Does Schema Markup Help AI Citations? What the Evidence Shows

Schema markup is code that labels the facts on a page for machines, usually written as JSON-LD in the page head. A lot of AI search advice says that adding it gets you cited in ChatGPT, Perplexity or Google AI Overviews.

The primary sources don't support that. Google says structured data isn't required for its generative AI features. The best-designed test of the idea, a matched-control study by Ahrefs, found no major uplift on any platform it measured, and a significant drop on AI Overviews.

That is not the same as saying schema is useless. It still has documented uses in regular Google Search, and Google says existing markup does no harm. The honest position is narrower: schema is not a lever for AI answers.

What this means for you: if a vendor or agency lists schema as a quick win for AI visibility, ask for the evidence. Then spend that time on the checks below, which have documented mechanisms behind them.

Google's documentation for site owners is direct on this point, as of October 2026.

• Eligibility is ordinary Search eligibility. Google's AI features page says a page must be "indexed and eligible to be shown in Google Search with a snippet" to appear as a supporting link in AI Overviews or AI Mode. It adds: "There are no additional technical requirements."
• No special optimization exists. The same page says there are no additional requirements to appear in AI Overviews or AI Mode, "nor other special optimizations necessary."
• Markup is named explicitly. Google's AI optimization guide (July 2026) says you don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search, including its generative AI capabilities.
• There is no AI schema type. Google says no AI-specific schema.org type exists.

FAQ markup has also lost its main Google use. Google deprecated FAQ rich results in May 2026, removed the FAQPage documentation on June 15, 2026, and removed Search Console API support in August 2026. That kills FAQ markup as a rich result. It says nothing against writing clear questions and answers on the page.

One caveat: these are Google's statements, not experiments. They are the best evidence available for Google's own surfaces, and they cover Google only. For a measured result, look at the Ahrefs study.

Ahrefs tracked 1,885 pages that added schema and compared each one with three matched control pages that didn't, between August 2025 and March 2026. Its summary: "Adding schema produced no major uplift in citations on any platform."

The only statistically significant result in the study points the wrong way for anyone selling schema as an AI tactic. The small increases on AI Mode and ChatGPT were not presented by Ahrefs as a meaningful lift.

The design matters here. A matched-control test compares pages that changed with similar pages that didn't, over the same period. That rules out a lot of the noise that sinks before-and-after anecdotes, such as an algorithm update or a seasonal swing.

It is still one study by one SEO tool vendor, on one population of pages. For a fuller breakdown of the method and its limits, see our digest of the Ahrefs schema study.

What this means for you: the result fits a wider pattern. The C-SEO Bench study (NeurIPS 2025) found most content tactics aimed at AI citation were largely ineffective and frequently harmful. Schema joins a long list of on-page edits with no credible evidence of lift.

Ahrefs disclosed an important boundary. Every page in its dataset already had more than 100 AI Overview citations in February 2025, before any schema was added. The study measured pages AI was already citing heavily.

That leaves a different question open. Does schema help a page that AI never cites get into answers for the first time? Nobody has published a test of that. It is untested, not refuted.

This matters because many readers are in exactly that position. A mid-sized brand at zero citations is the population the study excluded.

Don't read the gap as hidden support for schema, though. Absence of a test is not evidence of an effect. The fair summary is: for cited pages, adding schema did nothing useful and possibly hurt on AI Overviews. For uncited pages, nobody knows.

What to do with that: if you're not being cited at all, start with the causes that have documented mechanisms. A page AI crawlers can't fetch, or a page Google won't show with a snippet, can't be cited regardless of its markup. If you want a baseline first, the free checker on this site shows whether AI answers name you for up to three buyer questions.

AI answers come from four stages. Knowing which stage you're working on stops you spending effort where it can't land.

1. Parametric. What the model learned in training, frozen at a cutoff. Nothing you do this quarter changes it.
2. Access. Whether a crawler may fetch and store your page. You and your CDN control this.
3. Retrieval. Whether your page enters the model's context for a given query. The search operator controls most of this.
4. Generation. Which retrieved pages the model uses and cites. The model decides.

Schema is usually sold as a stage 3 or 4 tactic: label your facts and the model will prefer you. The experimental evidence on that kind of tactic is mostly null or negative.

The access stage is different. Its controls work by removing blocks, and the operators document them:

• robots.txt rules are per bot. OpenAI says GPTBot governs training, OAI-SearchBot governs ChatGPT search, and each setting is independent.
• A firewall or CDN can block a crawler that robots.txt allows.
• noindex, nosnippet, max-snippet:0 and data-nosnippet keep content out of Google's AI answers.
• Vercel's crawler study found that ChatGPT's and Claude's crawlers fetch JavaScript files but don't execute them, so content that only appears after JavaScript runs may be invisible to them.

There is one positive signal for on-page content. Stating a specific fact, such as a price or a one-line definition, in text a crawler can fetch gives the engine a sentence it can quote. Our comparison of structured data and plain-text facts covers the difference in more detail.

Work from the gates with documented mechanisms toward the ones without. None of these steps guarantees a citation. They remove the reasons you can't be cited.

1. Confirm Google eligibility. Check key pages for noindex, nosnippet, max-snippet and data-nosnippet. Templates often carry these by accident. Confirm index status in Search Console.
2. Check robots.txt bot by bot. Make sure OAI-SearchBot, PerplexityBot and Claude-SearchBot are allowed if you want to appear in those engines' search answers. Blocking GPTBot only affects training.
3. Check the firewall. A robots.txt allow means nothing if your CDN returns a 403 to the crawler. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.
4. Check rendering. Compare the raw HTML with the rendered page. Prices, product names and key claims that appear only after JavaScript runs may never reach non-Google crawlers. The same logic covers any markup a script injects after load.
5. State your key facts in plain text. Put prices, plan names, locations and one-line definitions in the HTML body, not only in markup or images.
6. Keep existing markup accurate. If you already have schema, make sure it matches the visible page. A JSON-LD checker can flag markup that disagrees with what the page says.
7. Add new markup for documented Search features only. If Google documents a feature you want, use the markup it asks for. Don't add schema in the expectation of AI citations.

• Calling schema a quick win for AI. No study supports that framing, and the strongest one found a significant drop on AI Overviews.
• Adding FAQPage markup for AI answers. Google has deprecated the rich result and removed the documentation. The markup now has no documented Google use.
• Ripping out existing schema. This overcorrects. Google says existing markup does no harm, and no evidence says removal helps. Our answer on removing schema markup covers the trade-off.
• Letting markup and page disagree. If your JSON-LD says one price and the pricing page says another, you're publishing two versions of the same fact. Fix the contradiction at the source.
• Skipping the access checks. Teams often spend weeks on markup while a firewall rule or a stray noindex keeps a page out entirely.
• Trusting an "AI readiness" score that rewards schema. If a tool scores you higher for adding markup, it's scoring a belief, not a measured effect. See how evidence grades the signals behind AI Overviews.
• Treating one study as final. Ahrefs' test is the best available, not the last word. Re-check when new matched tests are published.