Structured data

Structured data is machine-readable markup in a page's source that explains what the page says. To a machine, a number on a page is at first just a number. Structured data states that it is a price, in which currency, and which product it belongs to.

How does structured data work?

The common vocabulary is schema.org, embedded as JSON-LD in a script block in the page head. It describes types and their properties: Organization, Product, FAQPage, Article, DefinedTerm for a glossary term like this one.

What matters is the relationship between types. An organisation carrying the same identifier across every page is recognised as one entity rather than many with the same name. That linkage is the difference between decorative markup and usable information.

The rule here is strict: markup must describe what is actually visible on the page. Claims that exist only in the source count as manipulation and can get a site excluded from search features.

Why does this matter for AI visibility?

Because a model has to decide who a statement belongs to. If your site consistently declares who you are, what you offer and where you are based, a system has something solid to work from. Without it, it has to infer from prose, and inference leads to confusion with similarly named companies.

Structured data is therefore a component of Generative Engine Optimization, though not a substitute for the text. What gets quoted is sentences, not markup. The markup ensures a quoted sentence is attributed to the right author.

There is a useful side effect: recording your offering, remit and location in machine-readable form forces you to state them unambiguously first. That sharpens the pages more than it sounds like it would.

Where to start

Not with everything at once. Three types carry most of the weight.

Organization on every page, with the same identifier, name, location and contact details. That is what makes the entity exist at all.

The relevant content type: Article for posts, Product or Service for offerings, DefinedTerm for glossary entries.

BreadcrumbList for a page's place in the site structure, so a system can tell where it belongs.

Anything beyond those three is only worth doing once they ship cleanly and without contradictions, which the search vendors' own testing tools will confirm.

Common mistakes

Four patterns explain most cases where markup exists and achieves nothing.

Contradictions between pages. The company name in the imprint differs from the one in the markup, the address does not match, or two organisation entries exist with no shared identifier. To a machine those are different companies.

Markup with no visible content. Ratings, prices or questions and answers present in the source and absent from the page. This is the case that gets a site excluded from search features.

Auto-generated markup nobody checks. Many systems emit markup on their own, sometimes with placeholders or the wrong types. A look at the testing tools settles it in minutes.

And the most common mistake: markup instead of text. A page with exemplary JSON-LD and thin content does not get cited, because what gets cited is what it says. Markup attributes; it does not substitute. The same holds for Generative Engine Optimization as a whole.