Does schema markup help you get cited by AI? We checked

No. Based on the evidence we have, adding schema markup doesn't get you cited by AI search, and one large study found citations dipped slightly after sites added it. Ship the basics for classic search and leave it there.

That's an unpopular view when so many agencies are selling schema packages as AI optimization, so it's worth going through what the tests found.

What do the controlled tests show?

Three separate experiments have tried to answer this directly. All three landed in the same place.

The canary test. Researchers at Otterly made up a fact that didn't exist anywhere else online and put it only inside FAQPage JSON-LD on one page. Then they asked seven AI platforms to find it. None of them could. Google's AI Mode went a step further and invented a schema type that wasn't on the page at all.

The format isolation test. searchVIU built a page that showed a different price through each of eight delivery methods, then asked five systems what the price was. Not one of them found the price that only lived in JSON-LD. Visible text worked, and so did visible Microdata and RDFa. That suggests what matters is the text being on the page, not the markup format wrapped around it.

The falsification test. This one tells you the most. Williams-Cook put a fake address inside JSON-LD that was broken on purpose, with a made-up type and a context that doesn't resolve. ChatGPT and Perplexity returned the address anyway. Perplexity even said it found it "in the page's embedded structured data".

That result explains what's going on under the hood. If broken schema still gets returned, the system isn't reading it as schema at all. It's treating the script block as plain text, the same as everything else on the page.

What about the outcome studies?

Ahrefs compared 1,885 pages that added JSON-LD between August 2025 and March 2026 against 4,000 matched pages that didn't, using a difference-in-differences analysis. AI Overview citations fell 4.6%, and that result was statistically significant. AI Mode and ChatGPT moved about two points the other way, which is within the noise.

A separate study looked at 730 AI citations across 75 commercial queries, including professional services, and controlled for ranking position. Having schema made no significant difference. Ranking position made a huge one: a page ranked first was cited in 43% of queries, and a page ranked seventh in 5%.

What does Google say?

It says so plainly in its own documentation on AI features: there's no special schema.org structured data you need to add. Its AI optimization guide, published in May 2026, says the same thing and adds that you don't need to break content into small pieces either.

It's rare for the platform, the controlled tests and the outcome studies to line up this neatly. Normally at least one of them is murky.

So is schema worthless?

No, and this is the part that gets lost. Schema still does real work in classic search, where Google and Bing read it at crawl time and feed it into their knowledge graphs. That's what drives knowledge panel eligibility, logo association and entity disambiguation.

Classic search is also the way into AI answers. Google's AI Overviews pull from the regular search index, and a page only needs to be indexed and eligible for a snippet to be used. So schema helps indirectly, by helping you rank. It doesn't help by being read.

Where this really matters is budget. An hour spent on Organization markup is time well spent. A schema package sold to you as AI optimization isn't.

What should I ship?

Do these once and move on:

• Organization with a filled-in sameAs pointing at your real, verified profiles. This one does the most work.
• BreadcrumbList if your theme doesn't already output it.
• BlogPosting on articles.

And skip these:

• FAQPage. Google removed the feature on May 7, 2026, and deleted the documentation on June 15. It now only shows for government and health sites.
• ProfessionalService. schema.org deprecated it and folded it into Service. If you see it in 2026, someone copied a blog post from 2019.
• Service. It isn't in Google's rich results gallery, so it doesn't display anywhere.
• Review or AggregateRating about yourself. This breaks Google's review snippet policy if the reviews aren't shown on the site.

What should I be skeptical of?

Some very confident statistics in this space trace back to studies that don't exist. The widely quoted claim that schema increases AI citations by 44%, usually credited to BrightEdge, has no source anyone can find. Neither does the "73% analysis" that tends to get quoted next to it. And the claim that Microsoft confirmed schema helps its models is a misquote of a comment it made about content freshness.

If a vendor leads with any of those, you've learned a lot about the rest of their pitch.

What matters instead?

Rank position and off-site presence, by a long way. For commercial business queries, your own site makes up roughly 8% to 17% of what gets cited. Everything else is third-party: roundups, directories and review platforms. Our breakdown of what gets brands cited by AI search goes through the evidence.

That's a less satisfying answer than a markup checklist because it's harder work. It's also where the return is.

We handle the markup worth having as part of our Shopify development work. If you want a straight answer on what's worth doing for your store, that's what our free CRO audit is for.