By Mehdi HassanDoes schema markup help AI citations?
No, not in any way that's been measured. The best controlled evidence we have, Ahrefs' 2026 test of adding structured data to pages, found no significant lift in AI citations. Google's own documentation says pages need no special markup to appear in its AI features.
That doesn't make schema useless. It makes it the wrong lever for AI visibility. Our view, after building sites in Next.js, Framer and Webflow, is that schema is hygiene you set up once in your templates, while citations are won by rendering, crawler access and pages that answer questions clearly. The full build picture is in our guide to building an AI-search-ready website.
What did the Ahrefs schema test actually show?
Ahrefs ran a controlled test in 2026: it added structured data to a set of pages and compared their AI citations with similar pages left alone. The result was no significant lift. A controlled test is the right design for this question, because it isolates schema from everything else that differs between websites.
That's the key difference from most claims you'll read. A finding that cited pages often have schema tells you about the kind of site that gets cited, not about what schema does. Sites with clean structured data also tend to have strong content, good technical SEO, more links and more brand mentions. Any of those could be doing the work.
Type of evidence | What it can tell you | How much weight to give it |
|---|---|---|
Controlled test (Ahrefs, 2026) | Whether adding schema changes citations, other things equal | High: it isolates the variable. Result: no significant lift |
Correlation study | Whether cited pages tend to have schema | Low for causation: well-run sites do many things at once |
Platform guidance (Google) | What the engine says it needs | High for Google: no special markup required for AI features |
Vendor case study | That one site changed many things and traffic moved | Very low: no control, and schema is rarely the only change |
Why doesn't structured data move AI search?
Structured data doesn't move AI search because answer engines cite passages, and schema doesn't change the passages. An engine retrieves pages for a query, reads their text and quotes the parts that best answer it. JSON-LD tells a machine that a page is an Article by a named author, but it doesn't make the paragraph clearer or more trustworthy.
- Language models read text well. The original reason for schema was that machines struggled to parse pages. Modern models parse plain HTML fine.
- Retrieval uses the index, not the markup. Google's AI Overviews draw on its regular index. ChatGPT and Perplexity fetch and read pages directly.
- Schema can't fix a weak page. If the answer isn't on the page, or it's buried below three paragraphs of preamble, markup won't surface it. See answer-first page structure.
- Schema can't fix an invisible page. If your content is rendered by JavaScript that crawlers don't run, the markup is often missing too. Read about server-side rendering for AI crawlers.
What is schema markup still good for?
Schema markup is still good for rich results in classic search, for telling search engines exactly which entity you are, and for keeping facts like prices, authors and dates consistent. Those benefits are real and documented, they just aren't AI citations.
Schema type | What it does | Worth it on a marketing site? |
|---|---|---|
Organization | Names your company, logo and official profiles via sameAs | Yes, on every site |
WebSite | Declares the site name Google can show in results | Yes, once on the homepage |
BreadcrumbList | Shows the page path in search results | Yes, if you have nested pages |
Article / BlogPosting | Marks author, publish and update dates | Yes, on blog templates |
Product / SoftwareApplication | Can make pricing and reviews eligible for rich results | Yes, if you have real pricing and genuine reviews |
FAQPage | Marks up question and answer pairs | Optional: Google now shows FAQ rich results for only a small set of sites |
Why Organization schema is the one we always add
Organization markup with sameAs links to your LinkedIn, Crunchbase, GitHub or other official profiles is the cleanest way to tell a search engine which company you are, especially if your name is common or shared with another business. Entity confusion is a real problem for startups called something like Atlas or Nova.
It won't make an AI engine cite you, but it reduces the chance that your brand gets mixed up with another one in a knowledge panel or an answer. It takes minutes to add once in a site-wide template, and it rarely needs to change.
Is schema markup for AI Overviews any different from schema for SEO?
No. There's no special schema for AI Overviews. Google says pages need to be indexed and eligible to show a snippet, which is the same bar as normal search, and that no additional markup is needed. Any vendor offering AI-specific schema types is selling something Google has said doesn't exist.
The same goes for ChatGPT and Perplexity. Neither has published a schema requirement for citations. What they do need is access: their crawlers must be allowed in robots.txt and not blocked by your CDN, which is covered in our guide to allowing or blocking AI crawlers. The llms.txt file sits in the same bucket as schema; our llms.txt explainer covers what the studies found there.
How should you implement JSON-LD for AI visibility?
Implement JSON-LD in your page templates, generated from the same CMS fields that render the visible page, then leave it alone. That way schema stays accurate without anyone maintaining it by hand, and your effort goes to the content engines actually quote.
Add Organization and WebSite site-wide
One block in the root layout with your legal name, logo, URL and official profiles.
Generate page schema from CMS fields
Blog posts output BlogPosting with the author and dates from the CMS. Pricing pages output product data from the same fields that render the price table.
Render it on the server
Make sure the JSON-LD is in the HTML the server returns, not injected later by a tag manager. On Next.js, render it in the page component.
Validate once per template
Test one page per template type, fix errors, and add a check to your launch QA.
Then move on
Spend the rest of the budget on clearer answers, comparison tables and FAQs that come from real buyer questions.
What's our verdict on schema and AI citations?
Schema is worth doing and not worth obsessing over. If your site has no structured data, add the basics to your templates. If it already has them, don't buy a schema project in the hope of AI citations, because the controlled evidence says it won't deliver them.
