Websites and SEO
What is GEO and how to get your company cited by ChatGPT and other AIs
More and more people ask ChatGPT, Gemini or Perplexity a question instead of typing a search and opening ten pages. The answer comes ready-made, sometimes naming two or three companies. If yours is not among them, you do not even show up on page two: you simply do not exist in that conversation. GEO is the work of changing that.
What GEO is
GEO stands for generative engine optimization. It is the set of practices that help your company's content be found, understood and cited by AI assistants when they answer a question.
A warning right at the start: GEO is a new discipline. AI vendors do not publish a list of rules, the systems change often and nobody can guarantee that a company will be cited. What exists are practices consistent with how these systems work. Be wary of anyone who promises a position.
SEO and GEO: what changes
In traditional SEO, the goal is to rank well in a list of links. In GEO, the goal is to be part of the answer. The person may never visit your site, but they read your company's name associated with the solution to their problem.
The good news is that the two jobs overlap almost entirely. Many assistants search the web at the moment of the question and read the pages they find. If your site is not crawlable, fast and clear, it is read by neither the search engine nor the AI. That is why we treat GEO as a layer on top of well-done SEO, not as a replacement. The technical foundation is in Technical SEO for a company website.
How AIs reach your content
There are two paths, and they call for different strategies.
- What the model learned in training. Public texts that existed when the model was trained. You influence this only in the long run, by being mentioned consistently in relevant places.
- What the assistant looks up on the spot. Many assistants run a web search, read a few pages and assemble the answer, citing the sources. Here you have direct influence: your page needs to be found, read without difficulty and contain a passage that answers the question.
Practices that help
1. Let the AI bots in
Check the site's robots.txt file. Some sites block AI bots without the owner knowing, because of a platform or security-service default. Blocking is a legitimate decision, but it needs to be a conscious one: if the bot cannot read, the assistant cannot cite.
2. Deliver the content in HTML
Many bots do not run JavaScript. If the page text only appears after a script runs, the bot sees an empty page. The main content should be in the HTML delivered by the server.
3. Write passages that answer on their own
The assistant extracts pieces. A paragraph that opens with the direct answer, in one or two sentences, and only then gives detail is more likely to be used. Headings phrased as questions help match what people ask.
4. Be specific about what you do and for whom
"Complete solutions for your business" answers no question. "Collections automation for clinics that serve patients on WhatsApp" does. State clearly the service, the audience, the region and how it works.
5. Use structured data
Schema.org markup, in JSON-LD, tells machines what the page is: an organization, a service, an article, a question and answer, a review. Search engines have recommended it for years. For AIs, it helps reduce ambiguity, although no vendor guarantees it is decisive.
6. Show where the information comes from
Content with a source, a date, authorship and verifiable numbers is more trustworthy to people and to machines. Update the date when you truly review, not out of routine.
7. Stay consistent outside the site
Name, address, services and description should be the same on the site, on the Google business profile, on LinkedIn and in industry directories. Conflicting information confuses any system trying to understand who you are.
8. Consider an llms.txt
It is a recent proposal: a text file at the root of the site that summarizes, for language models, what the company does and where the main pages are. It is an emerging practice, with no adoption guaranteed by the major vendors. It costs little and does no harm, but do not count on it as a solution.
GEO done well is, in practice, SEO done well with one extra step: writing to be cited, not just to be found.
What does not work
- Hidden text with instructions for the AI. Besides being unethical, it is detectable and can hurt the site.
- Stuffing the page with keywords. It did not work well for search engines and works worse for models that read meaning.
- Publishing hundreds of generic AI-generated texts. Content that adds nothing has no reason to be cited. We discuss this in Does a company blog still work?
- Buying a "citation guarantee". Nobody controls the answer of a third-party model.
How to measure
Measurement is still imperfect, but you can keep track:
- Test questions. List ten to twenty questions a customer would ask and test them every month on the main assistants. Note whether the company appears, how it is described and which sources are cited.
- Visits from assistants. In the site's analytics, look at traffic coming from those services' domains.
- Brand mentions. What does the assistant say when asked directly about your company? If the description is wrong, the problem is in the public information available.
Results vary from one day to the next and from one assistant to another. Look at the trend, not at a single answer.
Where to start
Do three checks today: does robots.txt block any AI bot? Does the main text of your pages show up with JavaScript turned off? Does the home page say, in one sentence, what the company does and for whom? Those three answers already show the size of the work.
For a fuller analysis, delpra™'s SEO & GEO Radar evaluates your site against dozens of technical rules and AI-visibility rules and shows what to fix first.
Written by the delpra™ teamProcesses, automation and AI so operations scale efficiently.