Citation versus ranking
Google ranks pages; an assistant composes an answer and cites its sources. The logic changes: your position on a broad keyword matters less than being the most reliable source on one precise question your clients actually ask.
An assistant looks for pages that answer plainly, with verifiable facts and a clear origin. A 4,000-word “complete guide” that buries the answer serves less than a page that asks the question in its title and answers it immediately.
What makes a page citable also makes it better for Google and for your visitors. The two kinds of visibility are built together: the same groundwork serves both SEO and AI visibility, and it can be measured on both sides.
The structure of a citable page
The title is the question, phrased the way your clients ask it. The answer fits in three sentences on the first screen: the figure, the condition, the nuance. The rest of the page develops, section by section, each sub-question that naturally follows the first.
Every claim must be attributable: a dated figure, a named source, an identified author who signs the page. Assistants cite what they can verify; anonymous generalities do not get cited. A figure published with no date and no origin will be neither verified nor reused: it will be ignored.
Add a short FAQ on neighbouring questions, and links to your in-depth pages. The page you are reading applies exactly this structure. Question as title, direct three-sentence answer, sections that develop it, FAQ, author and date: the model is repeatable for every question your clients ask.
Structured data and machine readability
The invisible structure matters too. Structured data (Organization for the publisher, Article for the author and date) tells machines who is speaking and about what. Google states, however, that no special markup is needed for its AI Overviews, and it has not shown FAQ rich results since 7 May 2026. Clean HTML does the rest: a single h1 per page, a logical heading hierarchy, real text never locked inside images.
Dates matter too: a page that is dated and updated gets cited more readily than a page without an age. Author, date, organisation: the three basic trust signals, readable by a machine as well as by a human.
Crawler access and the llms.txt file
Crawlers need to be allowed in. Check your robots.txt: blocking AI crawlers on public content means giving up on citations. Allow them on what is public, protect what must be protected. On Google's side, AI Overviews rely on Googlebot's usual crawling; Google-Extended changes nothing there.
The llms.txt file, placed at the root of the site, lists your important pages for AI tools, following an open proposal. Google states that its search does not use it, AI features included; other tools may read it. It costs little, provided you expect nothing decisive from it.
These files are set up in a few hours. It is the easiest part of the job, and it is worthless without the pages behind it. The order matters: content first, technical files second.
Measuring citations
For Google, Search Console has published generative AI reports since late August 2026: impressions in AI Overviews and AI Mode, by page, country and device. For the other assistants, measurement remains manual but doable: every month, ask them the questions your clients ask, and record who gets cited. In parallel, track the traffic arriving from assistants in your analytics tool.
We track these citations and Google rankings in our clients' cockpits, alongside the other acquisition indicators. An indicator that is not tracked does not improve. The monthly reading also shows the questions where a competitor is cited instead of you: that is your next page to write.
The rest is consistency. Our production rule: four new answer pages per month, written and signed by a human, never published raw from an AI. A rhythm held over time is worth more than twenty pages published at once and then 6 months of silence.