Structured data: which types still work

What Google does with it, according to its documentation

Google recommends JSON-LD: a script block inside the page that visitors never see. The vocabulary is usually schema.org, but Google's introduction says its own documentation, not schema.org's, is definitive for how Google Search behaves.

It promises nothing in return. Its general guidelines say so plainly: even correct markup does not guarantee that the data will show up in results.

The same guidelines set the rule that matters most. Structured data must be a true representation of the page content, and must not describe anything readers cannot see. Misleading markup can lead to a manual action by Google, which makes the page ineligible for rich results.

The types that still do something

The list of types Google supports has grown shorter. For most sites, five types remain useful.

• Organization: the company's name, logo, address and identifiers. Google may use them in the panel that presents an organization.
• WebSite, on the home page: it sets the site name Google shows above each result. Its other use, the search box under the result, has not been shown since November 2024.
• BreadcrumbList: the breadcrumb trail, which shows where the page sits in the site hierarchy.
• Article: the headline, dates and author of an article, a news story or a blog post.
• Product: the price, availability and reviews of a product. Its values change with the catalog, and the block has to change with them.

FAQ markup no longer produces anything: Google's documentation changelog announced that FAQ results would stop appearing from May 7, 2026, then removed the documentation in June. Google says markup it no longer uses causes no problems. Removing it is not urgent, and adding it does nothing for Google Search.

The blocks nobody rereads

A structured data block is displayed nowhere. Nobody rereads it while browsing the site, and it goes stale with nothing to show it: a price changed on the page but not in the block, a replaced logo, an address that moved.

The format is strict. One extra comma makes the block unreadable, and a search engine drops it entirely, while the page shows no error at all.

Our own site ran into this. It exists in French and in English, and every visible string goes through two dictionaries checked against each other. The structured data blocks did not. Two names, the service's and its offer catalog's, were written in French inside a block declared as English, on the two pages that present the offer. No reading of the English page could have revealed it.

Since then, a check compares each page with its twin in the other language, field by field, and blocks publication when a string is identical in both. Names that do not translate, such as the brand, come from the same source as the page instead of a hand-written list.

These blocks also say who publishes the site and who wrote an article. Our guide to E-E-A-T explains what those trust signals cover.

What AI assistants read from it

For its generated answers (AI Overviews, AI Mode), Google states that no special schema.org structured data is required, and no AI-specific file either. Among the basic practices it lists is a condition you have already met here: structured data that matches the visible text on the page.

For other assistants, we promise nothing. A block consistent with the page gives them one more fact; a block that contradicts it gives them two versions of the same fact. Our guide to the llms.txt file covers another thing sites offer to assistants.

What the audit checks, and why

The audit reads the blocks on every page it analyzes, in the page as the browser renders it.

• Unreadable blocks. A block that fails to parse is reported, since a search engine ignores it entirely.
• The types declared on each page, and whether they fit what the page contains: a product page with no Product, an article with no Article.
• The fields of each type: those without which the block is useless, such as an Organization's name and URL, or a Product's image and offer, and those that complete it, such as a logo or a brand. For a product sold in variants, a field carried by the variants counts.
• Pages with no structured data at all, listed in the page-by-page analysis.
• The country and currency the blocks declare, which are among the clues to the country of the site's publisher.

These findings feed into the assessment of the pillar. No single rule in the scoring deducts points for missing structured data on its own: a site without it is still readable by search engines.

What is quick to fix, and what takes a project

A missing field or an unreadable block is fixed in the template that writes it, once for every page built on that template.

It becomes a project when the blocks are written separately from the data they describe: by hand, or by a plugin nobody checks. They then need to be generated from the same source as the page, so that the displayed price and the declared price can no longer drift apart.

What this check doesn't tell you

By Quentin Mathis, Z29K · updated September 28, 2026