- GEO (Generative Engine Optimization) aims to be cited in generated answers, not only ranked in classic results.
- Answer engines favour pages that answer directly, with figures, sharp definitions and a structure a machine can read.
- Good SEO remains the foundation: without an indexed, fast, trustworthy page, nothing happens.
- Three concrete actions: an llms.txt file, short answers at the top of each section, clean structured data.
What changed in search
For twenty years the game was to appear among the first ten links. Today a significant share of queries, especially questions, receives an answer written by a model, with a few sources cited on the side. The user reads the answer, sometimes clicks a source, often does not. For an expertise site like ours, this means fewer visits on some queries, but more qualified visits when the citation is there.
We measured the effect on our own articles: on "how" and "what is" queries, classic traffic fell, while contact requests mentioning "seen in ChatGPT" or "cited by Perplexity" appeared. Visibility changes shape; it does not vanish.
How an answer engine picks its sources
The exact mechanisms are not public, but the regularities are clear. Cited pages answer the question in the first lines of a section, without detours. They contain verifiable facts: figures, dates, definitions, comparisons. They are structured with explicit headings, tables and lists a model can extract cleanly. They are recent, or at least dated. And they come from sites the engine considers trustworthy, which brings us back to the fundamentals of search: authority, inbound links, consistency.
What does not work
Texts that circle the subject for three paragraphs before getting to it. Undated pages. Vague claims without a figure. Content copied from one site to another. And, paradoxically, visibly AI-generated text: answer engines readily cite human, precise, signed field reports.
The three actions to take this week
An llms.txt file at the root of the site
A text page describing your organisation, your main pages and your key content, in a format designed for models. It is the sitemap of answer engines. Ours is available at datasai.fr/llms.txt; it lists our services, the Academy and every article with its description.
A short answer at the top of each section
Every section heading poses an implicit question. The first two sentences must answer it, with a figure or a fact where possible. The development comes after. This discipline also improves human reading: nobody complains about a quick answer.
Clean structured data
Schema.org markup (Article, FAQ, Organization, Course, JobPosting) gives machines an unambiguous reading of your page. It does not guarantee a citation, but its absence makes one unlikely. Check it with Google's rich results test.
Ask three answer engines the question your customers ask before they find you. If none cites you, you know where to start. Repeat the test every month.
What GEO does not replace
GEO is not a new discipline that makes search optimisation obsolete. It is an extension. A slow, poorly indexed site with no inbound links and no clear pages will be cited by no one. Conversely, solid, precise, signed content serves both channels at once. The companies doing best are those writing for a demanding reader, not for an algorithm.
"Answer engines cite what they can verify. Give them facts, a date and a name."
Conclusion
Zero-click search is here, and it favours precise, trustworthy sources. For an expertise business that is rather good news: quality becomes the criterion again. An llms.txt file, direct answers at the top of each section and clean structured data are enough to begin. The rest is a matter of true content, regularly updated.
FAQ
Does GEO replace SEO?
No. Classic SEO remains the entry condition: indexing, speed, authority. GEO adds a layer of clarity and structure to be cited in generated answers.
How do we know if we are cited?
Through regular manual tests on answer engines, noting the questions asked and the sources cited. Tracking tools are emerging, but a hand-kept monthly table is enough to start.
Should we block AI crawlers?
If you want to be cited, no. You can however protect private pages and point crawlers to the summaries meant for models, as we do in our robots.txt.
Can AI-generated content be cited?
Rarely, unless it brings unique facts. Answer engines favour original sources: field reports, measured figures, argued positions.
