Does FAQ schema still work in 2026? Google cut it, AI didn't

Does FAQ schema still work in 2026? The rich result it was built for is gone: Google removed it from Search for every site on May 7, 2026, permanently. The markup itself is a different story. FAQPage is still a valid schema.org type, and the question worth asking now is whether that markup, and the Q&A content underneath it, still does anything for AI citations. The 2026 data on that is genuinely split.

Here's what actually changed, what the three most-cited 2026 studies found when researchers tried to isolate FAQ schema's effect on AI citations, and how to write FAQ content for extraction now that no rich snippet is waiting on the other end.

Does FAQ schema still work in 2026? What Google removed on May 7, and what it didn't

Google's own FAQPage documentation states plainly that the FAQ rich result "will no longer appear in Google Search results, on both desktop and mobile devices," effective May 7, 2026. The feature is gone for every site that carries the markup, with no exceptions. Search Engine Journal reports that Google will also drop the FAQ search-appearance filter, the FAQ rich result report, and Rich Results Test support for FAQ in June 2026, then remove FAQ data from the Search Console API entirely in August 2026. The whole reporting surface is being wound down on a schedule, not just the snippet.

What's notable is what Google didn't do: no blog post, no public rationale. Search Engine Journal notes Google "gave no explanation beyond adding the deprecation notice to its documentation." One line at the top of a docs page, and a feature that had been on Google's SERP for over a decade was gone.

The August 2023 restriction that set this up

This didn't come out of nowhere. In August 2023, Google announced it was restricting FAQ rich results to well-known, authoritative government and health sites, with the feature no longer shown regularly for anyone else, while pulling HowTo rich results off mobile entirely. That was Google narrowing the feature's blast radius three years before killing it outright. If you were paying attention in 2023, May 2026 wasn't a surprise, it was the second half of a sentence Google started then.

What's still valid: the markup, just not the snippet

FAQPage remains a real, valid schema.org type. Nothing about the removal makes the markup invalid or non-conformant, it just no longer triggers a visible feature in Google Search. And Google has already answered the obvious follow-up: does leaving unused FAQ markup on your pages hurt you? No. Structured data that isn't being used "does not cause problems for Search," Google has said, "but also has no visible effects." So there's no cleanup task here. Nobody needs to rip FAQPage JSON-LD out of a CMS template this week. The markup just stopped doing the one thing it used to visibly do.

Schema markup, AI Overviews, and why FAQ content still correlates with citations

This post goes deep on FAQ specifically. For the other schema types worth shipping in 2026, Article, Organization, BreadcrumbList, and how they hold up under the same scrutiny, see schema markup for AI Overviews, the broader companion to this one.

Google's own generative AI search optimization guide, last updated July 2026, is unambiguous about the current state of play: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." That line covers every schema type, not just FAQ. So if you're shipping FAQPage markup hoping it moves an AI Overview or an AI Mode result, Google itself says that's not how it works. The guide does still recommend structured data for classic rich-result eligibility, which is a separate, still-real benefit; it just isn't an AI-citation lever.

That leaves an open question the removal didn't answer: does the FAQ content itself, question-shaped headings paired with direct answers, still correlate with getting cited by ChatGPT, Perplexity, or an AI Overview, whether or not the JSON-LD wrapping it does anything? Three separate 2026 studies tried to measure that, and they don't agree with each other.

The correlational case for FAQPage

An AirOps analysis of 16,851 ChatGPT queries, covering 353,799 pages across ChatGPT's retrieval pipeline, found FAQPage was the second-highest-correlating schema type with citation: pages carrying it were cited 45.6% of the time. That trailed BreadcrumbList at 46.2% and edged out Organization at 44.3%. For comparison, pages with any JSON-LD at all were cited 38.5% of the time versus 32.0% for pages with none, a real gap the AirOps team says survives controls for word count and domain authority.

That's a genuine signal, but it's still a snapshot of pages that already have FAQPage schema, compared against pages that don't. It can't tell you what happens if you add the markup to a page that currently lacks it, because sites that ship clean FAQ schema also tend to ship clean content, fast pages, and real authors. The markup rides along with the quality signals that actually drive citations, which is exactly what the next study set out to isolate.

The controlled test that complicates it (Ahrefs, 1,885 pages)

Ahrefs ran the causal version of the experiment: 1,885 pages that added JSON-LD schema between August 2025 and March 2026, tracked against 4,000 matched control pages that didn't. The result: Google AI Mode citations moved +2.4% and ChatGPT +2.2%, both statistically indistinguishable from zero, while Google AI Overviews moved -4.6%, a small but statistically significant decline the authors don't attribute to the schema itself. Their read: "not much really changed. Schema had no clear positive or negative effect."

One detail matters more than the headline number. Every page in the study already had 100 or more AI Overview citations before schema was added, so this measures diminishing returns on pages already winning, not whether schema helps a new page get discovered in the first place. As Ahrefs puts it, "if a page is already getting picked up, our data suggests that adding schema isn't going to push it higher." That's a narrower claim than "schema doesn't matter," and it's worth holding onto that distinction before you generalize it to your own blog.

