SEO vs GEO: What Changes in the Age of AI Search

For two decades, digital visibility was dominated by SEO. Brands competed for rankings, clicks, backlinks and keywords, and the scoreboard was a results page everyone could see. AI is changing where that competition happens. Instead of scrolling through results, users increasingly ask a question and read a single generated answer.

This is not the end of SEO, and anyone selling it that way is overstating the case. But the two disciplines optimize for different outcomes, and treating them as one workstream is how brands end up ranking well and being invisible at the same time.

SEO earns a position on a page. GEO earns a mention in an answer.

SEO is concerned with rankings, click-through rate and organic traffic — a page competing against other pages. GEO is concerned with whether a model names you, cites you, describes you accurately and recommends you over a competitor — a brand competing against other brands, usually without a click involved at all.

Far enough to matter. Ahrefs found that only around 12% of URLs cited by AI assistants also rank in Google's top ten. That single figure is the strongest argument against treating AI visibility as a by-product of good SEO: roughly seven out of eight sources AI leans on are not the pages winning traditional search.

Meanwhile the ground under classic search is shifting too. Rankability's August 2026 analysis of 3,751 keywords over four years found search interest in "SEO" itself down roughly 30% from its mid-2025 peak, while AI search demand rose around 260% since 2022. Neither number means SEO stopped working. Together they mean the share of discovery it covers is shrinking — and the platforms are investing accordingly, as Google's $32B Wiz acquisition suggests.

A good deal, which is why the "SEO is dead" framing is unhelpful. Work that continues to pay:

1. Crawlability and site health. A page an AI crawler cannot fetch cannot be cited. The technical fundamentals are shared.
2. Genuine subject authority. Models weight corroborated expertise, much as search does — the corroboration just comes from more places.
3. Clear information architecture. Content organized around real questions is easier to both rank and quote.
4. Structured data. Schema helps machines interpret a page confidently, whichever machine is reading.

And the parts that do not transfer, which is where most of the surprise lives:

1. Keyword targeting. Buyers ask AI full questions, often several sentences long, with context a keyword tool never captured. Optimizing a page for a two-word phrase optimizes for the wrong unit.
2. Position as the metric. There is no position three in an answer that names two companies. You are in it or you are not.
3. Traffic as the proof. A recommendation can shape a purchase without generating a session. Judging AI visibility by referral traffic understates it badly.
4. Your own site as the main lever. AI systems frequently cite review sites, forums, editorial roundups and industry lists more than a brand's own domain — so a share of the work sits outside your CMS entirely.

In traditional search, a user saw ten blue links and chose among them. In AI search, they may see one answer, three recommendations, or a single generated comparison. Visibility becomes concentrated, and the question stops being do we rank? and becomes does AI recommend us?

Share of answer is the metric that follows from that: not where you sit in a list, but how much of the response is about you rather than a competitor. It behaves differently from rank in one important way — it is not stable. Ask the same question three times and you may get three slightly different sets of names, which is why any honest measurement runs each question repeatedly rather than once.

One difference has no real equivalent in SEO. Search engines return local-language results for local-language queries. AI systems often reason in English internally regardless of the prompt language, then answer in the user's language — which means a brand with deep local-language authority and thin English coverage can be strong in national search and absent from AI answers in its own market.

For Nordic, Baltic and smaller European markets this is a structural disadvantage rather than an oversight. We looked at the mechanism in ChatGPT has an English bias, and Nordic brands are paying the price.

The organizational question is where most GEO programmes stall, and it rarely gets asked until something has already gone wrong.

The instinct is to hand AI visibility to whoever owns SEO. That is half right. The technical foundations overlap, and an SEO team already understands crawlability, structured data and content architecture. But a meaningful share of the work sits outside their remit entirely: getting into third-party lists is closer to PR, fixing entity confusion touches brand and legal, and correcting how a model describes your pricing is a product marketing problem.

We see three patterns that work:

1. SEO owns it, with a PR dependency. Fine when the main gaps are technical and on-site. Breaks down when the finding is "you are absent from the five sources AI cites", because SEO cannot fix that alone.
2. Brand or comms owns it, with SEO support. Better suited to companies whose problem is misdescription rather than absence — the work is reputation management with a technical component.
3. A named owner with a cross-functional call. One person accountable for the number, able to pull in SEO, PR and product marketing. Slower to set up, and the only one we have seen survive a change of agency.

Worth saying because most advice only adds to the list. Two habits actively mislead once AI is part of discovery.

Judging content by traffic alone. A page that earns a citation and no clicks looks like a failure in analytics and may be doing more commercial work than a page that earns a thousand sessions. If your content scoring is traffic-weighted, it will steadily deprioritize the pages AI actually uses.

Reporting AI visibility as one number. A single score hides the only useful information — which stage you disappear at and which competitor replaces you. If a dashboard gives you a score with no breakdown, it is a comfort object rather than a measurement.

The practical answer is not to choose. Keep the SEO programme that works, and add a second measurement surface for AI answers, because the first will not tell you about the second.

That means a defined set of buyer questions, run repeatedly, across more than one AI system, with results tracked over time — the approach we set out in how to measure your brand's AI visibility. If AI has already formed a view of your category and you have not read it yet, the first job is simply to look. Our Visibility Snapshot is built for exactly that starting point, and AI search is becoming AI action covers where this is heading next.