Structured data for AI search - what still matters in 2026
3 min read Augustyn Głowacki
No AI engine has ever published a statement that schema.org markup improves AI answer citations, and the one controlled experiment on the question found no effect: Ahrefs tracked 1,885 pages that added JSON-LD, matched against roughly 4,000 similar pages that didn't, and measured Google AI Overviews at -4.6%, Google AI Mode at +2.4%, and ChatGPT at +2.2% - the last two statistically indistinguishable from noise (Ahrefs, 11 May 2026). Structured data still earns its keep for the rich results Google actually pays out for, and for keeping your own facts parseable. It is not a citation lever, and selling it as one isn't defensible after this test.
Does Google say schema helps AI Overviews or AI Mode?
No, directly. “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary,” and “there’s also no special schema.org structured data that you need to add” (Google, AI features documentation). Google’s May 2026 AI optimization guide repeats it under a section literally titled “Overfocusing on structured data”: keep using schema for rich-result eligibility, not because it does anything for generative AI search.
Do AI systems actually read JSON-LD at all?
The cleanest test found says mostly no. searchVIU built one page with eight product prices, each exposed through a different mechanism, then queried ChatGPT, Claude, Gemini, Perplexity, and Google AI Mode five to ten times each. Every system failed to surface the price when it existed only in JSON-LD - all five, zero hits. The same page’s visible HTML text was read correctly by ChatGPT, Gemini, and Google AI Mode. Prices injected into JSON-LD by client-side JavaScript scored the same zero, across every system including Gemini, the one engine in the test that does render JavaScript (searchVIU, October 2025, n=1 page). The systems read the visible page. Markup-only facts don’t reach them.
What happened to FAQPage and HowTo?
Both are dead as Google features, on record dates. HowTo was deprecated 13 September 2023, and the FAQ rich result stopped appearing in Google Search on 7 May 2026, with the remaining Search Console support following it through June 2026 (Search Engine Journal, “Google Drops FAQ Rich Results From Search”). No engine has ever stated FAQPage as an AI-citation signal, and the one controlled test of adding JSON-LD found no positive effect on any AI surface. Google’s standing guidance, unchanged since 2023, is not to panic-remove old markup either: “structured data that’s not being used does not cause problems for Search, but also has no visible effects.”
So what’s actually worth shipping?
Three types, each justified on grounds that don’t depend on any AI-citation claim. Organization with sameAs on your canonical entity page - Google states directly that it helps disambiguate a brand in search results. Article or BlogPosting on editorial content - no rich result requires it, but Google says it helps understand title, image, and date information, and it feeds Google News. A Person node with sameAs inside Article bylines - cheap, and covers author attribution without a site-wide profile-page retrofit. All three would be worth shipping even if every AI engine in the world ignored them completely, which is close to the honest state of the evidence today.
What’s the one rule that ties all of this together?
Never put a fact only in structured data - check every marked-up claim on a page against what the page visibly says. Every test I found agrees on this, independently: mirror the marked-up fact in visible text, or the fact doesn’t exist as far as an AI system is concerned. Markup describes your content for the systems that already trust your HTML. It doesn’t feed the ones that don’t.