Structured Data for AI Search: What Helps and What Doesn't

Which schema types actually matter, and why?

Four types cover most of what a typical business site needs:

• Organization — names your company as an entity: legal name, logo, official social profiles (via sameAs), contact details. This is the type most directly tied to how clearly an engine or a knowledge panel can identify who you are, separate from what any one page says about you.
• Product — price, availability, rating, and specification facts for something you sell. This is what makes a product page eligible for Google's product rich results.
• Article — publish date, author, headline for content pages. Lets Google show the right date and byline in search, and gives any system parsing the page a clean, structured timestamp instead of having to guess one from the text.
• FAQPage — question/answer pairs matched to visible on-page content. Eligible for FAQ rich results in Google Search, and a clean, literal match between the markup and what's actually on the page.

Schema.org itself — the vocabulary all four of these come from — is a shared, open specification maintained through a W3C community process, originally backed by Google, Microsoft, Yahoo and Yandex. As of 2024 it's used by over 45 million domains. It's a shared machine-readable vocabulary, not a search-engine-specific trick, and nothing in schema.org's own documentation claims special treatment for AI systems versus traditional search.

What did a controlled test actually find?

Ahrefs ran the closest thing to a controlled experiment on this question that's been published (Louise Linehan, 11 May 2026). The team tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026, matched each one against control pages from different domains with similar citation levels that never added schema — 4,000 control pages in total — then measured the citation change before and after, using a matched difference-in-differences design and four separate statistical tests to isolate schema's effect from broader platform trends.

The result: Google AI Mode citations moved +2.4% and ChatGPT +2.2% — changes Ahrefs itself calls statistically indistinguishable from zero. Google AI Overviews moved −4.6%, which Ahrefs reports as a small but statistically significant decline. Ahrefs' own summary: "Adding schema produced no major uplift in citations on any platform." In plain terms: adding schema to these pages didn't produce a detectable citation gain anywhere, and on one surface the pages that added it saw a small dip.

What this test does and doesn't tell you: every page in the dataset already had 100+ AI Overview citations before schema was added. The test measures whether schema moves the needle for pages already winning — it says nothing about whether schema helps a page that isn't cited yet get discovered at all. That's a different, still-open question the study doesn't answer.

A separate, earlier Ahrefs analysis (Louise Linehan, "AI Search Overlap," 11 August 2025, 15,000 long-tail queries) looked at a related but different question — how much AI citations overlap with Google's own top 10 rankings (about 12% on average, with Perplexity the outlier at 28.6%) — and didn't test schema specifically. The two studies are often conflated in secondary coverage; they measure different things.

What does Google say?

Directly, and without hedging. Google's generative-AI-features guide (updated 10 July 2026) states: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." Its dedicated AI-features-and-your-website page (updated 10 December 2025) repeats the same position: "You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add." Eligibility for AI Overviews and AI Mode support links requires only what ordinary Search already requires — the page indexed and eligible to appear with a snippet.

Google is explicit that this guidance is scoped to its own features. It says nothing about what other engines — ChatGPT, Perplexity, Claude, Gemini — do with structured data, and no comparable first-party statement from those companies was found for this post. Treat "no special markup needed" as Google's position on Google's features, not a blanket claim about every engine.

Where does schema still help, then?

Where it has always helped, on classic SEO and clarity grounds — and that classic value doesn't disappear just because it isn't a proven AI-citation lever:

• Rich-results eligibility. Correct Product, FAQPage, and Article markup is what makes a page eligible for Google's enhanced search listings (star ratings, FAQ accordions, image carousels) at all. Google's own documentation cites a Nestlé case study showing pages appearing as rich results had an 82% higher click-through rate than pages that didn't — a real, if Google-specific and single-case, number.
• Entity clarity. Organization schema with correct sameAs links to your official Wikipedia, Crunchbase, LinkedIn and Wikidata profiles is one of the clearest, lowest-cost ways to tell any machine reader which real-world company a page is actually about — see our guide to entity SEO for the full method.
• Reducing ambiguity for common names. If your brand or product shares a name with something else, schema is one of the few direct levers you have to disambiguate — not a ranking trick, just removing a source of confusion.

