Structured data for AI search: what helps you get cited

Structured data is a block of code, usually JSON-LD, that tells machines what a page is: who wrote it, what it covers, and which questions it answers. Google says its AI features need no special markup. Bing says schema helps its models understand content.

Most of what's written about structured data and AI blurs two separate questions: what markup does in Google's classic results, and what it does, or doesn't do, inside a generative engine. To pull them apart, we read what Google and Microsoft have actually published and measured our own blog. All 20 pages on our blog (10 posts, each with an English version) carry the same markup: Organization, Article, BreadcrumbList, and FAQPage. From July 17 to October 17, 2026, the 16 with data earned 9,191 Google impressions, and Search Console recorded zero rich-result appearances across the whole domain. Below: what structured data is, what each search engine says about using it for AI, which types are worth the effort, and the actual JSON-LD from one of our posts. It's the technical layer of GEO; if the term is new to you, start with what GEO is.

What is structured data?

Structured data is a machine-readable description of a page, written in a shared vocabulary called schema.org. Google defines it as a standardized format for providing information about a page and classifying its content. Instead of leaving a search engine to infer that a page is an article by a specific person at a specific company, the markup states it in fixed fields: type, author, date, organization, image. Google accepts three formats: JSON-LD, Microdata, and RDFa. It recommends JSON-LD wherever a site's setup allows, because it sits in its own <script> block and doesn't touch the visible HTML. That's the format we use across our site and the one in the example further down. Markup doesn't replace the text on the page. It restates it in a form a program can read without guessing.

Does AI read structured data?

It depends on the engine, and only two have said anything on the record. Google Search Central's page on AI features, last updated December 10, 2025, says there's no special schema.org structured data you need to add to appear in AI Overviews or AI Mode, and no new machine-readable files either. What it does ask is that your markup match the visible text. Microsoft adds a nuance: as Search Engine Land reported in March 2025, Fabrice Canel, then a principal product manager at Bing, said at SMX Munich that schema markup helps Microsoft's large language models understand content, and later confirmed it on LinkedIn. We found no public documentation from OpenAI or Perplexity on whether they use markup when choosing sources, as of October 20, 2026. So the claim that schema gets you cited in ChatGPT is, for now, an opinion.

Which structured data types matter for AI?

The ones that say who is speaking and about what, because those are what let an engine attribute an answer: Organization, Person, Article, and BreadcrumbList, plus FAQPage when the page genuinely shows questions and answers. Organization names the company behind the page, with its site, logo, and profiles. Person identifies the author and ties them to that company. Article dates the piece and links it to both. BreadcrumbList places the page within the site. Other types only make sense when the page actually is that thing: Product and Review on a store, Event on a calendar, LocalBusiness for a business with premises. And some types Google no longer shows at all. HowTo rich results were retired in September 2023, and in June 2025 Google dropped seven more from its results, including Course Info, Claim Review, and Estimated Salary, noting the change doesn't affect rankings.

Is FAQ schema still worth it?

Not for the expandable FAQ result in Google, in almost every case: since August 8, 2023, Google shows that rich result only for well-known, authoritative government and health websites. Our blog bears this out. All 20 pages carry FAQPage with three to seven questions each, and across 9,191 impressions in three months, Search Console shows no rich-result appearances. We keep it anyway, for three reasons you can check. In the same announcement, Google said unused structured data causes no problems for Search. The questions in our markup are the same ones on the page, so nothing is hidden. And Bing has said schema helps its models. Today, FAQ schema is an orderly description of what the page already answers.

What does the JSON-LD on an Aurora post look like?

It's a single block with an @graph of four nodes linked by @id identifiers, so the organization is declared once and the article and author point back to it. Here's the markup from our post on Google AI Mode, published October 8, 2026 (Spanish version), trimmed for reading: we removed the social profiles, the author details, and five of the six questions.

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "Organization",
      "@id": "https://www.auroraglobalgroup.com/#organization",
      "name": "Aurora Global Group",
      "url": "https://www.auroraglobalgroup.com/",
      "logo": { "@type": "ImageObject", "url": "https://www.auroraglobalgroup.com/assets/img/aurora-logo.png" }
    },
    {
      "@type": "Article",
      "headline": "Modo IA de Google (AI Mode) y la búsqueda sin clic",
      "datePublished": "2026-10-08",
      "dateModified": "2026-10-08",
      "inLanguage": "es",
      "author": { "@type": "Person", "@id": "https://www.auroraglobalgroup.com/sobre-aurora/#marc-ribo", "name": "Marc Ribó" },
      "publisher": { "@id": "https://www.auroraglobalgroup.com/#organization" },
      "mainEntityOfPage": "https://www.auroraglobalgroup.com/blog/modo-ia-google/"
    },
    {
      "@type": "BreadcrumbList",
      "itemListElement": [
        { "@type": "ListItem", "position": 1, "name": "Inicio", "item": "https://www.auroraglobalgroup.com/" },
        { "@type": "ListItem", "position": 2, "name": "Blog", "item": "https://www.auroraglobalgroup.com/blog/" },
        { "@type": "ListItem", "position": 3, "name": "Modo IA de Google", "item": "https://www.auroraglobalgroup.com/blog/modo-ia-google/" }
      ]
    },
    {
      "@type": "FAQPage",
      "mainEntity": [{
        "@type": "Question",
        "name": "¿Hay que hacer algo técnico para aparecer en el Modo IA?",
        "acceptedAnswer": { "@type": "Answer", "text": "No hay requisitos adicionales, según Google Search Central: la página tiene que estar indexada y ser apta para mostrarse con fragmento." }
      }]
    }
  ]
}

The part that matters most is the @id. The organization has one fixed identifier, the same on every blog post in both languages and on the homepage, and the author has another that points to the author's entry on our About page. Anyone reading the markup of any post can tell that one person at one company wrote them all, without having to infer it by comparing text.

How do you implement and test it?

Pick the types that describe what the page is, write them in JSON-LD in the <head>, and validate before you publish. In practice that's four steps. First, list your templates: blog post, service page, homepage, About page. Second, build one markup pattern per template that fills in from each page's visible data, so the title, dates, and questions always match the text. Third, validate with Google's Rich Results Test, which tells you whether a type is eligible for a rich result, and with schema.org's Schema Markup Validator, which checks any type. Fourth, watch the enhancement reports in Search Console. When we review a site in our generative engine optimization (GEO) audit, markup is one of the first things we check: an author with no profile page or a company declared three different ways shows up within an afternoon.

What mistakes make markup useless?

The most common one is marking up something the reader can't see: FAQs that aren't on the page, invented ratings, or an author with no page of their own. Google explicitly asks that markup match the visible text, and it's the only condition it sets for its AI features. The second is declaring the same company several times without an @id, with different names or logos depending on the template, so a machine sees three organizations where there's one. The third is dates that don't track reality: a dateModified that never changes when you update the article, or one that changes every day when nothing has. The fourth is using types that don't describe the page, like HowTo on an opinion piece, because an old guide recommended it. All four get fixed by removing what doesn't belong and connecting what's already there.

Frequently asked questions

Want to know whether your markup helps or gets in the way?

We review your site's structured data alongside the other signals Google, ChatGPT, and Perplexity use to choose sources, and tell you what to remove, what's missing, and what to fix first. Request a GEO audit.