Optimising a page to rank will not get it quoted inside an answer engine. The two systems want different things from your content, and most SEO advice still only addresses one of them.
A page that ranks third on Google can be completely invisible inside ChatGPT, and a page that ranks nowhere can be the one an assistant quotes. These are not the same competition, and treating generative engine optimisation as a rebrand of SEO is the most common mistake I see.
A search engine is building an ordered list of documents it thinks might satisfy a query. It rewards relevance, authority and a good enough match that a human will click through and finish the job themselves. The document is the unit.
A model composing an answer is looking for something narrower: an extractable, self-contained claim it can lift and attribute without inheriting risk. The sentence is the unit, not the document.
That single difference drives almost everything else. A paragraph that requires three earlier paragraphs of context to make sense is useless to a model even if the page ranks first, because it cannot be lifted cleanly. A specific, attributable, verifiable sentence sitting under a question-shaped heading is enormously useful even on a site with modest authority.
Answer the question in the first sentence under the heading, then support it. Inverted pyramid, not build-up. Most marketing content does the opposite and saves the claim for the end.
Make claims specific and attributable. “Cost per lead fell from $36 to $17 after reallocating placement spend” is quotable. “We dramatically improved efficiency” is not, because a model has nothing to stand behind.
Put the entity in the sentence rather than relying on pronouns and page context. A model lifting one paragraph out of your page does not carry your H1 with it.
Use question-shaped headings that match how people ask, not how marketers write. The heading is the retrieval hook.
Ship an llms.txt that states plainly what the business does, who it serves and what it has actually done. Publish complete structured data so the entity is unambiguous. Name AI crawlers explicitly in robots.txt rather than leaving it to a wildcard.
I have seen a site with all of that infrastructure in place and full schema coverage return a single session from an assistant in seventeen days. The plumbing was not the problem. Nothing citation-worthy was flowing through it. Infrastructure makes you eligible; substance is what gets you quoted.
Track citation presence as its own metric, per engine, per query set that matters to your business. Rankings and organic sessions will not show you this, and neither will any standard SEO tool. You have to ask the engines directly and log what comes back, the same way you would run a rank tracker.