Generative Engine Optimization
Being the source the assistant cites.
A growing share of buying decisions is made inside a generated answer, without a click. We work on what decides whether you are cited there: structure, factual density, sources that hold up.
When someone asks an assistant which agency to pick, or how two vendors compare, they get an answer built from a handful of sources. Two or three, rarely more. Being one of them is not luck, and it is not decided by the same things that decide a position in a list of blue links.
What actually changes
A search engine ranks pages; an assistant extracts facts. That difference has direct consequences for how you write and structure.
First, extraction cost matters. A page that states its answer in one self-contained paragraph, with figures and a date, is easy to quote. A page that spreads the same information across twelve paragraphs of preamble loses to a weaker but more legible one. Second, format matters: content that compares, lists and rates is heavily over-represented in citations, far beyond its share of the web. Third, everything happens server-side: assistant crawlers do not execute JavaScript. Information that only appears after the browser renders does not exist for them.
What we refuse to do
We do not manufacture content designed purely to be scraped: pages with no human reader, stacking unverifiable claims. It works for a few months, then becomes a liability: once a site is identified as an unreliable source, the reputation it built turns against it, and rebuilding takes far longer than losing it did.
We also do not promise a place in the answers. Nobody controls what a model cites. What can be worked on are the conditions that make citation likely, and those can be measured.
Who this is for, and who it is not for
It pays off if your customers research before they buy: professional services, software, business decisions, long purchase cycles. On those subjects the generated answer is often the first opinion received, and it frames everything after.
It is premature if your site is not yet indexed properly, or if ninety percent of your revenue comes from paid campaigns on impulse products. In both cases we will say so, and start somewhere else.
How it runs
-
01
Baseline citation audit
We put your commercial questions to the assistants and record who gets cited, from which source, in which position. That is your starting number. Most companies discover here that they are absent from answers about their own market.
-
02
Extractability review
An assistant does not render your page, it extracts facts from it. We check what is readable without JavaScript, how much work an answer costs to pull out, and where the information dissolves into prose that states nothing.
-
03
Factual density
Dated figures, citable sources, comparison tables, definitions that stand alone. A page that answers in one paragraph gets quoted; a page that needs three readings gets skipped.
-
04
Structured data and access
Markup that matches what the page actually shows, a current sitemap, and an orientation file for assistant crawlers. Then we verify real access with the agents themselves, not just what the rules declare.
-
05
Monthly citation tracking
The same measurement as step one, every month. It is the only honest read on this work: neither traffic nor Google rankings tell you whether you are being cited.
Frequently asked questions
Is this not just SEO under a new name?
They overlap heavily: a well-structured site starts ahead. But the objective differs. SEO chases a click from a list of results; here you are trying to be the source behind an answer that often produces no click at all. In practice: factual density matters more than volume, a standalone definition beats an introduction, and a dated sourced figure is quotable where a claim is not.
How do you measure a result with no click?
By direct sampling: we ask the assistants the questions and record who is cited. It is manual, repeatable, and currently the most reliable measure available. Tools promising automated citation tracking exist, but their readings drift between sessions: we use them alongside, never as the reference.
How long before anything moves?
Faster than classic SEO on thinly covered subjects, slower on crowded ones. Expect two to three months to appear on niche questions, longer on competitive ones. We decline to give a timeline before the baseline audit: without knowing who holds the ground, the estimate would be invented.
Should we block or allow assistant crawlers?
Allow them, if the goal is to be cited. Blocking GPTBot or ClaudeBot means opting out of generated answers. The debate about paying for content is real, but it is settled at the level of a news publisher, not a company looking for customers.
Do you work with clients outside France and Algeria?
Yes. This work is largely language-independent in method and entirely language-dependent in execution: the questions have to be sampled in the market's own language, and the content written in it. We are based in Algiers and work remotely with clients across French-speaking and English-speaking markets.