On May 7, 2026, Google stopped showing FAQ rich results in Search. Within a week, half my inbox was clients asking the same thing: should we rip the FAQ schema out? The answer is no, and getting it wrong is the most expensive structured-data mistake I'm seeing on B2B sites right now.
1 · What Google actually retired on May 7
Read the deprecation notice closely and the scope is narrow. Google retired the display of FAQ rich results—the expandable accordion that used to show under your blue link. It did not touch how it crawls or stores the underlying data.
The timeline on Google's own documentation runs in three phases: no display from May 7, then the Search Console rich-result report and Rich Results Test support both dropping in June. Here's the exact wording:
As of May 7, 2026, FAQ rich results are no longer appearing in Google Search. We will be dropping the FAQ search appearance, rich result report, and support in the Rich results test in June 2026.
Notice what that sentence is about: appearance, reporting, testing. Three display-and-tooling features. The schema type itself—FAQPage on schema.org—is untouched and still crawled.
And leaving the markup in place costs you nothing. Search Engine Journal's coverage put it plainly:
The markup won't cause problems, but it also won't produce visible results in Google Search.
The plain-language version I give non-technical clients: Google turned off a button that used to show your FAQs in blue-link results. The data behind that button is still there, still readable, and still useful—just to a different audience now.
2 · The 'schema is dead' panic was a category error
Within 48 hours, German SEO Twitter had reached the conclusion that schema was over. I watched a few accounts with real followings tell people to strip FAQPage sitewide. That advice conflates two different things, and the distinction is the whole point of this article.
- A rich-result eligibility rule governs whether Google paints a visual feature onto the SERP. That rule changed: FAQ no longer qualifies.
- An AI-crawling trust signal governs whether a machine can parse your content cleanly enough to quote it. That did not change at all.
Google's own John Mueller said the quiet part out loud when the questions started:
Google is not killing schema.
Retiring one rich-result type is refinement, not abandonment. If you deleted your FAQPage markup because a thread told you to, you optimized for a display feature that no longer exists while throwing away a signal that's becoming more valuable every quarter.
3 · How AI engines actually use your structured data
The shift to understand is from a display layer to an entity-verification layer. Google's rich result was a display use of your schema. AI engines use the same data for something completely different: confirming who you are and what you claim before they put your name in an answer.
Every major engine reads structured data at index time. As SEOcrawl put it:
FAQPage schema isn't going away, it's changing audience. ChatGPT, Claude, Perplexity and Gemini read structured data when they index web pages.
AI Mode runs on Google's Gemini models, Perplexity runs its own retrieval stack, ChatGPT Web pulls live pages—and all of them face the same problem before composing an answer: is this entity who it says it is, does this passage actually answer the question, and how confident can I be quoting it? Structured data lowers the cost of answering all three. The markup remains crawlable to the bots that matter here:
The markup continues to be crawlable by Bingbot, PerplexityBot, and the various retrieval-augmented generation crawlers indexing the open web.
Google removed its own display feature. It did not—and could not—remove the data from everyone else's index.
4 · FAQPage's second life
Here's the field observation that should stop you from touching that markup. Across the audits we run, pages that stripped FAQPage right after the May announcement started showing citation drops in Perplexity within four to six weeks. Not Google rankings—Perplexity citations. Pages that kept the markup held steady.
I'll be honest about scope: this is pattern observation across a sample of audits, not a controlled study, and citation behavior is noisy. But the mechanism explains why it happens. Retrieval-augmented generation chunks a page into passages, then ranks those passages against the user's question. A clean question-and-answer pair is the easiest possible thing to chunk and the easiest to match to query intent.
FAQ schema provides exactly this—a machine-readable format that clearly delineates questions from answers, eliminating ambiguity in content parsing.
Ambiguity is the friction every RAG system is trying to eliminate. FAQPage hands it a pre-separated question and answer with zero guessing. Strip that, and the engine has to infer the boundaries from your prose—and inference is exactly where citation confidence leaks away.
The aggregate numbers point the same direction. One tracker's audit data found that pages with valid FAQ, HowTo, and QAPage schema appear 20–30% more often in AI-generated summaries than comparable unstructured pages. Treat the exact figure as directional, but the direction is consistent with everything else here.
5 · The 2026 B2B priority schema stack
If FAQPage is no longer your headline act, what should a B2B site prioritize? Here's the stack I deploy, ranked by AEO impact, with one implementation note each.
a · Article—with datePublished and author Person
This is the freshness and authorship backbone. Google's documentation flags datePublished and dateModified as recommended fields, and they're the two signals that corroborate freshness against the May 2026 core update. Implementation note: the author must be a Person object with a real name and a sameAs to a verifiable profile—not a bare string, and not your company name in the author slot.
b · Organization—with sameAs, founder, and contactPoint
This is your entity-disambiguation anchor, and on a B2B site it's the highest-leverage block you own. The sameAs property is how AI confirms your site, your LinkedIn, and your Wikidata entry are one organization:
The 'sameAs' property links your entity to your external profiles. This is how AIs understand that your LinkedIn page, Twitter profile and website represent the same organization.
