GEO guide: getting cited by the AIs.

What GEO is and why it concerns you

GEO stands for Generative Engine Optimization: the set of choices that make your site understandable and quotable by generative artificial intelligences — ChatGPT, Gemini, Perplexity, Copilot and the rest. It is the natural sequel to SEO: if before the goal was to appear among the ten blue links, today a new goal is added, being the source the AI uses to build its answer. The way people search is changing: increasingly they do not open ten tabs, but ask a question and read a single answer. If you are not in that answer, for that user you do not exist.

AIs read differently from people

A human reader forgives a confused text: they skip, search, guess. A language model does not: it takes what it finds and uses it. This is why GEO rewards content that answers in a direct, verifiable way. A page that clearly states who you are, what you do, where you operate and on what terms is usable material. A page full of slogans and short on facts is not. The paradox is that writing for AIs means writing better for people too: clear sentences, concrete data, statements that can be quoted without needing interpretation.

Structured data is your interpreter

Structured data (schema.org) is a language that explains to machines what a page contains: that this is an article, that that is a company with this address, that those are frequently asked questions with their answers. It does not change how the site looks, but it removes ambiguity. An AI having to answer "who builds websites in the Vicenza area" trusts a page that structurally declares its nature, area of activity and services far more than one leaving everything implicit in the text. It is invisible work that makes an enormous difference.

Being found by AI crawlers

AIs read the web with their own crawlers, and you decide whether to let them in. A robots.txt file blocking them — perhaps because of a choice made years ago and never revisited — excludes you from generated answers. Many sites have this problem without knowing it. On the opposite front, the llms.txt file is gaining ground: a plain-language map telling models what the main contents of the site are and how they are organised. It is not yet a universal standard, but it costs little and clarifies a lot. If you want to dig deeper, we have also written about how to be found by ChatGPT.

Being quotable: facts, not adjectives

A model gladly cites whoever gives it solid material: numbers, dates, procedures, short answers to precise questions. It reluctantly cites whoever writes about being an "industry leader" without a shred of data. Concretely: put FAQs where they are needed, explain prices or at least the criteria by which they are formed, state where you operate and who you work with, keep dates updated. Every statement an AI can report without risking a mistake is an opportunity for your name to appear in the answer. Every vague sentence is an opportunity lost.

Reputation outside the site still counts

AIs do not only read your site: they read what others say about you. Reviews, mentions in the local press, presence in industry directories, consistent company profiles. If your name appears uniformly across several independent sources, the model has more reason to trust it. If instead you exist only on your own site, you are a single voice. Looking after your local presence and your reviews is no longer only SEO: it is material AIs use to decide whether you are a reliable source.

Where to start

There is no need to redo everything. You start with a check: can AI crawlers access the site? Do the main pages have correct structured data? Do the questions clients actually ask have a clear, short answer somewhere? Once these three points are sorted, you move on page by page, prioritising those that answer commercial questions. It is the same approach as SEO — slow, incremental, measurable — but with one more goal: not just being found, but being described correctly by those who today answer in place of the search engines.