The number making the rounds is that pages with FAQPage schema get cited by AI at a 41% rate versus 15% without, roughly 2.7 times more. It is repeated in GEO decks as if the markup itself is the lever. It is not. When researchers ran the controlled version of that test, adding schema to thousands of pages, the citation lift was close to zero. The 41/15 gap is real in the data that produced it, but the most defensible reading is that FAQ schema marks pages that already have good question-and-answer content, and the content is doing the work. Signals runs an aged Reddit account marketplace plus an editorial network for AI brand mentions across Reddit, Quora, Product Hunt, and Threads, and we watch which signals actually move citations versus which ones just correlate. This one is a correlation people keep selling as a cause. Here is what the evidence supports, and the move that does earn the citation. :::key-takeaways - The 41% vs 15% FAQ schema citation stat comes from a single GEO vendor's 50-site study. It is correlational, not a controlled before-and-after test. - When Ahrefs added schema to 1,885 pages against a control group, AI citations barely moved: AI Overviews fell 4.6%, AI Mode rose 2.4%, ChatGPT rose 2.2%. The last two are statistically indistinguishable from zero. - A clean test showed ChatGPT and Perplexity returning data from deliberately invalid, fabricated JSON-LD. LLMs read the text inside schema, not the schema vocabulary. - Google fully removed FAQ rich results from Search as of May 7, 2026. The SERP reward for the markup is gone. - The real lever is visible question-and-answer content with self-contained answers. Keep the FAQPage markup as cheap parsing hygiene, not a citation strategy. ::: ## Does FAQ schema actually cause AI citations? No. The honest answer from the strongest available evidence is that FAQPage schema does not independently cause higher AI citation. It correlates with it, because the same teams that add structured data also write clean question-and-answer copy, use question-shaped headings, and do the rest of the GEO work. When you isolate the markup from the content, the markup's effect shrinks to near nothing. That distinction matters because it changes where you spend your time: on the answers, not the tags. This is the central confusion in 2026 GEO advice. A correlation between schema and citation gets reported as "add schema to get cited," and operators bolt JSON-LD onto thin pages expecting a lift that never comes. The markup is genuinely cheap and harmless to keep. It is simply not the thing pulling your brand into the answer. The content inside the markup is. ## Where does the 41% vs 15% number come from? It comes from one vendor study, and it is correlational. The figure traces to a [Relixir analysis of 50 sites around Google's Gemini 2.0 rollout](https://relixir.ai/blog/faq-howto-schema-google-ai-mode-gemini-2-study-2025), which reported FAQ-schema pages cited at a 41% rate versus 15% without, the highest per-schema lift in their set. Relixir is a GEO software vendor, the sample is 50 domains, and the study compares pages with schema to pages without it rather than measuring the same pages before and after adding it. That design cannot separate the markup from everything else those pages have in common. Pages that carry FAQPage schema tend to be the better-resourced, better-structured pages on a site. They were more likely to get cited for reasons that have nothing to do with the `