Google frames the schema's purpose the same way—disambiguation, not decoration: 'Adding organization structured data to your home page can help Google better understand your organization's administrative details and disambiguate your organization in search results.' Implementation note: put one canonical Organization block on your home page and reference it by @id everywhere else, rather than redefining it per page.
Why this matters more than it used to: entity establishment is now described as a prerequisite for AI Overview citations and AI Mode answers, not a big-brand vanity signal.
Entity establishment is no longer a vanity play: it is the prerequisite for AI Overview citations, Knowledge Panel cards, and AI Mode answers.
Gemini is trained on the Knowledge Graph. If your entity isn't established there, you're functionally invisible to a chunk of AI-generated answers—no amount of good prose fixes that.
c · FAQPage—retained on service and product pages
Keep it. Put it where buying questions live: service and product pages, not your blog index. Implementation note: write the answers to be self-contained, so a single extracted chunk reads as a complete answer without the surrounding page.
d · BreadcrumbList—for hierarchy signaling
Breadcrumbs tell a crawler where a page sits in your information architecture, which helps an engine judge topical context. Implementation note: the breadcrumb path should match your actual URL hierarchy—mismatches read as noise.
e · HowTo—for process-oriented content
For anything procedural—onboarding, setup, migration—HowTo gives the same clean step separation that makes FAQ so quotable. Implementation note: only mark up genuine sequential steps; don't wrap a marketing list in HowTo and call it a process.
6 · Before and after: a real implementation walkthrough
Take a German B2B SaaS service page—anonymized, but representative of what we see weekly. Before, the page was well-written and completely opaque to a machine. The FAQ section was plain HTML:
<section class="faq">
<h2>Häufige Fragen zur API-Integration</h2>
<h3>Wie lange dauert die Einrichtung?</h3>
<p>Die meisten Teams sind in unter 4 Stunden live.</p>
<h3>Unterstützt ihr Reverse-Charge-Rechnungen?</h3>
<p>Ja, OSS- und Reverse-Charge-Rechnungen werden automatisch erstellt.</p>
</section>A human reads that fine. A retrieval system has to guess where each answer starts and stops. We added one JSON-LD block in the <head>—no visible page change:
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{
"@type": "Question",
"name": "Wie lange dauert die Einrichtung der API-Integration?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Die meisten Teams sind in unter 4 Stunden live."
}
},
{
"@type": "Question",
"name": "Unterstützt ihr Reverse-Charge-Rechnungen?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Ja, OSS- und Reverse-Charge-Rechnungen werden automatisch erstellt."
}
}
]
}
</script>The structural outcome: in a Perplexity citation test on the integration-timeline query, the marked-up page was quoted with the exact 'under 4 hours' figure pulled cleanly from the answer field. The before-state page, with identical prose, was paraphrased vaguely and cited a competitor for the specific number. Same words on the page. The difference was whether the answer arrived pre-separated or had to be inferred.
7 · How this interacts with the May 2026 Core Update
The May 2026 core update—rolling out as I write this, expected to close around June 5—scores E-E-A-T holistically. Correct schema doesn't replace that scoring; it corroborates it. Your author Person markup backs your credential claims. Your datePublished backs your freshness. Your Organization block backs your topical authority. Schema is the machine-readable witness for things your prose asserts.
There's an observed stability effect, too. During the rollout, pages carrying FAQ, Article, and Review schema have held up better through the volatility:
Pages with FAQ schema, Article schema, and Review schema are holding better because Google's systems can understand and categorise them more accurately during a turbulent rollout period.
The mechanism is intuitive: when an algorithm is re-sorting everything, the pages it can categorize confidently move around less. But corroboration only works if the schema is correct. In this context, 'correct' means something specific:
- No mismatched types—don't mark a product page as an
Articleor a blog post as aProduct. - No missing required fields—an
Articlewithout anauthor, or anOrganizationwithout aname, corroborates nothing. - No claims the page doesn't make—schema that contradicts the visible content reads as a manipulation signal, not a trust signal.
Incorrect schema during a core update is worse than none—it gives the system a confident wrong answer about what your page is.
8 · The five-question schema audit—run it in 20 minutes
Open your highest-value service page, view source, and find your JSON-LD blocks. Then answer these five questions honestly.
- Did you strip FAQPage after May 7? If yes, put it back on service and product pages. You optimized for a dead display feature and deleted a live AI signal.
- Is your author a Person with a sameAs? A bare name string or a company-as-author does nothing for E-E-A-T corroboration. Link to a real, verifiable profile.
- Does your Organization block have sameAs, founder, and contactPoint? This is your entity anchor. Without sameAs, AI engines can't confidently confirm you're one organization across the web.
- Do your Article pages carry datePublished and dateModified? These are the freshness signals the core update checks against. Missing dates leave your recency unverified.
- Does any schema claim something the page doesn't say? Mismatched types and phantom fields are worse than no markup during a volatile rollout. Fix or remove them.
If you want to keep score on the part that actually matters now—whether ChatGPT, Perplexity, Gemini, and AI Overviews are citing you and how that moves as you fix schema—that's measurement, not a one-time check. We built Cited to track exactly that across engines in real time, so you can see a schema change show up as a citation change instead of guessing.
Five questions, 20 minutes. If you come out of it unsure whether your structured data is corroborating your content or quietly contradicting it, that's the gap a full audit closes.