Your Next AEO Budget Should Not Fund Schema

The Gist

• Schema doesn’t move citations. A matched Ahrefs study of 1,885 pages found JSON-LD structured data produced no measurable AI citation lift in AI Mode or ChatGPT, and a slight decline in AI Overviews.
• Mentions beat backlinks. Ahrefs’ analysis of 75,000 brands found branded web mentions correlate with AI visibility at 0.664, versus 0.218 for backlinks.
• No single playbook works. Citation behavior varies so widely across AI Mode, ChatGPT and Perplexity that per-engine technical optimization can’t keep up — earned authority is the only strategy that pays off on all platforms at once.

There’s a ticket in the work backlog for the tech and SEO teams regarding FAQ schema across 400 product pages. Someone has proposed a CMS migration, on the theory that the current one is “not AI-ready.” Nobody in the room can define AI-ready, but everyone nods, because the traffic charts have been sloping the wrong direction since AI Overviews showed up and doing something feels better than doing nothing.

In this article, my goal is to save that team in Manhattan about six months of endless meetings and endless decisions.

This AI panic is legitimate and scary, and arguing about the CMS is not going to fix the problem. From what we’ve witnessed, the AI answer engines are not evaluating your technical implementation and CMS.

AI agents are evaluating whether anyone else on the internet is mentioning/linking to you in context that is relevant to your brand and your industry, which ironically is the same thing search has quietly rewarded for 20 years.

Ahrefs Data Shows Schema Markup Doesn’t Move Citations

For two years, the standard advice for AI visibility has been structured data. Add JSON-LD, mark up your FAQs, help the machines parse you. It sounded right. Schema was built to make content machine-readable, and AI engines are machines reading content.

Then somebody ran the experiment. In May 2026, Ahrefs published a study tracking 1,885 pages that added JSON-LD between August 2025 and March 2026, matched against control pages with similar citation levels that never added schema, measuring citation changes across Google AI Overviews, AI Mode and ChatGPT for 30 days before and after. That is a matched difference-in-differences design, which is specifically built to isolate what the schema itself did rather than what the site was already doing.

The result was nothing. AI Mode moved about +2.4% and ChatGPT about +2.2%, close enough to zero to be statistical noise across thousands of URLs. AI Overviews actually ticked down slightly.

Four days before that study published, Google deprecated FAQ rich results entirely.

To be fair to schema, the Ahrefs authors flag a real caveat: every page in their dataset was already receiving 100+ AI Overview citations before the schema was added, so the study can’t speak to whether markup helps pages that haven’t been discovered yet. Structured data still does useful work for crawling, parsing and classification. But classification is not selection. Being legible to a model is not the same as being chosen by one, and the gap between those two things is where most AI visibility budgets are currently disappearing.

Google’s own guidance says no special schema is required for AI Overviews or AI Mode. They have been saying it the entire time. We just preferred the version of the story where the fix was a sprint ticket.

Did Ahrefs’ Matched Study Find Schema Markup Improves AI Citations?

No — Ahrefs’ difference-in-differences study of 1,885 pages found JSON-LD schema produced statistically insignificant citation changes in AI Mode and ChatGPT, and a slight decline in AI Overviews.

Related Article: What Did 1 Billion Google AI Mode Users Search for Over the Past Year?

What Actually Predicts AI Citations: Brand Mentions or Backlinks?

Here is the finding that should reorganize your quarter. Ahrefs analyzed 75,000 brands across AI Overviews, ChatGPT and AI Mode and found that branded web mentions correlate with AI visibility at 0.664, while backlinks correlate at 0.218. YouTube mentions came in highest of all, at 0.737.

Read that carefully, because two different conclusions live inside it and only one of them is right.

