Being cited by an AI is decided in the page, not in the tag

The question arrives just like that: “how do we get the site listed on ChatGPT?”. It assumes a desk, a submission, a setting. There is none. You do not register with a generative engine: it reads what is published, and repeats what it can repeat without risk.

Everything hangs on “without risk”. A machine that cites stakes its own credibility. It therefore favours what is verifiable, dated and attributed — and sets aside anything that forces it to interpret.

The standalone sentence test

Take any page on the site. Find the sentence that answers directly the question the heading poses. Cut it from its context and read it again.

If it remains accurate, it is quotable. If it begins with “which is why”, “in that case” or “as we saw above”, it is not: it depends on a paragraph the machine will not carry along with it.

This flaw is pervasive in corporate writing, because commercial rhetoric builds tension: it sets up a problem, worsens it, then delivers the answer at the end of the section. A machine cuts before the end.

A statement that needs its run-up to be true is not a statement. It is a conclusion.

What prevents a quotation

• The orphan figure. “70% of companies…” without a source will not be repeated, or will be repeated with a caveat that works against you.
• The missing date. On a regulatory or technological subject, an undated page is treated as out of date by default.
• The unverifiable claim. “Leader”, “benchmark”, “recognised expert”: a machine can neither confirm nor cite it.
• Text that exists only on screen. Content produced by script after load is not guaranteed to be read.
• The page that answers three questions at once. It will be a candidate for none of them.

What triggers one

Conversely, four moves change the yield of an existing page without a full rewrite.

• Put the question in the heading and answer it beneath. Not a hook: the question as actually typed, and the answer in the two sentences that follow.
• Attribute every figure. Source name, year. One line is enough, and it is worth more than the number itself.
• Date it visibly. Publication date, and date of last review when the subject moves.
• Name the author. An identifiable person, not an anonymous editorial desk. That is what separates a considered view from mass-produced content.

Before and after, on an ordinary case

An IT provider’s page announces: “Our audits quickly identify the weaknesses in your information system and significantly improve your security.” Nothing in it is false. Nothing in it is quotable either: not what is measured, not in what time, not what “significantly” covers.

The same thing written to be quoted: “An infrastructure audit covers six areas — network, workstations, backups, access rights, security, obsolescence. It produces a report ranked by criticality. Allow two to four weeks depending on the size of the estate.”

The second version is no longer. It is more exposed: it says things that can be held against it. That is exactly what makes it quotable.

The limit of the exercise

None of these corrections manufactures a position. They make quotable what deserves to be. A company with nothing particular to advance will at best be cited as one among others — which is worthless, because a written answer does not distinguish between equivalents, it keeps one.

These corrections assume a prior question has been settled, and it is the only one that really matters: should they come before or after repairing the technical groundwork.

This work does not excuse you from understanding what actually changes: a generative engine does not rank pages, it repeats statements from them.

The useful work therefore starts upstream: knowing what you assert that others do not. The formatting comes afterwards, and it is the easy part.

Common questions