share weights each cited position on a flatter curve than organic search and is compared to the same market's organic click share; surfaces are measured separately and never pooled. We publish the result and the reasoning, never the method that produces them.

Frequently asked

Does SEO still matter for AI?

Yes, decisively — as the foundation, not the finish line. In a UK health market we measured, a page ranking in Google's top three was cited by the AI Overview about 67% of the time; a page at position 11 or worse, about 16%. Rank is the single strongest predictor of AI citation. But ranking well and still losing the answer to a community thread is the commonest outcome we see, so rank earns you the shortlist, not the selection.

Is SEO dead because of AI?

No. The claim collapses on contact with data: where you rank on Google is still the biggest lever on whether an AI answer cites you, and our own top-ten citation share (57%) lands on the independent industry benchmark (BrightEdge's 54%). What has changed is that ranking is now necessary rather than sufficient — the answer layer selects from your ranked sources with a bias of its own, and that bias differs on every surface.

Does being #1 on Google get you into the AI answer?

It gives you the best odds, not a guarantee. In the market we measured, the site that leads organic search — a roughly 5% share of the clicks — takes under 1% of Google's AI Overview attention, because the Overview reaches past it for community discussion. Being first buys the click; it does not buy the answer.

What is the difference between SEO and GEO?

SEO earns you the rank that makes you retrievable; GEO — generative engine optimisation — is winning the selection that happens after retrieval. They are coupled, because rank strongly predicts citation, but they are not the same job: a brand can dominate search and be almost absent from the AI answer, and the gap between the two is now a metric in its own right.