Back in August 2023 I was mid-way through an audit for a legal services client when Google quietly dropped the bomb: FAQ rich results were being "limited to authoritative government and health websites." Overnight, a tactic I'd baked into something like 400 site builds over the prior two years just... stopped working. No accordion snippets. No doubled SERP real estate. The sites kept their FAQ schema in the markup, but the visual reward vanished.
Here's what I didn't do: panic and rip the schema out.
That instinct turned out to be right, for reasons I didn't fully understand until AI Overviews started rolling out properly in 2024 and I started watching citation patterns across client sites obsessively. FAQ schema isn't performing the way it used to. But it's doing something else now, and if you understand what that is, you can build for it deliberately.
What Google Actually Killed (and What It Didn't)
The September 2023 announcement from Google Search Central was specific. Rich result eligibility for FAQ markup got pulled from most commercial sites. The dropdown accordion in the SERP, the thing that doubled your listing height and crushed CTR for everyone below you, that's gone for 99% of publishers.
But the schema itself? Still perfectly valid. Still crawlable. Still indexed. Google didn't say "stop using it." They said "we're not showing the fancy SERP treatment for it."
That distinction matters enormously right now.
Think of it like this: structured data is a translation layer between your content and machines. AI systems, whether that's Google's own Gemini powering AI Overviews, or the crawlers that Bing uses to feed Copilot, or the web-browsing mode in ChatGPT, they are all machines trying to parse intent and extract factual answers at scale. Schema markup is you speaking their language directly.
I pulled the structured data reports from four e-commerce clients last March using Schema Markup Validator and cross-referenced them against their AI Overview citation frequency (tracked manually, I know, painful). The sites with clean, valid FAQ schema on high-volume category pages were getting cited at roughly 2.3x the rate of pages with identical content but no structured data. Small sample. But consistent enough to make me pay attention.
How AI Systems Are Actually Pulling Citations
This is the part where most schema guides get hand-wavy. Let me be specific about the mechanism as best I understand it.
AI Overviews and tools like Perplexity aren't scraping SERP rankings and picking the top result. They're doing something closer to retrieval-augmented generation: pulling chunks of content that match the query's semantic structure, then synthesising an answer. The question is what signals help a chunk get retrieved in the first place.
Content that is structured as a question-answer pair is easier to retrieve cleanly. The FAQ schema tells the crawler "this text block is a question, and this adjacent text block is its answer." That pair is then a coherent unit. Compare that to burying the same information in the third paragraph of a 1,400-word blog post with no markup, the answer might be there, but the machine has to work harder to isolate it.
Seahawk had a fintech client last year where we'd built out a product comparison page with roughly eighteen FAQ items in schema, covering things like "what happens to my money if the app goes bust" and "is this FCA regulated." The page wasn't ranking in the top three organically. But it was getting cited in AI Overviews for regulatory questions because the answer was clean, bounded, and marked up explicitly. That's the shift. Citation doesn't always follow ranking anymore.
The Authoritative Source Problem
Here's the thing though. AI systems are increasingly source-sensitive. Perplexity, for example, tends to favour sites with clear E-E-A-T signals: demonstrated expertise, named authors, external links pointing in. Schema alone won't rescue a thin site. But schema on a content-credible site is amplifying the signal.
The FAQ schema question-answer format pairs very naturally with Person schema on the author, and Organization schema on the publisher. Running all three together on a page tells a much richer story to any machine reading it. I started doing this as standard on Seahawk client builds about eighteen months ago and I won't stop.
What Makes an FAQ Item Actually Citation-Worthy
Not all FAQ content is created equal. I've audited too many sites where someone has stuffed schema around questions like "what are your opening hours" and "do you offer free shipping." That's fine for local SEO hygiene. It won't earn a citation from an AI system trying to answer a substantive query.
Here's what the citation-earning FAQ items tend to have in common, based on watching this play out across a few dozen sites:
- Specificity over generality. "How long does FCA authorisation typically take for a new payment firm" beats "how do I get regulated." The more specific the question matches how a real person types a query, the better.
- Self-contained answers. The answer should make sense without reading the rest of the page. If your answer says "as mentioned above," it's not citation-ready.
- A number, a name, or a date. Concrete facts anchor an answer. "Typically 12 to 18 months according to the FCA's own guidance" is the kind of answer a generative model wants to reproduce.
- No hedging fluff at the start. "That's a great question! There are many factors to consider..." is death. Start with the answer.
- Length around 40 to 80 words. Long enough to be substantive. Short enough to be extractable.
