Here's the answer, and it's a test you can run tonight: could a machine, or a hurried stranger, answer who you are, what you do, for whom, at what price, and why anyone should believe you, from your pages' text alone? Most brand sites fail. Vibes-only homepages, facts trapped in images, critical answers buried in marketing prose. When AI describes you wrong, the error usually started on your own site.
Four page types earn most owned citations: a substantive FAQ answering real buyer questions plainly, not five marketing softballs. An evidence page banking your press, credentials, and results. A candor page that says who you're for and who you're not for, because models reward that honesty with trust. And the definitional resources your content engine builds. Build these four before touching anything cosmetic.
Write everything for extraction: the direct answer in the first two sentences, headings that ask the buyer's question, facts in text rather than images, dates on everything.
Then the boring win: schema markup, an afternoon of structured data that makes your facts machine-certain, your organization, your founder, your services with honest pricing, your FAQ. One rule: schema declares facts, it doesn't create them. Marking up reviews you don't have teaches every machine to distrust your markup entirely.
And the technical floor: make sure AI crawlers aren't blocked by accident, and treat llms.txt honestly, as a twenty-minute hedge, not a strategy, since no major provider confirms using it.
The full owned-layer rebuild is Volume 5 of our Primacy Playbook.
Your site is testimony. Write it for the witness stand.