No AI vendor publishes an exact ranking formula, but the pattern that keeps showing up is: assistants lean on sources that are already well-indexed, structurally clear, and topically authoritative on the specific question asked — often pulling from top organic and featured-snippet-style results rather than obscure pages. Freshness, direct phrasing, and consistent facts across multiple sources also appear to help.
Treat it like SEO's stricter cousin: rank well organically, answer one question per page in plain language, and back claims with citable data. There's no confirmed way to game it — only to remove friction between your content and the model.