One study even found a negative correlation, and why that's not the contradiction it looks like

SE Ranking analyzed 129,000 domains and 216,524 pages across 20 niches and found pages carrying FAQ schema averaged fewer ChatGPT citations than pages without it: 3.6 versus 4.2. On its face that looks like it contradicts the AirOps finding above. It doesn't, once you notice what kind of page tends to carry FAQ schema in the wild. SE Ranking's own explanation is that FAQ schema clusters on simpler support and help-center pages, the kind that answer a narrow, low-authority question and get cited less regardless of what markup sits on them. The dip tracks page type, not the schema.

SE Ranking's researchers flagged the same limitation the other two studies share: their dataset doesn't pin down which ChatGPT version or exact timeframe produced the numbers, so, in their own words, "these patterns should be treated as directional correlations rather than proof of how ChatGPT's ranking algorithm works." Read all three together and there's no real contradiction: FAQ schema correlates positively when it's attached to substantive Q&A content on a strong page, correlates negatively when it's attached to a thin support page, and does close to nothing when a controlled test adds it to a page that was already winning. None of that is about the JSON-LD. It's about what the FAQ content underneath it actually is.

Writing FAQ schema for citation, not for a rich snippet that no longer exists

The practical shift is small but real: stop writing FAQPage markup to satisfy a Google feature and start writing it to satisfy an AI model reading your page for an extractable answer. That means the same content discipline, aimed at a different reader.

Match every answer to visible text, word for word

Google's structured-data policies are explicit: "Don't mark up content that is not visible to readers of the page," and "your structured data must be a true representation of the page content." Violating that can trigger a manual action that strips rich-result eligibility, and while the FAQ rich result itself is gone, the policy still applies to every schema type you ship, including FAQPage. Write the FAQ answer once, in the body copy or the rendered FAQ block, and let the JSON-LD mirror it exactly. Don't maintain two versions of the same answer that can drift apart.

Answer in the first sentence, then support it

An AI model extracting an answer from your page behaves the same way a featured-snippet crawler always did: it wants the direct answer in the first sentence under the question, not three sentences of preamble first. Write the question as the heading, answer it plainly in the first line, then add the supporting detail, the number, the date, the source, underneath. That structure is also exactly what makes an answer independently quotable if a model lifts just that paragraph.

Keep it dated, specific, and independently readable

A vague FAQ answer isn't citable; a specific one is. "FAQ schema helps SEO" tells a model nothing it can repeat with confidence. "Google removed the FAQ rich result on May 7, 2026" is a fact a model can quote and attribute. This is the same discipline that governs every claim on a page, not just FAQ answers, which is why fact-checking every stat and link before a post ships matters more than the schema wrapped around it. A model won't confidently repeat a claim it can't verify, and neither should you publish one.

This page is the demo: how Lyra's own FAQPage JSON-LD gets generated

Here's the part you can check yourself instead of taking our word for it. This post's frontmatter carries a faqs list, four question-and-answer pairs, nothing else. app/blog/[slug]/page.tsx, the page template for every post on this blog, reads that list and does two things with it automatically: it renders a visible FAQ section under the article body, and it emits a matching FAQPage JSON-LD block in the page's <head>, built with the exact same question and answer text. One list in the frontmatter, two synced outputs. There's no second place to edit an FAQ answer and no way for the visible copy and the markup to drift apart, because they're generated from the same source at build time.

That's also why the FAQ content in this post's body never repeats the frontmatter list as prose: writing an "## FAQ" section by hand here would just duplicate what the template already renders below the article. The four questions above exist in exactly one place in this file.

We're not claiming this setup gets this post cited more, the studies above are too mixed for anyone to make that promise honestly. What it does prove is the actual argument of this piece: shipping clean, matching FAQ markup on every post costs nothing, breaks nothing, and takes no extra editorial work once it's part of how the page is built rather than a checklist item someone remembers half the time. This is the same mechanism Lyra runs for every blog she writes: she drafts the FAQ pairs as part of the post, fact-checks the answers against a current source, and lets the CMS template generate the schema, so the markup is never a separate task that gets skipped under deadline.

Should you remove FAQ schema in 2026?

No. Keep shipping it, for three reasons that don't depend on any of the studies above resolving in your favor. It's free: if your CMS or template already generates it from your existing FAQ content, there's no added cost to leaving it in place. It's safe: Google has confirmed unused structured data causes no problems, so there's no downside case for removal. And it's aimed at the right reader now: even in the study that found a negative correlation, the explanation traced back to page quality, not the markup itself, so the fix is writing a better answer, not deleting the schema around it.

What you shouldn't do is spend more editorial time perfecting FAQPage JSON-LD than the underlying answer deserves, and you shouldn't expect a SERP rich result that Google has explicitly discontinued. Ship the markup as hygiene. Spend the real effort on the same things the answer engine optimization work has always required: a direct answer in the first line, a heading phrased the way someone would actually search, and a fact a model can verify and repeat. That's the part Google's removal didn't touch, and it's the part the mixed data above says still matters. For the engine-by-engine version of that playbook, see how to get cited by ChatGPT, Perplexity, and Claude.

Every post Lyra writes ships FAQPage JSON-LD generated from the same Q&A pairs a reader sees on the page, matched, fact-checked, and never a separate task someone forgets.

Try Lyra → · Talk to the founder