Structured data was originally built for search engine rich results, and answer engines inherit some of that same parsing infrastructure, so marking up entities, products, organizations, and authorship in JSON-LD gives retrieval and extraction systems a clean, unambiguous fact to pull instead of requiring them to correctly parse and interpret free-form paragraph text.
Schema markup should be treated as a clarity aid, not a ranking or citation guarantee: no controlled study has shown a specific schema type reliably increasing AI citations, and claims of fixed percentage lifts from any single markup type should be treated skeptically. Its value is reducing misinterpretation risk for the systems that do read it, not manufacturing visibility on its own.