Study of 2M AI citations: domain authority beats SEO signals
What this answers
- Does adding FAQ blocks, structured data or fast Core Web Vitals help a page get cited by AI engines?
- Not on its own, per a new study of 2 million AI citations — those signals' positive correlation with citations reverses or falls to zero once the analysis controls for which domain the page is on.
- What page feature best predicts whether ChatGPT, Claude, Gemini or Google AI cite it?
- How closely a page's wording overlaps with the language of real user prompts, per the study — the strongest of more than 60 features tested, ahead of any traditional SEO signal.
Discovered Labs co-founders Ben Moore and Liam Dunne, in an arXiv preprint, analyzed roughly 2 million citations from ChatGPT, Claude, Google AI and Gemini over six months, joined to 10,000 pages across 19 B2B SaaS workspaces. Testing 60-plus page features across nine statistical methods, they found prompt-content alignment — word overlap between a page and the wording of real prompts — was the strongest predictor (beta +0.37), while traditional SEO signals like FAQ blocks, structured data and Core Web Vitals showed positive citation correlations that reversed or collapsed to zero once domain-level effects were controlled for; domain-level AI authority outweighed the next-strongest feature sixfold in SHAP value. Discovered Labs is an AEO/GEO agency for B2B SaaS companies, and the study’s authors are its founders.
Why it matters: It's the largest citation-driver study on the wire's record, and it lands on the same conclusion as September's GEO-scoring study: SEO-shaped signals keep failing to predict AI citations under rigorous testing, while matching how people actually phrase prompts does.
The record: ChatGPTAnthropicGeminiGoogle
Via arXiv AI-search citations ↗
Posted to the wire September 28, 2026. Edited by Joe Balewski.