Quick answer: FAQ schema — FAQPage structured data — marks up question-and-answer content so systems can parse it as discrete, attributable pairs. Google retired FAQ rich results from its own search results in May 2026, but the same markup can still help AI answer engines identify quotable answers, provided the marked-up text matches what's actually visible on the page.
What is FAQ schema, exactly?
FAQ schema is a block of JSON-LD structured data using schema.org's FAQPage type. It lists a page's questions and answers as a machine-readable array — each one a Question entry with a nested acceptedAnswer — separate from, but matching, the plain-text Q&A a reader sees on the page. The point of the markup is to turn an ambiguous block of prose into an explicit, structured set of question-answer pairs a machine can parse without guessing where one answer ends and the next question begins.
Does FAQ schema still do anything in Google Search?
Less than it used to, and the timeline matters. Google restricted FAQ rich results to "well-known, authoritative government and health websites" back in August 2023, which already excluded most sites from the visual snippet. Then, in 2026, Google went further: its FAQPage documentation added a deprecation notice on May 8, 2026, stating the feature "will no longer appear in Google Search results," and removed the documentation entirely on June 15, 2026. The FAQ rich snippet — the expandable questions that used to show up directly on the results page — is gone.
That doesn't make the markup pointless. It changes what it's for.
So why mark up FAQs at all in 2026?
Because the audience for structured data was never only Google's rich-result renderer — it's any system trying to parse a page's content, and that now includes AI answer engines. A well-formed FAQPage block hands ChatGPT, Perplexity, and Google's AI Overviews an already-segmented set of question-answer pairs, which is exactly the shape those systems need when they're extracting a quotable passage rather than rendering a snippet. That's the mechanism covered in how AI search engines discover and cite content: a passage generally has to be well-structured and clearly attributable before it's even a candidate for citation, independent of whether Google shows it as a rich result.
This is the same shift Generative Engine Optimization is built around: optimizing for a system that extracts an answer, not one that only displays a ranked list. It's also worth being honest about the limits of any single markup type — see does llms.txt actually matter in 2026? for a similar case where a plausible-sounding technical fix turned out to matter less than it sounded like it should.
Why must FAQ schema match the visible page exactly?
This is the part worth getting right before anything else, because getting it wrong risks a manual action rather than just wasted effort. Google's general structured data guidelines state plainly: "Don't mark up content that is not visible to readers of the page," and that "your structured data must be a true representation of the page content." Marking up an answer that doesn't actually appear as visible text on the page — or that says something different from the on-page copy — isn't a gray area; Google treats it as a policy violation that can cost a page its eligibility for rich results altogether, independent of the FAQ-specific deprecation above.
In practice: every question and answer in the markup should be copy-pasteable, word for word, from something a visitor can actually read on the page without expanding a hidden accordion the crawler can't see, and without paraphrasing the answer to sound better in the markup than it reads on the page.
Which questions are actually worth marking up?
Not every question deserves the markup. A useful filter:
- It's a real question, not a section header wearing a question mark. "What is X?" earns markup; "Our approach" dressed up as "What is our approach?" doesn't.
- The answer stands alone. If the answer only makes sense after reading three paragraphs above it, it isn't a self-contained Q&A pair — it's a heading.
- It reflects something people actually ask. Questions pulled from real search queries, support tickets, or "People also ask" boxes are worth the markup; invented questions that exist only to pad a schema block aren't.
- The answer is direct. One or two clear sentences up front, not a paragraph that circles the point before landing on it.
What does the markup actually look like?
The structure is simple enough to describe in plain terms rather than raw code:
| Element | What it holds | Example |
|---|---|---|
| @type: FAQPage | Marks the whole block as a set of questions and answers | Wraps the entire FAQ section |
| mainEntity | An array — one entry per question | A list of every Q&A pair on the page |
| Question / name | The exact question text, matching the visible heading | "Does FAQ schema still help in Google Search?" |
| acceptedAnswer / text | The exact answer text, matching the visible paragraph | "Less than it used to for Google's own rich results, but..." |
Each question-answer pair in the markup should be a near-exact echo of one visible heading-and-paragraph pair on the page — not a summary, and not an expanded or improved version of it.
How do you validate FAQ schema?
Google's Rich Results Test historically checked FAQ markup, but with the FAQ feature itself deprecated, that specific check is being phased out of Search Console and the testing tools through mid-2026 along with the feature. The more durable check is schema.org's own validator or a generic Schema Markup Validator, which confirm the JSON-LD is syntactically valid FAQPage markup regardless of whether Google's tools still render a preview for it. Beyond syntax, the only validation that matters is the manual one: read the markup next to the live page and confirm every question and answer matches, word for word.
What are the most common FAQ schema mistakes?
- Marking up hidden content. Answers that only exist inside a collapsed accordion with no server-rendered text violate the visibility rule above.
- Using FAQ markup for non-FAQ content. Product specs or navigation labels reformatted as fake questions aren't FAQ content, even wrapped in FAQPage schema.
- Letting the schema drift from the page. A copy edit that updates the visible answer but not the JSON-LD leaves mismatched markup live indefinitely.
- Stacking duplicate FAQPage blocks. More than one FAQPage entity on a page describing the same questions creates conflicting markup rather than added credibility.
FAQ schema is one piece of a larger structured-data picture, not the whole of it — see structured data for LLMs: what actually helps for how it fits alongside Article, BreadcrumbList and Organization markup.
Where does ButterBlogs fit?
Schema markup is one of the steps ButterBlogs adds automatically once a post is drafted in a brand's persona — alongside internal links and meta tags — rather than a separate task left for after publishing, covered in more detail in the guide to AI blog writing. Since the markup and the visible answer have to stay in sync by rule, keeping FAQ sections current is also part of what makes Revise Existing Post useful on older posts whose answers have quietly gone stale.
Frequently Asked Questions
Should I remove FAQ schema now that Google doesn't show FAQ rich results?
No — removing it doesn't help, and the same markup can still assist AI answer engines in parsing question-and-answer content, even without Google's rich-result snippet as an outcome.
Can mismatched FAQ schema get a site penalized?
It can cost a page its eligibility for rich results through a manual action, per Google's structured data guidelines, since markup that doesn't match visible content is treated as misleading regardless of intent.
How many questions should a FAQ section have?
As many as genuinely stand alone as real questions with direct answers — a page with two solid Q&A pairs is better served than one padded to five with invented, low-value questions.
Do FAQ schema and the on-page FAQ heading need to say exactly the same thing?
Yes — the markup should be a near-exact echo of the visible question and answer, not a paraphrase or an expanded version of it.
Does FAQ schema help with ChatGPT or Perplexity citations?
It can help indirectly, by giving those systems an already-segmented, attributable answer to extract, though structure alone doesn't guarantee a citation on its own.
If FAQ schema is going to stay in a post, it has to stay accurate. See ButterBlogs' pay-per-blog pricing, or start free and let the pipeline add matching schema to every post it publishes.