Schema.org markup affects AI citations indirectly but reliably — it collapses the model's attribution problem by making the author, publisher, and answer text machine-readable, and it is the primary signal Google uses to render both classic rich results and the source strip in AI Overviews.
Google's structured-data introduction is explicit that JSON-LD is the recommended format, and that markup is what makes a page eligible for rich-result treatments. The AI-features documentation confirms that pages surfacing in AI Overviews follow the same eligibility rules as regular Search.
The specific types that carry the most weight for answer engines are QAPage with an acceptedAnswer and answerCount, Article with author and datePublished/dateModified, Organization with sameAs, and Person with sameAs. Together they let the model answer three questions instantly: what is the direct answer, who wrote it, who publishes it.
Schema does not by itself force a citation. It also does not overrule content quality — Google's structured-data guidelines require that markup reflect content actually visible on the page. What schema does is remove a class of failure modes: without it, the same fact can appear on your page and be attributed to a competitor whose page ships QAPage JSON-LD.
Validate every deployment with the Google Rich Results Test. If a page fails validation, treat it as if the markup were missing.