Get cited by AI. And make sure what it says about you is true.

The basics, no smoke

SEO for AI, explained the way we’d explain it to your CFO

What is SEO for AI?

SEO for AI, also called GEO (Generative Engine Optimization), is the work of getting AI assistants —ChatGPT, Gemini, Perplexity, Copilot and Google’s AI Overviews— to mention and cite your brand when someone asks them about what you sell. It isn’t about appearing in a list of ten blue links, but about being inside the answer the model writes. It is achieved with content structured into passages a model can retrieve and cite, with a brand entity that is consistent across every source the models consult, and with proprietary data nobody else has.

How is GEO different from traditional SEO?

Classic SEO optimises so that Google ranks a URL; GEO optimises so that a language model picks your text when composing an answer. Three things change. The unit of work is no longer the page but the passage: blocks of 80 to 150 words that answer a question in a self-contained way. The authority signal is no longer only the link, but entity consistency —your website, your profiles, directories and the press all saying the same thing about you—. And measurement moves from rankings to citations: how often, in which assistants and with what text you are mentioned. Eighty per cent of the technical and editorial work serves both disciplines; what changes is how it is structured and how it is measured.

How does ChatGPT or Gemini decide whom to cite?

Models combine what they learned in training with real-time searches. For a commercial question (“which SEO agency in Madrid do you recommend”), the assistant runs several searches, reads the best-ranked pages and third-party rankings, and writes an answer citing the sources that gave it clear, recent and verifiable passages. That is why appearing in sector listings and media, having visible dates and real authorship, and answering the question in the first lines instead of paragraph eight weighs so much. The 2026 citation studies agree that the cited passage is usually self-contained, around 100–120 words, and that content opening with the explicit question is reused almost twice as often.

How is AI visibility measured?

With a fixed panel of questions. We define between 40 and 80 real questions from your category —the ones a customer who doesn’t know you yet would ask— and run them every month against the same assistants, under the same conditions. We measure whether you appear, where in the answer, with what text and from which source you are cited, and what they say about your competitors. That turns GEO into a time series, like any other KPI, and shows which changes moved the needle and which didn’t. Without a fixed panel, what you measure is noise: assistants change their answers from one day to the next.

How long before it shows?

Entity corrections (every source describing you the same way) and content restructuring show up in assistants with live search —Perplexity, AI Overviews, ChatGPT’s search mode— within four to eight weeks. Whatever depends on model training takes longer and cannot be promised with a date. That is why we work with a question panel and a monthly review: by the third month there is enough data to decide whether to continue, change the approach or stop.