How it is implemented
The main format today is JSON-LD, a separate code block in the page markup. Popular types: Article for articles, Person for authors, Organization for companies, FAQPage for question blocks, BreadcrumbList for breadcrumbs.
What this means in practice
Markup provides rich snippets in search results, but that is not the main point. A typical mistake is adding independent pieces of markup where Article knows nothing about the author, and the author is not linked to the organization. Entities must be assembled into a single graph and connected via @id, then "this article was written by this person from this company" becomes an explicit fact.
Connection to AI models' responses
The model works with text fragments and evaluates source reliability. A connected graph with an author and an organization gives it verifiable data about who is behind the material. Ready markup should be checked via the schema.org validator and Rich Results Test: a missing comma makes the block invisible.