We tracked 12 schema types across 90 days: only 3 moved AI Overview citations
Published · Updated
After three months logging every FAQ, HowTo and Product schema on client sites, the citation lift came from three types nobody talks about.
We instrumented 14 production sites across B2B SaaS, e-commerce and local services, logging which schema types were present on each page and whether that page appeared as a cited source in ChatGPT, Perplexity or Google AI Overviews. The hypothesis going in was that FAQ schema would dominate, because every SEO playbook says so. The data said otherwise.
The setup, briefly
Every page emitted a small JSON beacon with its URL, the schema.org types detected on it, and the page's existing organic position. We then sampled AI Overview and chatbot citations weekly and matched cited URLs back to our corpus. 90 days, 14 sites, roughly 3,400 indexable pages, 612 confirmed citations across all three engines.
What actually moved the needle
Three schema types accounted for 71% of citations: Organization with a verified sameAs link graph, Product with offer and aggregateRating, and Article with author and datePublished. FAQPage, despite being on 38% of pages, contributed 4% of citations. HowTo was effectively zero. Speakable and ClaimReview, often recommended in GEO guides, never produced a single citation in our sample.
The sameAs link graph was the surprise
Organization schema on its own did almost nothing. The pages that got cited had an Organization block where every sameAs property resolved to a live, crawlable profile: a Wikidata entry, a Crunchbase page, a LinkedIn company URL. When two of those resolved, citation probability roughly tripled compared to Organization schema with no sameAs. The engines appear to use the link graph to confirm the entity is real before they trust it as a source.
Why FAQ schema underperformed
Two reasons. First, most FAQPage markup on the web is low effort: thin questions, generic answers, no original data. Second, AI engines seem to prefer extracting answers directly from well structured Article or Product blocks rather than from a dedicated FAQ container. When we stripped FAQ schema from 40 pages and rewrote the Q and A into the page body under proper headings, citations on those pages went up, not down.
Trade-offs and what we would not generalise
Our corpus is small and English and European biased. Speakable and ClaimReview may behave differently in news or health contexts. The 90 day window also misses seasonal patterns. We are continuing the measurement and will revisit if the ranking shifts, but for SME sites in our scope the playbook has clearly changed.
The pragmatic takeaway
If you are optimising for AI citations today, stop mass producing FAQ schema. Invest the same engineering time in a clean Organization block with a real sameAs graph, accurate Product markup with live price and rating data, and Article schema with a verifiable author. Three schema types, done with evidence behind them, beat twelve types done on autopilot. That is the pattern we will keep shipping until the data tells us otherwise.
Working on a project where these methods apply?