What is structured data?
Structured data is machine-readable markup. Almost always the schema.org vocabulary expressed as JSON-LD, that states what a page's content is: an Article, an Organization, a Person, a Product, a BreadcrumbList. Instead of leaving a machine to infer meaning from prose, you declare it.
It sits in the page's head or body as a small JSON block. It changes nothing a human sees, its entire audience is software.
What it actually does
Two proven jobs. First, eligibility for rich results: breadcrumbs, article treatments, and other SERP features only appear when the matching, accurate schema is present (presence makes you eligible. It doesn't guarantee display). Second, entity understanding: it disambiguates your organization and author and helps connect them in the knowledge graph.
This estate ships Organization, Person, BlogPosting, and BreadcrumbList JSON-LD. Generated from canonical data so the markup can't drift from what's on the page. Those are the real, defensible uses.
What it does NOT do
It is not a proven AI-citation lever. There's no controlled evidence that adding schema lifts your odds of being cited in AI Overviews, ChatGPT, or Perplexity, citation tracks answer quality, not markup (see the cornerstone). And it actively backfires when misused: AggregateRating or Review schema on pages that aren't eligible for it is a documented demotion flag.
So the failure mode isn't too little schema. It's inaccurate schema bolted on to fake signals.
Which types are worth adding, and how do you test them?
Lead with the types that earn rich results or anchor your entities: Organization and Person (who you are, for the knowledge graph), BreadcrumbList (navigation in the result), Article/BlogPosting (article treatments), and Product, Recipe, or FAQPage only where the page genuinely qualifies. Skip speculative markup with no eligible display.
Then test it: Google's Rich Results Test shows which rich results a URL is eligible for, and the Schema Markup Validator catches syntax errors. Validating before you ship is how you stay clear of the inaccurate-schema failure mode, the markup either earns a real feature or it shouldn't be on the page.
How to use it well
Mark up what is genuinely on the page, accurately, using only eligible types; keep the markup in sync with the visible content; and never assert ratings, reviews, or facts the page doesn't actually carry. Generating it from the same source as the page. As we do. Is the cleanest way to guarantee it never drifts into a false claim.
Accurate schema for its real jobs, paired with genuinely good content, is part of our Search & AI visibility practice. And it pairs with shipping llms.txt as hygiene rather than as a lever.