The structured data types that change how models describe you
Start with an Organization node carrying name, url, logo, description and sameAs. Add FAQPage wherever you already answer questions in prose, and Article with a named author and dateModified on every post. Those three cover most of what an assistant needs to describe and attribute you correctly.
Organization: the identity anchor
sameAs is the field teams skip and the one that matters most. It links your site to your LinkedIn, Crunchbase, GitHub and Wikidata records, which is how a model resolves 'Geolix' from a string into a specific entity rather than a coincidence of letters.
FAQPage: free passage structure
If a page already answers questions, FAQPage markup makes each pair independently addressable. It is the cheapest structural win available, and it forces you to write question-shaped headings anyway.
Article: authorship and freshness
author with a real Person node, datePublished, dateModified, publisher. Freshness affects which of two similar sources gets quoted, and dateModified is the only machine-readable claim you have about it. An article with no author node reads as unattributable.
Validate, then re-validate after deploys
Broken JSON-LD is worse than none: the parser drops the whole block. We see valid markup shipped and then quietly broken by a template change three sprints later. Add a parse check to CI.
Want the version of this measured against your own site? Run a free audit.