Google killed FAQ rich results for most sites in 2023. If you're still adding FAQ schema hoping for those accordion snippets, stop — Google's own docs confirm they've "reduced" the feature to a small slice of government and health publishers. But here's what most SEO writeups skip: the same structured markup Google walked away from is now one of the cleanest signals you can send to AI assistants. We've been watching this shift at Nuclear Marmalade, and the opportunity is real.
Why did Google stop caring about FAQ schema?
Google deprecated FAQ rich results because the feature got abused fast. Sites were stuffing fifteen fake questions onto every page, and Google's quality signals couldn't reliably tell genuine FAQs from keyword-stuffing in a trench coat. So they pulled it — no rich result, no visual accordion, no direct ranking benefit. Overnight, FAQ schema went from reliable search real-estate to dead weight for most publishers.
Except it wasn't actually dead. It just changed audience.
What does AI actually do with FAQ schema?
AI assistants — ChatGPT, Perplexity, Google's AI Overviews, Claude — don't render results the way a browser does. They parse content and extract meaning. FAQ schema hands them exactly what they want: a machine-readable list of questions paired with direct answers, already structured the way a language model processes information. When an AI is synthesising an answer about, say, AI voice agents, and it crawls a page with clean FAQ schema, there's no guessing at context. No scraping a paragraph and hoping.
This is why our work building structured content products keeps circling back to the same principle: structure isn't just for humans.
Does FAQ schema actually influence AI citations?
Yes — and most content strategists haven't caught up with this yet. LLMs extract structured content preferentially because it's lower-effort to parse than prose. A FAQPage schema with Question and Answer entities maps almost perfectly to how a retrieval-augmented system chunks and indexes content. We've tested this across several client sites. Pages with clean FAQ schema get pulled into AI-generated answers at noticeably higher rates than equivalent pages without it — even when the prose is nearly identical.
Honest caveat: we don't have a controlled double-blind study. Nobody does yet. But the pattern is consistent enough that we now treat FAQ schema as standard on every content build.
How is this different from traditional SEO?
Traditional SEO was about signals to one algorithm. You optimised for crawl budget, backlinks, keyword density, Core Web Vitals. The target was narrow, the rules were (sort of) published. Generative Engine Optimisation — GEO — means being useful to many different models, each with slightly different training data and retrieval logic. FAQ schema works across all of them because it's a shared standard. It's in the schema.org vocabulary. Every major AI crawler understands it.
The shift I keep coming back to — and it's something I talk about in the Nuclear Marmalade founder background — is that we've gone from optimising for one gatekeeper to being genuinely legible to machines in general. FAQ schema is one of the few tactics that bridges both worlds: it still does mild technical good for Google, and it does real, measurable good for AI.
What's the right way to implement FAQ schema in 2025?
Don't bolt it on as an afterthought. The mistake most sites make is generating FAQ schema from questions nobody actually asks, shoved into a JSON-LD block that has nothing to do with the surrounding content. AI models are getting better at detecting when structured data doesn't match the page — and they'll deprioritise it.
Here's what actually works:
Write the questions first, as if a real customer sent them. Then write the answers as complete, standalone sentences — not "See above" or "It depends." Then wrap the FAQPage schema around that real content. The schema should reflect the page, not replace it.
One gotcha worth catching early: if your CMS auto-generates FAQ schema from a template, check whether it's outputting duplicate @id values. Silent killer — validators pass it, parsers choke.
Should every page have FAQ schema?
No. This is where a lot of sites overcorrect. FAQ schema belongs on pages where you're genuinely answering questions — service pages, how-it-works pages, comparison pages. Slapping it on your homepage or a product listing is noise. AI systems are good at recognising noise.
The pages that benefit most are ones where someone might type a question into an AI assistant and your answer is the best one available. Think about what you know that others don't. For us at Nuclear Marmalade, that's stuff like how long AI phone agent builds actually take, what they cost in real terms, and where they break in production — not the polished version, the real one.
If you're figuring out which pages deserve FAQ treatment, the blog has more on structuring content for AI visibility.
What about Google's AI Overviews — does schema help there?
AI Overviews is a different beast. Google pulls content into it based on a mix of existing quality signals and retrieval logic they haven't fully disclosed. FAQ schema helps at the margin — it makes answers easier to extract — but the bigger lever is still E-E-A-T: experience, expertise, authoritativeness, trustworthiness.
Where FAQ schema genuinely earns its keep in AI Overviews is specificity. Vague answers don't get cited. "It depends on your situation" never appears in an AI Overview. Answers like "Most implementations take 3–6 weeks and cost between $8,000–$25,000 depending on integration complexity" do. Be specific. Be direct. Let the schema package that specificity cleanly.
We saw this firsthand with a client in professional services — we cut their content production time from two weeks per page to three days by systematising the FAQ-first approach, and their AI Overview appearances went up meaningfully in the following quarter. That's the one result I keep pointing to when people ask if this is worth the effort. It is.
Key Takeaways
- Google dropped FAQ rich results for most sites in 2023 — but AI assistants actively prefer FAQ schema because it maps cleanly to how they parse and extract answers.
- The markup isn't about gaming a search result anymore. It's about being legible to language models across every platform that might cite you.
- Write the questions from real customer intent first, then write complete answers, then add the schema. Never the other way around.
- Don't put FAQ schema on every page — only pages where you're genuinely answering something a person would search for.
- Specificity is the whole game. Vague answers don't get cited by AI. Numbers, timelines, and real-world details do.
If you're building a content strategy that needs to perform in both search and AI environments, reach out to Nuclear Marmalade — we build this stuff, we don't just write about it.