Structured data for AI: what it is and why it matters
Structured data is a machine-readable description of your content in JSON-LD format that helps AI tools like ChatGPT, Perplexity, and Google AI correctly identify what a page contains. For AI search it plays two roles: it makes facts easier to pull out (price, author, date, the steps in a procedure) and it helps classify the type of content correctly. The available tests and hands-on experience from 2025–2026 suggest that valid structured data can improve the odds that AI tools understand your content correctly and use it — but it is no guarantee of a citation.
This guide covers which types to implement in 2026, how to write them properly in JSON-LD, what changed when FAQ rich results ended, and how to validate them. It adds the technical layer to the broader content for the AI era.
Key structured data types for AI in 2026
For most sites it is practical to start with these six types and add others as the content calls for them:
The core types for AI search
- Article for an article or blog post; author, publication and update dates, main image.
- Product for a product in an online store; price, availability, ratings — all of it has to match the visible content.
- Organization at the site level, to identify the brand; logo, social profiles (sameAs), contact details.
- BreadcrumbList breadcrumb navigation across every page in the hierarchy — it helps AI understand the site structure.
- HowTo only where the page contains a genuine step-by-step procedure.
- FAQPage only where the FAQ is visible to users too (careful: no more rich results in Google — see below).
For specific content, add Recipe (a recipe), VideoObject (a video), LocalBusiness (a physical location), Event (an event), or DefinedTerm (a term definition).
JSON-LD as the preferred format
JSON-LD is the standard choice for structured data in 2026. Here is why:
- Google and Bing support JSON-LD as a matter of course, and for AI tools like ChatGPT, Perplexity, or Claude the format can make your content easier to read by machine, assuming they can reach the page.
- It lives in a separate script (
<script type="application/ld+json">), apart from the HTML — easier to maintain and less likely to break when the template changes. - Multiple structured data blocks on one page can go into one or several
scriptblocks with no effect on the layout.
Microdata and RDFa work too, but the recommended and widely used route is JSON-LD.
A short, simplified example of what JSON-LD looks like for the Article type (in production you would usually add mainEntityOfPage, publisher, and other fields depending on the page type):
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Article",
"headline": "Article headline",
"author": { "@type": "Person", "name": "Author name" },
"datePublished": "2026-06-07",
"dateModified": "2026-06-07",
"image": "https://example.com/og/article.png"
}
</script>You can have several such blocks for different types (Article, BreadcrumbList, Organization) in the head or at the end of the body.
Heads-up: FAQ rich results in Google are no longer a win in 2026
This is a significant change worth knowing about:
What this means in practice: do not roll FAQPage out across the board for the sake of the SERP. Implement it only where the FAQ is genuinely visible to users and where it helps them navigate the content. For AI citations, what helps most is that the answers are short and to the point.
Structured data must match the text
This is Google’s core rule: structured data has to match exactly what the user actually sees. Adding values that are not on the page is spam and can lead to a manual action.
- For Product that covers price, availability, and ratings.
- For FAQPage, the questions and answers.
- For Article, the author and the publication and update dates.
- For Organization, contact details and social profiles.
How to validate structured data
Before you ship and after every substantial change, use two tools side by side:
| Aspekt | validator.schema.org | Google Rich Results Test |
|---|---|---|
| What it does | Stricter check of Schema.org compliance | Confirms rich result eligibility in Google |
| What it catches | Structural errors, properties that don't exist | Missing required fields for rich results |
| When to use it | Always — the baseline validity check | When you're targeting a rich result in Google |
| AI check | Doesn't cover AI tools | Doesn't cover AI tools |
Once it is live and validated, it also pays to ask ChatGPT or Perplexity directly about the content of the page and watch whether they mention it, and which values they pull from it. No validator will confirm that AI is going to start citing you. Treat a manual query in ChatGPT or Perplexity as a rough check only — and only in modes that have access to the web or a current index.
Common mistakes
Blanket FAQPage for the sake of rich results
Google FAQ rich results are essentially over for ordinary sites in 2026; FAQPage no longer helps with how you appear in the results.
Fix: Implement it only where the FAQ is visible.
Structured data that doesn't match the text
Inaccurate or padded values are spam — you’re risking a manual action.
Fix: It always has to match the visible content exactly.
Relying on schema instead of content
Without solid, credible content, structured data on its own will not win you a citation.
Fix: Structured data is an aid, not a substitute.
Not validating after you ship
An error in your JSON-LD can invalidate a given structured data block or keep it from being processed correctly.
Fix: Run both validator.schema.org and Rich Results Test.
What’s next: put structured data in context
Structured data is the technical layer that helps AI understand your content. Without solid content and credibility it is not enough on its own. For the full picture of AI visibility, see the practical SEO for AI checklist; credibility is covered in E-E-A-T for AI.
