Schema Markup and AI Citations: What the New Data Means

Schema markup just got the kind of test the SEO industry has been asking for. Ahrefs tracked 1,885 pages that added JSON-LD and measured what happened to citations in Google AI Overviews, Google AI Mode, and ChatGPT. The lift was flat. That study, picked up by Search Engine Journal on May 16, lands at the same time Google is pulling back visible rewards tied to several schema types, including FAQ rich results.

If you have been told that schema markup is the fastest way to get cited by AI search, this is the moment to slow down and look at the data. The findings do not say schema markup is dead. They say one specific sales pitch around schema markup is harder to defend, and that your team’s time may be better spent elsewhere.

Why the new schema markup data matters now

Two things happened back to back. Google removed FAQ rich results from Search a few weeks ago, which we covered in our piece on the FAQ rich results deprecation timeline. Then Ahrefs published its citation study on May 11, and Search Engine Journal followed up with a wider analysis on May 16 looking at what both events mean together.

The pattern is not new. Google has been narrowing what visible structured data does for years. HowTo rich results were limited to desktop and then deprecated. Course Info, Claim Review, and Estimated Salary were retired in 2025. Practice Problem markup was deprecated in 2026. The markup still validates, but the visible reward in the SERP goes away. That alone made the AI citation story the big remaining argument for adding more schema markup. Now that argument has its first real test, and the result is not flattering.

What the Ahrefs schema markup study actually tested

Ahrefs analyst Xibeijia Guan and content director Louise Linehan ran the test in two passes. The first pass looked at 6 million URLs and found that pages cited by AI were almost three times more likely to have JSON-LD than pages that were not cited. That sounds like proof, but the team flagged it as correlation. Sites that bother with structured data also tend to publish stronger content, build links, and keep their pages clean. Schema markup might be riding the wave instead of moving it.

So they ran a second test designed to isolate the effect of adding schema markup. They picked 1,885 pages that added JSON-LD between August 2025 and March 2026. Each page was matched against three control pages that never added schema, with similar pre-period citation levels. They measured how many times each page was cited 30 days before and 30 days after the schema went live, then ran a difference-in-differences analysis to strip out platform-wide trends.

The headline result: Google AI Mode citations moved by 2.4 percent, ChatGPT by 2.2 percent, and Google AI Overviews by minus 4.6 percent. The first two are statistically indistinguishable from zero, which means they look like normal random noise. The AI Overviews decline is statistically significant on paper, but Ahrefs said it cannot confidently attribute that to schema markup, since both treated and control pages were already on a downward trend before the schema was added.

One important detail: every page in the dataset had more than 100 AI Overview citations before any schema was added. These were pages already in the consideration set. Ahrefs was clear that schema markup might still help pages that are not yet visible to AI systems. The study just cannot answer that question with this data.

How AI systems actually use schema markup at retrieval time

The Ahrefs report also pointed at a separate experiment from searchVIU. They tested five AI systems, including ChatGPT, Claude, Perplexity, Gemini, and Google AI Mode, to see whether any of them used JSON-LD when fetching a page in real time. None did. During direct retrieval, every system extracted only visible HTML content. JSON-LD and hidden Microdata were ignored.

That covers one stage of the pipeline. It does not rule out schema markup playing a role in earlier stages like indexing or entity understanding. But it does explain why adding schema markup to a page that is already cited produces no real lift. The AI system reads what a human reader sees, then writes its answer from that. Content structure, clear headings, and direct answers in prose may matter more for citation than the schema markup wrapped around them, as we covered in our AI search visibility playbook.

The schema markup debate among practitioners

Senior SEO voices have been blunt. Lily Ray, VP of SEO and AI Search at Amsive, flagged on LinkedIn that around 168,000 pages now use the phrase “FAQ schema is critical for GEO.” She called it a familiar cycle. “Anything that can be spammed in SEO, will be spammed,” she wrote, pointing to her 2019 warning about FAQ schema overuse.

