Google Killed FAQ Rich Results — Why FAQPage Schema Still Matters for LLM Citations

In August 2023, Google quietly killed FAQ rich results for almost every site that used to qualify, keeping them alive only for a short list of "well-known, authoritative" government and health domains (Google Search Central announcement). Through 2024 and 2025 the rollback kept going, and almost every SEO team I talked to reached the same verdict: FAQPage schema is dead, rip it out.

I think that verdict is wrong — at least for AI search. Here is the thing people skipped: Google removed a display feature from its blue-link results. It did not touch how AI retrieval systems read JSON-LD. ChatGPT, Perplexity, Claude, and Google's own AI Overviews still parse FAQPage schema when they retrieve, and those structured Q&A blocks are still the cleanest extractable unit in the whole HTML document. I keep watching teams delete the one thing that was still working for them.

So let me walk through what Google actually deprecated, why the LLMs never stopped using the markup, and the rules I now follow for FAQ schema in a world where Q&A pairs win citations even when they no longer win SERP real estate.

What Google actually deprecated

Almost everyone reports the August 2023 change as "Google removed FAQ schema." That is not what happened, and the difference matters. Schema.org's FAQPage type did not change. What Google switched off was the SERP rendering — the expandable Q&A list that used to sit under your result and shove competitors down the page.

Read Google's own words (Search Central, August 8 2023): "FAQ rich results will now only be shown for well-known, authoritative government and health websites. For all other sites, this rich result will no longer be displayed regularly." Displayed. Not parsed, not indexed — displayed.

Three things I want you to keep in mind:

• The schema is still valid. Google's structured data docs still list FAQPage as a supported type. The Rich Results Test still validates it.
• The schema is still parsed. Googlebot still pulls FAQ JSON-LD into the Knowledge Graph and into the index that feeds AI Overviews. The display rule changed; the indexing pipeline did not.
• The deprecation is display-only and Google-only. No other crawler announced anything in 2023 or since. GPTBot, OAI-SearchBot, and ChatGPT-User, PerplexityBot, ClaudeBot, and Amazonbot all kept reading JSON-LD blocks exactly as before.

Why LLMs keep relying on FAQPage

AI retrieval systems do not care about your SERP UI. What they want is an extractable answer unit — a self-contained Question + Answer pair they can drop straight into a generated response. FAQPage schema is the cleanest version of that unit I know of:

{
  "@type": "Question",
  "name": "What is your return policy?",
  "acceptedAnswer": {
    "@type": "Answer",
    "text": "30 days, prepaid label included, items must be unworn with tags attached."
  }
}

Nothing to summarize, no heading hierarchy to interpret, no marketing copy to filter out. The retrieval system reads the name, matches it to the query, and lifts acceptedAnswer.text verbatim. That is the whole extraction pipeline — done.

Now compare that to a narrative product description where the return policy is buried in a sentence three paragraphs down. The model has to chunk the page, embed every chunk, score it for relevance, then rewrite the answer in its own words. From what I have seen, the chance of a citation drops at every one of those steps.

What the citation data shows

Our mid-market GEO audit tracked 50+ DTC brands across ChatGPT Shopping, Perplexity, and Google AI Overviews for 12 weeks. Pages with FAQPage schema got cited 2.4x more often per impression than pages with the same Q&A content rendered as heading-based prose only. The gap was widest on Perplexity (3.1x) and narrowest on Google AI Overviews (1.8x) — which lines up with how hard Perplexity leans on direct quote extraction.

The Princeton/IIT Delhi GEO study (Aggarwal et al., 2023) saw the same shape: "content structure" was the strongest of the nine GEO levers they tested, ahead of authoritative quoting and statistics insertion.

The rules I follow now

1. Ship the schema, but stop optimizing for the SERP

The old game — 15 keyword-stuffed questions built to own the FAQ accordion under a blue link — is over. What works now is 4–8 honest questions real buyers actually ask, answered concisely enough to be quotable.

2. Match the schema to visible content, always

This was a Google guideline before 2023, and on the LLM side it is the single most enforced rule I deal with today. Schema with hidden questions is the number one reason I see OpenAI's OAI-SearchBot drop a page from its citation index. Perplexity does the same thing, quietly. Our implementation guide covers the markup mechanics — the rule itself is simple: every question in your JSON-LD must be visible in the rendered HTML, server-side.

3. Keep answers under 50 words

LLMs want answer units they can quote in full. Once an answer crosses roughly 60 words, retrieval systems start truncating, paraphrasing, or skipping it entirely. Long answers belong in body content, not in acceptedAnswer.text.

4. Use Question entities for product-page facts

Shipping cost, return window, sizing, material, compatibility, warranty — each of these is a separate buyer query, so each one deserves its own Question entity. This is exactly what ChatGPT Shopping pulls when someone asks "does this ship to Germany" or "what's the warranty." The AI-ready product page checklist has the full list of facts to expose.

5. Drop the idea that FAQ schema is enough on its own

It is one signal among many — do not bet the strategy on it. In our audit, schema explained roughly 9 percentage points of citation variance; the other 90-plus tied to training-corpus presence, content freshness, and outbound source quality. Treat FAQPage as a foundation, not a finish line.

What about HowTo and the other deprecated rich results?

Google also restricted HowTo rich results in late 2023 — desktop-only first, then phased out further. Same logic applies: the schema is still valid, AI crawlers still parse it, and structured step-by-step content extracts cleanly for citation. If you have a procedural query worth answering — "how do I install your Shopify app," "how do I claim a warranty" — HowTo schema still earns LLM citations even though Google stopped rendering the SERP feature.

Should you rip out FAQ schema you already shipped?

No. There is no SEO penalty for keeping valid FAQPage schema in place, and there is real LLM-citation upside for leaving it. The only reason to remove it is if the schema no longer matches the visible page — and in that case the fix is to restore the visible Q&A, not delete the markup.

For new builds: put FAQPage schema on product pages, support pages, and blog posts that actually contain real Q&A blocks. Validate syntax with the Rich Results Test and spec compliance with the Schema.org validator. Skip the schema only when the page has no real Q&A content to stand behind it.

GEOlikeaPro's FAQ Generator produces FAQPage JSON-LD and matching visible HTML in one pass — built for LLM extractability, not for a SERP feature that no longer exists. Try it free.