Agencies sell schema markup as an AI-visibility fix. A controlled study of 1,885 pages found it didn't move AI citations. Here's what the data actually shows, and why we still add schema, just not as the headline.
If you've spoken to an SEO or "AI visibility" agency in the last year, there's a decent chance schema markup came up. Schema is the invisible code (JSON-LD, usually) you add to a webpage to spell out, in a machine-readable format, what the page is about: a business, a product, an FAQ, a review. It's often pitched as one of the highest-leverage things you can do to get cited by ChatGPT or Google's AI answers.
A study published by Ahrefs in May 2026 put that claim to a real test, and the result was blunt: adding schema markup to a page did not meaningfully increase how often it got cited by AI.
The headline number. Ahrefs tracked 1,885 web pages that added JSON-LD schema between August 2025 and March 2026, matched each one against control pages that didn't add schema, and measured the change in AI citations over the following 30 days. Citations on Google AI Mode rose 2.4% and on ChatGPT rose 2.2%, both changes small enough to be statistically indistinguishable from random noise. Citations on Google AI Overviews actually fell 4.6%, a small but real decline the researchers couldn't fully explain.
Ahrefs started with a simpler, and more commonly cited, fact: pages that get cited by AI are almost three times more likely to carry schema markup than pages that don't. That correlation is real, and it's the number that ends up in sales decks. But correlation isn't the same as cause. Sites that bother adding schema also tend to be more technically sophisticated, publish stronger content, and earn more links, exactly the kind of sites AI already prefers to cite for other reasons.
So Ahrefs ran a second study designed to isolate schema's actual effect. They found 1,885 pages that switched from having no JSON-LD to having it, matched each one against three similar control pages (same rough citation level, never added schema), and tracked citations on Google AI Overviews, Google AI Mode, and ChatGPT for 30 days before and after the change.
| AI source | Effect on citations after adding schema | Verdict |
|---|---|---|
| Google AI Overviews | −4.6% | Small, real decline vs. matched controls; unexplained |
| Google AI Mode | +2.4% | Statistically indistinguishable from zero |
| ChatGPT | +2.2% | Statistically indistinguishable from zero |
They ran the comparison four different ways (a straight before/after test, a difference-in-differences model that strips out platform-wide trends, a week-by-week event study, and a version with a different before/after window), and all four told the same story: no meaningful citation gain on AI Mode or ChatGPT, and a small AI Overviews decline that's statistically real but too small to build a strategy around.
Every page in the dataset was already being cited heavily by AI before the test started (100+ AI Overview citations as of February 2025). Ahrefs is explicit that their data can't tell you what schema does for a page AI has never seen before. A separate test they cite, from searchVIU, found that when five major AI systems (ChatGPT, Claude, Perplexity, Gemini, Google AI Mode) fetch a page live, none of them actually read the JSON-LD, they extract only the visible HTML text. So the mechanism by which schema would help discovery is itself unclear.
If an agency is charging you specifically to add schema markup as your AI-visibility fix, this study is the reason to ask harder questions. For a business that's already readable to AI (clear category, consistent listings, plain-English service pages), adding schema on top isn't going to be the thing that gets you cited. The businesses in this study that were already winning stayed roughly where they were, schema or not.
That doesn't mean schema is worthless. It has real, separate uses: it helps some rich results appear in regular Google search, it feeds voice assistants and knowledge panels, and it makes it easier for various systems to parse who you are. We still add basic schema (Organization, LocalBusiness) for our clients as part of general good housekeeping. We just don't sell it, or bill for it, as the AI-visibility lever. The things that actually move AI citations, per this study and others, are the boring fundamentals: a correctly categorised Google Business Profile, consistent NAP details across directories, and website copy that names your services and suburbs in plain English.
This is one study, from one vendor (Ahrefs, which sells SEO tools, though the methodology and caveats are unusually transparent for an industry blog post). It covers pages that were already getting cited by AI, not pages starting from zero, so it can't tell you whether schema helps a brand-new or currently-invisible page get discovered in the first place. The 30-day measurement window may be too short to catch a slow-burn effect. And it pools all schema types together (Article, FAQ, Product, LocalBusiness, HowTo), so it's possible one specific type helps more than the average shown here. Treat "schema doesn't move AI citations" as the current best evidence for already-visible pages, not a permanent, universal rule.
Skip the guesswork. Your free AI Visibility Audit scans ChatGPT and Google AI for your business and tells you the top three things worth fixing, based on what actually correlates with getting cited.
Get your free AI Visibility Audit