How to Structure FAQ Schema for AI

Every “AI SEO” guide tells you to add FAQ schema. Almost none of them tell you that Google actually pulled the plug on FAQ rich results in Search, or that schema alone won’t save a page with weak content underneath it. Here’s the honest, current version.

First, the part most guides skip: Google doesn’t show FAQ rich results anymore

This matters enough to lead with it. Google restricted FAQ rich results back in August 2023 to “well-known, authoritative government and health websites” only, which quietly cut off the vast majority of businesses, blogs, and agencies from the feature. As of May 7, 2026, Google removed FAQ rich results from Search entirely, even for the government and health sites that had kept them. Search Console’s FAQ reporting and the Rich Results Test’s FAQ check are being retired in June 2026, and Search Console API support for FAQ data goes in August 2026.

So if the reason you’re adding FAQ schema is “to get the expandable dropdown in Google Search,” stop. That feature is gone, for everyone, and it isn’t coming back. FAQPage is still a valid Schema.org type and won’t cause errors or penalties if you keep it, but it no longer buys you extra SERP real estate on Google.

So why bother with FAQ schema at all

Because the audience for it has shifted from Google’s rich result renderer to AI systems parsing your page: Google’s own AI Overviews, ChatGPT, Perplexity, and similar tools. Google has confirmed it will keep using FAQ structured data to understand pages even without the visible dropdown, and other engines like Bing, plus various AI crawlers, still process FAQ markup on their own terms.

The honest caveat here too: large language models tokenize a page as text, including script tags, but they don’t semantically validate JSON-LD as structured data the way Google’s Knowledge Graph does. So schema mostly helps Google’s own AI pipeline directly, and helps ChatGPT and Perplexity indirectly, through cleaner content parsing and better entity understanding. It’s one layer in a stack, not a standalone lever, and it’s now doing that job with zero help from a Google SERP feature.

The format that actually gets extracted

Since there’s no rich-result renderer to design for anymore, structure each FAQ entry for how an AI model extracts a passage, not how Google used to display a dropdown:

• Question: phrased exactly the way a person would type or ask it, not how a marketer would title a section.
• Answer: self-contained, roughly 40 to 60 words, opening with the direct answer in the first sentence and one supporting sentence with a specific fact, number, or named detail.
• No dependency on surrounding content. If removing the paragraph above the FAQ changes the meaning of the answer, rewrite it. AI extraction pulls the passage alone, not the page around it.

Where to actually use it

Any page that answers real questions is a candidate: product pages with common objections, blog posts with a genuine FAQ section, landing pages, and dedicated help pages. Don’t bolt FAQ schema onto a page that doesn’t structurally contain question-and-answer content just to check a box. Google has said this exact pattern, generic FAQ blocks added purely to grab extra space, is part of why the feature got restricted in the first place.

Pair it with the schema that actually establishes trust

FAQ schema tells an AI system what the content is. It doesn’t tell the system whether to trust it. Pair it with:

• Article schema with named author attribution. An unattributed page is a much harder sell for citation, regardless of content quality.
• Organization schema to correctly attribute the content back to your brand as a distinct entity.
• Person schema with a “knowsAbout” field if you’re marking up an author, since it tells AI systems what domain that author is credible on.

This is the same logic behind E-E-A-T: AI systems, like Google, are trying to answer “can I trust this enough to repeat it,” and unattributed, unstructured content fails that test even when it’s accurate.

The mistake that undoes all of it

Schema-marked content still has to match what’s actually visible on the page. If your JSON-LD claims a fact the visible text doesn’t support, or your FAQ schema answers a question your on-page content doesn’t actually cover, that mismatch gets flagged during quality review and can hurt more than having no schema at all. Schema documents what’s there. It doesn’t invent what isn’t.

How to check if it’s working

Forget FAQ impressions in Search Console, that data is being retired anyway. Query the platforms your buyers actually use, ChatGPT, Perplexity, Google AI Overviews, with the exact phrasing your FAQ answers, monthly. Track whether your page gets cited, misquoted, or ignored entirely. This is the only reliable signal now, because none of the platforms expose clean citation-rate reporting the way the old FAQ rich result report once did.

If you’ve added FAQ schema and nothing changed, the problem usually isn’t the schema. It’s that the underlying content wasn’t specific enough to be worth citing in the first place, and there’s no longer a Google dropdown to fall back on for visibility. Structure gets you into the running. Content depth and clarity are what actually win it.