The Technical Implementation Still Has to Be Right
I've seen FAQ schema deployed wrong so many times that I now include a schema QA step as a hard gate before any Seahawk site goes live. Common failure modes:
- Using
FAQPageschema on pages that aren't actually FAQ pages (product pages with one FAQ widget at the bottom, for example, Google doesn't love this). - Nesting the JSON-LD incorrectly so the
mainEntityarray is malformed. The Rich Results Test will catch this immediately. - Including questions in schema that aren't visible on the page. This used to work. It now violates Google's guidelines and can trigger manual actions.
- Using duplicate question text across multiple pages. Pick the canonical home for each question.
- Forgetting to update schema when the page content changes. I've seen sites with schema still referencing 2021 pricing.
Run your implementation through both the Rich Results Test and the Schema Markup Validator. They catch different things. Do both.
Yoast, Rank Math, and Manual JSON-LD
If you're on WordPress, Rank Math's schema module handles FAQ schema generation cleanly and lets you bind it to a custom block. I used this on a 200-page knowledge base build for a SaaS client earlier this year and it held up well. Yoast does it too but the UI is clunkier for bulk implementation.
For headless builds or anything outside WordPress, I write JSON-LD manually and drop it in the <head>. More control. Worth the ten minutes.
What About ChatGPT and Perplexity Specifically?
Neither ChatGPT's browsing mode nor Perplexity crawl in quite the same way Google does, and they don't formally "support" schema in the way Google's structured data documentation describes. So why bother with schema if you're chasing those citations?
Because schema disciplines your content structure. It forces you to write clean question-answer pairs. And that underlying structure is what gets cited, whether or not the tool formally reads the JSON-LD.
Perplexity in particular tends to pull from pages where the answer appears near the top of the content, is separated visually (headers, clear paragraphs), and matches the query closely. FAQ schema encourages exactly that content architecture. The schema is the scaffolding. The content it enforces is the actual citeable asset.
I tested this last autumn with a client in the property sector. Two pages, nearly identical Domain Authority, similar backlink profiles. One had FAQ schema with well-written Q&A pairs. One had an unstructured FAQ section written in flowing prose. Over 90 days, Perplexity cited the structured page 11 times versus twice for the prose version. I tracked these manually using Perplexity's own search interface, typing in the key queries every fortnight. Tedious. But telling.
Should You Add FAQ Schema to Every Page?
No. And I'd push back on any agency that recommends it as a blanket tactic.
Use it where:
- You have a genuine question-and-answer content format
- The page targets informational or consideration-stage queries
- You can write answers that are specific, self-contained, and factually grounded
- The page has broader E-E-A-T credibility behind it
Don't use it on pure product listing pages, homepage hero sections, or anywhere you're shoehorning questions in just to have schema. Google will ignore it and Perplexity won't care.
The sites in my network that are getting consistent AI citations aren't the ones that sprinkled schema everywhere. They're the ones that built dedicated, high-quality Q&A content on specific topics and then marked it up correctly.
FAQ
Is FAQ schema still worth implementing in 2025?
Yes, but not for the SERP accordion it used to generate. The value now is in making your question-answer content legible to AI retrieval systems, including Google's AI Overviews, Perplexity, and Bing Copilot. Sites with clean, valid FAQ schema on well-written Q&A pages are earning citations at a measurably higher rate than equivalent pages without it.
Will FAQ schema help my rankings directly?
Not directly, no. Schema is not a ranking factor in the traditional sense. What it does is improve how machines parse and retrieve your content, which can increase your citation rate in AI-generated answers, which in turn affects visibility for zero-click and near-zero-click queries. It's an indirect pathway, but it's real.
How many FAQ items should I include per page?
There's no magic number. I typically aim for five to twelve well-written items rather than twenty thin ones. Quality matters far more than volume. Each item should earn its place by answering a genuinely distinct question with a specific, self-contained answer.
Can I use FAQ schema on a page that also has HowTo or Article schema?
Yes. Multiple schema types on a single page are fine as long as they each describe something real on the page. Don't stack them arbitrarily, but a long-form article that includes a FAQ section at the bottom can legitimately carry both Article and FAQPage schema.
What's the fastest way to check if my FAQ schema is valid?
Go to Google's Rich Results Test and paste your URL. It'll tell you whether your schema is parseable and whether it would have been eligible for rich result treatment (even if that treatment is now limited). Also run it through Schema.org's validator for a second opinion.
---
The landscape shifted. AI answers ate a huge chunk of the zero-click space and Google's SERP accordion is mostly gone for commercial publishers. But structured data as a practice isn't a relic. It's just doing a different job now: helping machines find, parse, and cite your content instead of helping Google render a fancy dropdown. Get the implementation right, write answers worth citing, and the schema does its work quietly in the background. That's still a reasonable trade.