Short answer: no, not directly. Adding JSON-LD schema to a page does not, by itself, make ChatGPT, Google AI Overviews, or Perplexity cite you more often. That is not our opinion — it is what a matched difference-in-differences study of 1,885 pages found when Ahrefs tracked schema additions against a control group between August 2025 and March 2026. The uplift most agencies (including, at times, us) have promised clients from adding structured data to a page turns out to be close to zero, and on one major platform it was slightly negative.
That is an uncomfortable thing to publish on a blog that also tells you how to write FAQ schema that AI engines cite. So this post is the nuance behind that headline: what the 2026 research actually measured, why schema behaves the way it does inside an AI retrieval pipeline, and what to do with your limited content hours instead.
What the Ahrefs study on 1,885 pages actually found
Ahrefs published its research on May 11, 2026. The methodology was straightforward: find pages that added JSON-LD schema during the study window, match them against a control group of roughly 4,000 similar pages that did not, and measure the change in AI citation rate across three platforms.
| Platform | Citation change after adding schema | Statistically significant? |
|---|---|---|
| Google AI Overviews | −4.6% (roughly 12 fewer daily citations per page, on average) | Yes |
| Google AI Mode | +2.4% | No — indistinguishable from noise |
| ChatGPT | +2.2% | No — indistinguishable from noise |
Read that table twice. Two of three platforms showed no meaningful movement at all, and the one platform that did move, moved in the wrong direction. Ahrefs’ own conclusion was blunt: adding schema markup did not increase AI citations on any of the three platforms it tested.
Why doesn’t structured data help the way we assumed it would?
The mechanical explanation is the interesting part. When researchers at SearchVIU tested how ChatGPT, Claude, Perplexity, Gemini, and Google AI Mode actually retrieve a page in real time, every single system extracted only the visible HTML content. JSON-LD, hidden Microdata, and hidden RDFa were all ignored during that live fetch.
That lines up with a separate experiment from Otterly, which tested seven AI platforms’ ability to even read schema markup when explicitly asked to. Only one — Gemini 3 — could reliably fetch and parse it. Google AI Mode, when pushed, hallucinated schema types that were not present on the page at all. ChatGPT, Claude, Perplexity, and Microsoft Copilot could not fetch it.
If most of the systems doing the citing cannot reliably read your JSON-LD in the first place, it stops being surprising that adding more of it does not move citation rates. The invisible markup you spent an afternoon adding to your <head> is, for most AI answer engines, simply not part of what gets read.
Then why do schema-heavy pages get cited more often?
This is the part that keeps the myth alive, and it is worth sitting with because it is a genuine, measurable pattern — just not a causal one. Ahrefs found that pages cited by AI systems were almost three times more likely to carry JSON-LD than pages that were not cited.
That correlation is real. The explanation the study lands on is also the more boring, more useful one: schema markup tends to live on better-maintained, more technically sophisticated sites. Those same sites usually have stronger content, clearer information architecture, more consistent publishing, and better overall E-E-A-T signals — all of which are things AI systems and traditional search algorithms actually do reward. The schema is a marker of a well-run site, not the reason the site gets cited.
It is the same trap as noticing that companies with expensive office chairs tend to be more profitable. The chairs are not the cause. Something else that correlates with buying good chairs — scale, discipline, cash flow — is doing the work.
So should you rip out your schema markup?
No. This is where we want to be precise rather than provocative. Schema markup still does real work, just not the work most people think:
- It remains the mechanism behind classic rich results in traditional Google search — star ratings, FAQ dropdowns, breadcrumbs — and those still move click-through rate.
- It gives Google’s traditional indexing systems (which are not the same pipeline as AI Overviews’ live retrieval) cleaner signals about what a page is, which can support conventional rankings.
- It costs very little to maintain once it is set up correctly, so the ongoing cost of keeping it is low even if the upside is smaller than advertised.
What changes is the priority. If you are deciding between spending the next two hours adding review schema to twenty product pages, or spending those same two hours rewriting the actual visible copy on your five most important pages, the research now clearly says: rewrite the copy.
What actually correlates with AI citations, then?
Strip out schema and look at what these same studies, plus the retrieval-behavior research, consistently point to instead:
Visible, well-structured answers
Since AI systems read rendered HTML, not markup, the thing that actually mirrors a good FAQ schema entry is a visible question posed as an H3, followed immediately by a direct, two-to-three sentence answer. This is the same principle we covered in how structured data helps your content appear in AI answers — the structure that matters is the one a human reader (and therefore a text-extraction model) can actually see on the page.
Freshness and maintenance signals
Pages that are visibly updated, with dates, revised numbers, and current examples, get pulled into AI answers more consistently than static pages, schema or no schema.
Genuine topical depth over volume
A handful of thorough, specific, well-organized pages consistently outperform a large number of thin pages padded with markup. This is one of the recurring themes in our own rundown of GEO myths wasting small business budgets — effort spent on technical add-ons instead of the actual writing is effort misallocated.
What about the “December uptick” agencies keep citing?
If you have seen a case study floating around claiming a client’s AI citations jumped after a schema rollout, the Otterly experiment actually captured one of those moments — and then kept watching. Google AI Overviews citations for the tested brand rose 611% and Google AI Mode rose 42% over the three months following a schema addition. On the surface, that looks like exactly the proof schema advocates want.
Look closer and the story falls apart. ChatGPT citations for the same brand, over the same period, fell 71%. Gemini fell 35%, despite being the one platform that could actually read the schema correctly. Perplexity and Copilot moved anywhere from flat to down 64%. If schema markup were the cause, it should have moved every platform in the same direction, since the underlying page and its markup did not change per platform. Instead, the swings tracked competitor content changes and algorithm updates during that window far more closely than they tracked the schema rollout itself. The researchers’ own conclusion was that schema functions as an SEO lever with indirect AI visibility benefits through better organic ranking, not as a direct GEO growth tactic. A single chart showing “citations up 611%” makes a great slide in a sales deck; it makes a much weaker case once you see the other six numbers sitting next to it.
A practical checklist for this week
- Keep the schema you already have. Do not spend new hours adding more of it in the name of AI citations.
- Audit your five highest-traffic pages for whether the actual visible text answers the core question in the first two sentences under each heading.
- Add a visible date or “last updated” line to your evergreen pages, and actually update the numbers on them once a quarter.
- If you use FAQ schema, make sure the same questions and answers also exist as plain, visible text on the page — not just in the markup.
- Reinvest the time you would have spent on markup into rewriting thin sections of your most important pages.
Structured data was never a magic switch for GEO, and for the first time in 2026 there is a large enough study to say that plainly instead of guessing. Treat it as SEO hygiene, not AI strategy, and put your writing hours where the research says they actually pay off: in the visible copy itself.