Structured Data

Prose is ambiguous to a machine; markup is not. Google describes structured data as a standardized format for providing information about a page and classifying its content, and uses it both to understand the page and to gather facts about the people, books or companies it describes^[1]. The vocabulary is shared: schema.org was founded by Google, Microsoft, Yahoo and Yandex and is developed through an open community process^[3].

The visible payoff is rich results. Google cites site owners’ own measurements, such as Rotten Tomatoes recording a 25% higher click-through rate on pages with structured data than on pages without^[1]. Those are case studies Google chose, not a forecast, and Google does not guarantee that correct markup produces a rich result at all^[2].

For AI answers its role is narrower than often claimed. Google says no special schema.org markup is needed to appear in AI Overviews or AI Mode, and asks that structured data match the visible text^[5]. Microsoft’s guidance for Copilot treats schema as a way to label content as a product, review, FAQ or event so that machines can interpret it with confidence^[6]. Both point the same way: markup helps a system read a page correctly; it does not make a weak page worth citing.

The format matters less than the match. Google supports JSON-LD, Microdata and RDFa and recommends JSON-LD^[2], whose current version, JSON-LD 1.1, has been a W3C Recommendation since 2020^[4]. What decides eligibility is that the markup is true to the page: no content the reader cannot see, nothing misleading, and the most specific type that applies^[2].

Per page and per type: is the markup valid, are the required properties present, and does every value match what the page shows. The Rich Results Test checks one URL during development; Search Console’s rich result reports follow validity across the site after release, which matters because templates and serving can break markup that passed once^[1].

Parity is the check that validators skip. Compare the price, availability, rating, author and dates in the markup with the rendered page; a mismatch is a quality problem even when the syntax is valid. A structured data manual action costs a page its eligibility for rich results, not its ranking in web search^[2].

Count what passes the rules, not what is present. QueryRev’s study applies that to another machine-readable file: of the 848 llms.txt files found, 87.1% had at least one section that lists links and 82.5% a summary^[7]. A script tag of JSON-LD is the same: present is not the same as usable.

Sample data: an invented site, to show the method.

A shop marks up its oak bed frame as a Product with an Offer. A sale ends and the page returns to the full price, but the price in the JSON-LD comes from a cache that nobody cleared. The page shows one price and tells machines another.

The Rich Results Test still passes, because the markup is valid. A parity check that reads the rendered price beside the markup finds it, and the lasting fix is to generate both from the same field.