Structured data is a standardized format for stating facts about a page in machine-readable form, usually as a JSON-LD block using the schema.org vocabulary.Google's introduction to structured data and schema.org's getting-started guide cover the format and the vocabulary. Its job in AI search is disambiguation: telling every machine reader in the pipeline, crawlers, knowledge graphs, retrieval systems, and the agents starting to transact, exactly what a thing is, instead of leaving them to infer it. That is the whole job. Markup clarifies; it does not persuade, rank, or substitute for the content itself.
Schema as entity glue
The highest-value markup for AI search connects your pages to the knowledge layer. Organization markup on your site, with sameAs links pointing at your Wikidata entry, knowledge panel, LinkedIn, and Crunchbase-class profiles, asserts in machine-readable form that all of those records describe one company. That is the same consistency work as Entities: Becoming a Thing the Machine Knows, expressed as data, and it is the cheapest insurance against the wrong-domain failure described there.
Run everything else through the redundancy principle: state each important fact in prose, in markup, and in feeds where feeds apply. Machine readers differ in which channel they consume, and a fact stated three ways survives any single reader ignoring one of them. The prose half of that principle is the fact-source pages from Content Architecture: Building for the Chunk; never let markup state a fact the page's text does not.
The types that earn their keep
Four types cover most brands. Organization, as above. Product with offers, carrying price, currency, and availability as data. Article with a real author, linking the content to a named person for the credibility stack in Evidence, Expertise, and Original Research. And FAQPage where genuine questions get genuine answers, with the caveat that its visible rich-result payoff in Google has narrowed over the years, so treat it as disambiguation rather than a traffic play. Beyond these, add a type only when it maps to real page content; speculative markup of everything on the page adds maintenance cost without a mechanism behind it.
What each platform actually reads
Honesty about the evidence: it is thin. Google is the one platform with clear documentation, and its position cuts both ways. Structured data feeds Google's systems and rich results, but for its generative AI features Google states that structured data is not required and there is no special schema to add for AI, only the general reasons to keep using it.The mythbusting section of Google's AI features guide, under "overfocusing on structured data". The other AI platforms publish nothing definitive about consuming schema.org markup at answer time. The defensible assessment: markup reliably helps the crawling-and-knowledge layer that AI products build on, and direct answer-time consumption is unproven. Fund it accordingly, as cheap hygiene rather than as a visibility lever.
Feeds and the agentic buyers
For commercial visibility, product feeds are ahead of page markup. Google states that Merchant Center feeds and Business Profiles help products and local businesses appear in its AI responses, which makes the feed a direct input to shopping answers rather than an indirect signal.The local business and ecommerce section of Google's AI features guide. The same machine-readable pricing, specs, and availability serve the buying agents that check out on a user's behalf: an agent completing a purchase needs data it can parse, and a price trapped in an image or a quote form is invisible to it. Commercial Visibility: Products, Comparisons, and Agentic Buying covers that arena in full.
llms.txt, without the hype
llms.txt is a proposal: a markdown file at your site root listing your most important content, so an LLM can find the good parts without crawling everything.The llms.txt proposal site, which specifies the format and collects tooling. It is an idea, not a standard: no major AI platform documents consuming it, and Google states outright that Search, including its generative features, ignores such files.Google's AI features guide, mythbusting section: no need to create machine-readable files or Markdown to appear in Google Search, which ignores them.
The cost of shipping one is an hour, so shipping one loses you nothing beyond a small maintenance duty. Just book it as an experiment with no expected return until a platform documents support, and be suspicious of anyone selling llms.txt as a visibility tactic today; Myths and Misconceptions keeps the running list of claims like that.