The wrong conclusion is “links are dead, chase mentions instead.” That reads the metric as the mechanism. The right conclusion is that the unit of measurement changed while the underlying work stayed identical. A link and an unlinked mention are usually the same event: somebody at a publication your buyers actually read decided your company was worth referencing. Sometimes they hyperlink it. Sometimes they don’t. The engine counts both. You did not do two different things to earn them.

Nobody has ever gotten mentioned in a trade publication by upgrading their CMS.

Seer Interactive pushed this further with a hypothesis that, if it holds, is the whole ballgame: the model may decide which brands to recommend first, drawing on training data, and only then search for sources to support those choices. The citation is the bibliography, not the brainstorm. Under that model, page-level optimization cannot rescue a brand the system has never encountered. You are trying to win a footnote in an argument that was settled before the retrieval step ran.

Do Brand Mentions Correlate With AI Citations More Than Backlinks?

Yes — Ahrefs’ analysis of 75,000 brands found branded web mentions correlate with AI visibility at 0.664, compared to 0.218 for backlinks.

Why Google AI Mode, ChatGPT and Perplexity Cite Different Sources

The other reason to stop engineering for citations is that there is no single target to engineer for.

Moz’s 2026 analysis found 88% of Google AI Mode citations do not come from the organic top 10, and ChatGPT shows only about 6.5% URL overlap with Google’s top results. Cross-platform, one March 2026 analysis documented citation volume variance of up to 615x for the same brand between platforms. ChatGPT leans on Bing’s index and heavily on parametric knowledge from training. Perplexity retrieves live and rewards recency. Google runs query fan-out, splitting your search into sub-queries you were never tracking and citing whatever surfaces across them.

You cannot build four technical playbooks for four black boxes that change monthly. What you can do is be the kind of company that gets referenced by the sources all four of them already trust. That is one strategy with four payoffs instead of four strategies with none.

Do Google AI Mode, ChatGPT and Perplexity Cite the Same Sources?

No — Moz found 88% of Google AI Mode citations fall outside the organic top 10, ChatGPT shows only 6.5% URL overlap with Google’s results, and cross-platform citation volume for the same brand can vary by up to 615x.

FAQ: Does Schema Markup Actually Improve AI Citations?

Editor’s note: These questions address common misconceptions about technical AEO tactics, based on recent Ahrefs, Moz and academic research.

What Role Should Your CMS Actually Play in AEO?

None of this means the technical layer is worthless. It means it’s a floor, not a ceiling.

View All

Your CMS needs to render content server-side or fast enough that crawlers get the text. It needs to not block AI crawlers in robots.txt or at the WAF, which a surprising number of enterprise sites do accidentally after a security review. It needs clean HTML, real headings, stable URLs. Content should not be buried 900 words down a page, since cited passages skew heavily toward the top of source documents.

That’s the list. It takes a competent developer a few weeks. It is table stakes, and table stakes have never won anybody a hand.

The on-page work that does move citation rates is editorial, not structural. The original Princeton and Georgia Tech GEO research found that adding inline citations to primary sources improved AI citation rates by 40%, adding specific statistics improved rates by 37% and adding named expert quotations improved rates by 22%. Notice what all three have in common. They are not formatting. They are evidence of having done actual reporting. You cannot template your way into a named expert quotation.

Where to Redirect Your AI Visibility Budget

If you’re a CX or martech leader staring at a flat AI visibility line, here is the honest version of what to do.

Spend two weeks fixing the technical floor and then stop touching it. Redirect the rest of that budget toward earning references from sources the engines already ingest: trade publications in your category, credentialed practitioners publishing under their own names, communities where your buyers actually argue about this stuff, video, and original data nobody else has. One substantive placement in a trusted outlet outweighs a hundred directory listings, and it does so on every platform at once.

This is not a new discipline wearing a new acronym. It is the same earned-authority work that separated real brands from optimized ones back when the only surface was ten blue links. The interface changed. The scoreboard changed. The job did not.

The anxiety about AI Overviews is well founded. Just point it somewhere useful.

Learn how you can join our contributor community.