Google Killed FAQ Rich Results in May 2026. Your FAQ Schema Still Has a Job, Just Not the One Everyone's Telling You About.

The page that isn't there anymore

On May 7, 2026, FAQ rich results stopped showing up in Google Search. Google had added a deprecation notice to its FAQPage structured-data documentation warning it was coming, and then the phase-out kept going in stages. June removed the FAQ search-appearance filter, the FAQ rich-result report, and Rich Results Test support for it. August pulled FAQ rich-result data out of the Search Console API entirely.

Here's the detail that tells you how seriously Google means it. Go looking for that original FAQPage documentation page today and it isn't there. The URL that used to explain the feature now redirects to Search Central's changelog, where the entry reads, plainly, that documentation for the FAQ rich result feature was removed because the feature is no longer shown in Google Search results. Google didn't just stop showing the snippet. It took down the page that used to explain it.

That's not the same as Google saying FAQPage is dead. FAQPage is still a real, valid schema.org type. Google isn't asking anyone to rip existing markup out, and nothing in its guidance suggests unused structured data causes a problem sitting on a page. What ended is narrower than that: the little accordion of questions that used to show up under your listing in Google's regular search results. That visual payoff is gone. The type itself, and the choice to keep using it, wasn't touched.

Does it still do anything for AI search?

If you've had a FAQPage block sitting on your site for the last couple of years on the theory that it helps you show up in ChatGPT or Google's AI Overviews, this is the moment to ask whether that theory ever had anything under it.

Google answered this one directly, in its own generative-AI search optimization guide, last updated July 10, 2026: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." Not "helps a little." Not required.

That's not the same as saying schema does nothing. The same guide still recommends keeping structured data as part of ordinary SEO, because it's what qualifies a page for rich results in regular Search, the classic kind, not the AI kind. Not required for AI features, still worth having for the search engine that isn't AI. Two different jobs, two different answers.

The one real study, and the asterisk on it

Google saying "not required" is one thing. Whether it actually moves anything is a separate question, and Ahrefs ran the closest thing to a real answer.

They tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026, with the Article, FAQ, Product, HowTo, and Organization types pooled together, against 4,000 pages that didn't add anything. They measured citations across Google AI Overviews, Google AI Mode, and ChatGPT. On AI Overviews, citations dropped 4.6%, the only result that cleared statistical significance, and it went the wrong direction. AI Mode moved 2.4%. ChatGPT moved 2.2%. Both close enough to zero that four separate statistical tests, run four different ways, agreed neither was a real effect. Ahrefs' own summary: adding schema produced no major uplift in citations, on any platform.

Here's the part that matters most if you're the plumber this study wasn't really built to describe. The pages Ahrefs tracked were already getting cited a lot, at least 100 AI Overview citations each before anyone touched their schema. That's a study about whether schema moves an already-successful page further. It says almost nothing about whether it helps a page that's currently getting cited zero times start getting cited at all, which is where most small business sites actually sit. And FAQ schema wasn't isolated from the other four types in the pool, so this isn't even a clean read on FAQPage specifically. The honest verdict is no evidence of an uplift. Not proof it can't work. Those are two different sentences, and worth keeping different.

The unsourced 3.2x you'll see everywhere

You'll run into a specific, unverified number while researching this yourself: a claim that FAQ schema makes a page 3.2 times more likely to get cited by AI, sometimes given as 2.5x or 2.7x, attributed to a Princeton University and Moz study said to have analyzed 500,000 URLs and been published at WWW 2026.

I went looking for it. No paper, no proceedings entry, no Moz publication, anywhere. What I found instead is a wave of blog posts repeating the same suspiciously specific and entirely unsourced numbers: questions in the 3-to-7 range with 100-to-200-word answers producing the 3.2x, fewer than 3 questions producing 1.4x, more than 10 producing 2.1x. None of them link back to where any of it came from.

There is a real Princeton study in this space, from Princeton and Georgia Tech together, on what actually moves AI citations. It found real gains from adding inline citations to primary sources, around 40%, specific statistics, around 37%, and named expert quotations, around 22%. Nothing in it about FAQ schema. No 3.2x anywhere in the paper. The circulating claim reads like someone borrowed that real study's credibility for a number it never produced, with "Moz" bolted on for good measure. Separately, Joost de Valk traced one tool's version of the same claim back to a source URL that just 404s, and noted that none of OpenAI, Anthropic, Google, or Perplexity has published anything saying any of them treat JSON-LD as more trustworthy than the plain rendered HTML on the same page. Nobody saying they favor it isn't the same as somebody saying they don't.

I couldn't find where the 3.2x came from, and that's the actual finding here, not that the number is fabricated. I can't prove that either. What I can tell you is that nobody repeating it can point to where it's from, and that's my call on it: unverifiable, not fabricated, and reason enough to leave it out of anything you tell a client.

So why do we still generate this block

AEO Checker still writes a FAQPage JSON-LD asset into every report we generate.

Because it's a small, real job, not a big fake one. Structured Data is 28 of the 100 points in our score, the single largest module we run. FAQPage is 8 of those 28 for most site types (5 if you run a podcast), roughly 8% of the whole report. It's not the center of gravity. It's a real chunk of a real module, no more, no less.

And it isn't a check for whether the tag exists. No FAQPage markup at all is a fail, worth zero. Present with fewer than three actual question-and-answer pairs gets half credit. Three or more pairs, with at least three quarters of the answers stating something directly, gets the full 8 points. Three or more pairs where more than a quarter of the answers hedge, redirect, or bury the actual answer behind a long windup sentence gets docked two of those eight points, and the report names the specific question doing the hedging. We're not scoring whether you remembered to add the tag. We're scoring whether the answers underneath it actually answer.

The generator that writes this asset for you works the same way. If your site has no FAQ at all, it writes three to five pairs from facts it found during the crawl. If you already have one, it keeps every existing question exactly as written, in order, leaves the answers that already lead with a fact untouched, and only rewrites the ones that hedge. When it hits a question it can't answer with anything it actually found on your site, it doesn't make something up. It tells you which fact you need to go supply.

What to actually check on your own site today

Two separate things, two separate modules, worth checking on your own before anyone charges you to tell you about them.

First, open your page's source and look for a JSON-LD block whose top-level type is FAQPage. If it's there, count the question-and-answer pairs. Fewer than three is a half-credit situation. Read the answers themselves and ask whether each one states the actual fact in the first sentence, or winds up to it instead.

Second, and separately, look at the visible page, not the markup. Are there headings phrased as actual questions, and is there a direct answer sitting right underneath each one, in text a visitor and a crawler both see? That's a different check entirely, scored in a different part of the report, and it matters whether or not the schema is there at all.

Neither one comes with a promise that ChatGPT will cite you 3.2 times more often, an unsourced number nobody has ever traced to a study. What they will do is make sure the small, honest job FAQ schema is actually built for gets done right, instead of half-done and forgotten because someone once told you it was magic.

If you want a number to hold your own two checks against, the free Grade Check at aeocheck.net runs your homepage through that same scoring engine and hands you back a letter grade and a score out of 100. It won't name which question is hedging or break that score down module by module. That level of detail is what the paid report adds. But the free check takes about a minute, costs nothing, and gives you a real baseline to check your own findings against.