What actually gets cited: patterns across 40+ AI answer engine queries

We track which pages get pulled into AI-generated answers across every domain we run AEO on. This is a working note on the patterns that have held up across more than 40 tracked queries, not a final answer — the citation behavior of these engines changes month to month, and this note will be updated as it does.

What consistently gets cited

• Direct-answer paragraphs near the top of a page. Pages that open with a one-to-two sentence direct answer get pulled far more often than pages that build up to the answer narratively.
• Specific numbers over qualitative claims. "13,000 clicks in 90 days" gets cited where "significant growth" does not. Engines appear to prefer extractable facts over framing.
• Recently updated pages on otherwise-stable topics. A page that's been quietly refreshed in the last 30-60 days outperforms an older, unmaintained page covering the same question — even when the underlying answer hasn't materially changed.

What doesn't move the needle as much as expected

• Backlink volume, in isolation, correlates weakly with citation frequency for AEO specifically — it's a far stronger signal for traditional Google ranking than it is for whether ChatGPT or Perplexity choose to cite a page.
• Long-form depth doesn't help if the direct answer is buried. A 3,000-word page that takes 800 words to state its core claim gets outcited by an 800-word page that states it in the first paragraph.

The open question

We don't yet have enough data to say definitively whether entity authority (being a recognized, citable source generally) outweighs page-level formatting, or vice versa, when the two are in tension. Current working hypothesis: formatting determines whether a specific page gets cited; entity authority determines whether a domain gets considered as a candidate source in the first place. Both matter, but they're solving different parts of the problem.