Your customer no longer opens Google. She types her question into ChatGPT, reads the answer and makes a shortlist - and that answer mentions three providers. If you are not one of them, you have lost the conversation before it began. Your website may rank perfectly: in the AI answer that sits in front of it, you were not present.
That is the shift this piece is about. Not „SEO is dead" - but a second layer above search that follows its own rules. This layer is called GEO, Generative Engine Optimisation. Ignore it and you are not found less well; you simply stop being mentioned at all in a growing share of research.
What is Generative Engine Optimisation, exactly?
GEO is the practice of shaping content so that AI answer systems such as ChatGPT, Perplexity, Claude or Google AI Overviews cite your brand as a source - the core craft of an AI agency that builds visibility rather than just rankings. The difference from classic SEO is the measure of success: not the click, but the mention in the generated answer. The closely related term Answer Engine Optimisation describes the same goal from the user's side - to appear directly in the answer rather than hope for a link.
This is not theoretical. According to Similarweb's analysis, traffic from AI answers more than tripled between September 2024 and September 2025 - and ChatGPT alone accounts for roughly 87 per cent of all AI referrals. It is still a small share of total traffic, but it is the fastest-growing channel there is. Waiting until it is large means arriving too late.
Why is a good Google ranking no longer enough?
Because the AI answer comes before the ranking. When someone asks ChatGPT „which CRM suits a small team?", they get no ten blue links but a finished recommendation with three or four names. The remaining providers do not exist in that moment - regardless of who sits at position one on Google.
This decouples two things that belonged together for twenty years: ranking and visibility. You can rank superbly and be absent from the AI answer. You can rank modestly and still be cited, because your paragraph answers the question more cleanly than the competition. The rules by which a language model selects are not the rules of Google's ranking algorithm - and that is precisely the opportunity.
How does an AI decide whom to cite?
Language models favour content they can extract cleanly and attribute unambiguously. The most robust study on this is the Princeton paper „Generative Engine Optimization", which tested nine optimisation methods across more than 10,000 queries. Its clearest finding: adding evidenced statistics and citable sources to the text lifts visibility in AI answers by 30 to 40 per cent - even with no other change.
Translated, that means concrete figures with a source, clear definitions and directly quotable statements beat cloudy marketing prose. A model would rather lift the sentence „68 per cent of searches end without a click (source X)" than „many users click less often today". The first is a building block for an answer; the second is filler.
What role do AI crawlers and schema play?
Before a model can cite you, it has to be allowed to read your page at all. That is exactly what the AI crawlers govern: bots such as GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot or Google-Extended (Gemini) fetch the content that later lands in answers and training. Block these bots in robots.txt - often out of old habit - and you make yourself invisible to AI search without noticing.
The second technical layer is schema markup. Structured data translates your content into machine language: what is a definition, what is an FAQ, who is an author with what expertise. This raises the chance that a model grabs the right passage and attributes it correctly. The much-discussed llms.txt belongs here too - pleasant as an overview, but so far without a proven citation effect, so not a lever to bet on.
Are ChatGPT and Perplexity the same game?
No, and the difference matters strategically. Perplexity is an answer engine that searches the web in real time for every query and shows its sources openly - there you get cited quickly, often a day after publishing, and freshness counts heavily. ChatGPT cites visible sources less often, but because of its sheer user base it sends by far the largest share of AI traffic.
In practice: Perplexity rewards fresh, well-structured content almost immediately and works as a fast test channel. ChatGPT is the reach lever, but it moves more slowly and leans more on established authority and brand mentions across the web. Both reward the same foundation - clearly structured, citable content - only at a different pace.
Three levers to get started with GEO
Write the answer up top. Open every important page with a precise, self-contained answer in two or three sentences - the passage a model could lift verbatim without adding anything. That passage is the one that gets cited, and your brand name travels with it.
Back everything with figures and sources. Replace generic claims with concrete, dated numbers and a source link. According to Princeton this is the single strongest lever for AI visibility - and better journalism into the bargain.
Open the door for AI crawlers. Check your robots.txt, let GPTBot, PerplexityBot, ClaudeBot and the rest in, and add clean schema markup. Without this step everything else is wasted, because the models never get to see your content.
Search is not disappearing; it is gaining a new first layer - the AI answer. If you want to know how your brand appears there as a source, just drop us a message. 💌