What the structured data check tells you
Structured data is markup that describes a page in a vocabulary machines understand: this page is a product with this price, this article was published on this date, this business opens at these hours. Search engines, shopping tools and assistants read it. The structured data validator reads every item on a page in all three common syntaxes, checks it against the schema.org vocabulary and against the properties search engines ask for, and tests the page's hreflang language alternates while it is there.
What we check
- JSON-LD. Every
application/ld+jsonblock, read as strict JSON. A syntax error is reported with the block number, line and column so you can find it. Blocks that use@graphare split into their items, and references by@idare followed one level. A block without a schema.org context is flagged, because its terms then mean nothing. - Microdata. Items marked with
itemscopeanditemtype, theiritempropvalues read by the HTML standard's rules, including properties pulled in withitemref. - RDFa Lite. Items marked with
vocab,typeofandproperty, plus theschema:prefix. - The vocabulary. Each type and property is looked up in the schema.org vocabulary we keep on our side. Misspelt types, unknown properties and properties used on a type that doesn't define them are warnings.
- Our rules. For the most used types we check the properties search engine documentation lists as required and recommended, nested ones such as an offer's price, and the kind of value expected: URLs, ISO 8601 dates and durations, numbers and ISO 4217 currency codes.
- hreflang. Alternates from link tags in the page, Link headers and the site's sitemap. Codes must be a valid language with an optional script and region, or x-default. Each alternate is fetched once and must answer 200, be canonical to itself and link back. We also look for the page listing itself, an x-default, one code pointing to several pages and relative addresses.
Why the syntax and the vocabulary both matter
Structured data fails in two different ways. A syntax problem, such as a trailing comma in JSON-LD or an itemprop outside any item in microdata, means the markup can't be read at all, so nothing in it counts. A vocabulary problem means the markup is read but says something schema.org doesn't define, so readers skip that part. Our rules add a third layer: properties that are valid but missing, which search engine documentation asks for before it uses an item. Fixing them in that order, syntax first, gives the quickest improvement, and the result is laid out the same way so you can work from the top down.
Reading the result
JSON syntax errors come first, since they hide a whole block. Then each item gets a card with its types,
syntax and verdict: valid, warnings or errors. Problems name the property they're about, such as
offers.price, and say what's expected without repeating your content. The hreflang table lists each
alternate with where it was found and three checks: answers, canonical to itself and links back.
The check has no score. Required properties and invalid JSON are failures, because search engines skip those items; recommended properties and unknown terms are warnings. Search engines decide whether to show rich results, and the result says so rather than promising them.
What the check doesn't do
We read the page as served, with an honest user agent, and don't run JavaScript, so structured data added by scripts isn't seen. We follow at most two redirects to reach the page. We never fetch schema.org or any other context document while checking; the vocabulary we use is a copy published under the Creative Commons Attribution-ShareAlike licence, credited on every result. When an alternate can't be fetched the result is marked partial and the alternate shows as not checked.