Joost de Valk, the founder of Yoast, was sharper. “The GEO industry is replaying early SEO, just faster,” he said in a blog post. Gianluca Fiorelli, a strategic SEO consultant, called the Ahrefs work “one of the more honest pieces of research to come out of the AI Search space in 2026.”

Not everyone agrees with the gloomy reading. Google’s Search Relations team has said at recent Search Central Live events that supported schema markup types are still worth using. The truth sits in the middle, and that middle has clear rules you can act on this week.

A five-step schema markup plan for the next 30 days

Here is how to update your structured data strategy without overreacting. Each step takes one focused work session, not a sprint.

Step one: audit your active schema types. Pull a list of every structured data type on your site, then mark each one as active rich result, active for entity understanding, or no longer rewarded. Product, Review, Event, Video, Organization, Person, and Article are still active. FAQPage and HowTo are not driving rich results in the same way. Keep the active set and stop adding new instances of the deprecated types.

Step two: shift effort from schema markup to visible content quality. If the AI systems read visible HTML at retrieval, that is where your investment goes. Strong H2 and H3 headings, direct first paragraphs, named author bylines, and citation of primary data are the levers. We walked through this content shape in our piece on SEO content fundamentals for bloggers.

Step three: separate fact pages from opinion pages. AI systems tend to cite pages that resolve a question quickly with a clear answer. That format works for definitions, comparisons, and how-to pages. Opinion and analysis pages serve a different job and should be measured differently. Build the right format for the right job before you add any markup.

Step four: test schema markup on pages that are not yet cited. The Ahrefs study only measured pages that were already being cited at scale. The open question is whether structured data helps pages get into the consideration set in the first place. Pick five to ten pages with low or no AI citations, add the right schema markup, and track results in 30 and 60 day windows. Keep five matched control pages untouched so you can see the gap.

Step five: track citations across both surfaces. Keep Google rankings and clicks in view, then add a second track for AI citations. We laid out the full measurement model in our AI Mode attribution rebuild. The point is to keep your editorial decisions grounded in both signals at once. The new schema markup data is a reminder that single-signal decisions can send you in the wrong direction.

What schema markup still does well

Schema markup is not a dead tactic. It still helps Google understand what a page is about, what entity it represents, and how it connects to your brand. That work feeds knowledge graphs, voice assistants, and the parts of search where rich results still exist. If you sell products, run events, host videos, or publish reviews, your schema markup is doing real work that you can see in your CTR data.

The piece that has weakened is the claim that adding schema markup to a page that is already visible in AI search will push it higher in citations. The Ahrefs data does not support that. Stop selling that promise to your stakeholders. Sell the version of schema markup that is doing the work, which is helping search engines parse and classify your content cleanly.

That cleaner pitch also matches what we have heard from Google’s own guidance. Structured data helps machines read pages. It does not promise rich results forever, and it does not promise AI citation lift. If your technical SEO baseline is strong, the rest of the value comes from content depth, author credibility, and entity clarity, not from one more JSON-LD block.

AI search rewards the same things great SEO has always rewarded: real expertise, clear writing, fresh data, and pages that resolve the query. Schema markup supports that work. It does not replace it. When you treat schema markup as plumbing instead of a magic lever, your team stops chasing the next shiny structured data type and starts investing in things that compound. For ecommerce sites, Product, Review, and Offer schema markup still drives visible CTR gains, and our ecommerce SEO keyword playbook pairs that markup with the content templates that earn AI citations. Our analysis of AI search terms reporting in Google Ads describes the parallel shift on the paid side.

Want help running the schema markup audit on your own site, or a partner to test the pages that are not yet getting cited? Book a free consultation and our team will walk through it with you. Let’s Grow!

Sources: Ahrefs research published May 11, 2026 (“We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved.” by Louise Linehan and Xibeijia Guan); Search Engine Journal coverage by Matt G. Southern, “SERP FAQ Removal & New Data Challenge Schema’s AI Search Value”, May 16, 2026.