How do you validate your own markup?

1. Open Google's Rich Results Test (search.google.com/test/rich-results) and enter either a live URL or paste raw HTML/JSON-LD.
2. Choose smartphone or desktop view and run the test.
3. Read the output in three parts: eligible rich result types (what Google says it can show), errors (will block the rich result entirely), and warnings (won't block it, but means some recommended fields are missing).
4. Fix errors first. Google's own guidance on required vs. recommended properties is direct: "it is more important to supply fewer but complete and accurate recommended properties rather than trying to provide every possible recommended property with less complete, badly-formed, or inaccurate data." A smaller, fully correct markup beats a large, partially broken one.
5. Re-test after every template change. A site-wide theme or CMS update can silently break markup across every page that uses it — a quick recheck after any such change catches this before it compounds.

The results panel splits into exactly those three parts — eligible types, errors, warnings — for whichever URL you just ran, so running it on your own page takes under a minute.

Checklist before you ship schema

• Organization schema present sitewide, with a correct sameAs list pointing to your real, live profiles.
• Product schema present on every page that sells something, with price and availability kept current (stale pricing in schema is worse than no schema).
• Article schema present on blog/content pages, with a real publish date and author.
• FAQPage schema present only where the markup's Q&A pairs match visible, on-page text — not hidden or fabricated to game a rich result.
• Every template re-validated in the Rich Results Test after a site or theme change.
• No claim made internally or externally that schema is a proven AI-citation lever — the current evidence doesn't support that claim.
• Schema validation folded into the broader technical AI-readiness checklist for your site, not treated as a standalone fix.

Tools to use

• Google's free Rich Results Test (search.google.com/test/rich-results) — no login required.
• Schema.org's own documentation for the correct property names and types per schema (schema.org).
• Your CMS's built-in schema fields, where available, to avoid hand-coding JSON-LD per page.

Common mistakes

• Treating schema as an AI-ranking trick rather than what the evidence supports: machine-readable clarity with a real but narrower, Google-specific rich-results payoff.
• Marking up facts that don't match the visible page content (price in schema doesn't match the price shown, FAQ answers in markup that don't appear on the page) — this risks a manual action from Google, not just a wasted effort.
• Adding every schema type available instead of a few done correctly, against Google's own stated preference for complete, accurate core properties over broad, sloppy coverage.
• Never re-validating after a CMS or theme update, so broken markup goes unnoticed for months.

FAQ

Does adding schema markup get my page cited more by ChatGPT or Google's AI features? The best controlled evidence available — Ahrefs' matched test of 1,885 pages — found no meaningful citation lift. Google also states directly that no special schema markup is needed for its AI features. Treat schema as clarity, not a proven citation lever.

Is schema markup a waste of time, then? No. It's still the basis for Google's classic rich-results eligibility and for making your brand entity unambiguous to any machine reader. Those are real, if narrower, benefits than "get cited by AI."

Which schema types should a small team prioritise? Organization (sitewide), Product (on sales pages), Article (on content pages), and FAQPage (only where it matches real on-page Q&A). Most sites get more value from doing these four correctly than adding many types badly.

How do I check my schema is actually valid? Google's free Rich Results Test — paste a URL or raw code, no login needed — shows eligible rich result types plus any errors or warnings.

What to do next

1. Run your most important pages through Google's Rich Results Test and fix any errors first, warnings second.
2. Add or correct Organization schema with real sameAs links to your actual public profiles.
3. Stop describing schema internally as an AI-citation lever — describe it as clarity and rich-results eligibility, which is what the evidence actually supports.

ContextOwl sells one-time AI visibility reports. The FAQ Schema (JSON-LD) Generator is a free tool that builds valid FAQPage markup from your own questions and answers.