ChatGPT and Perplexity follow patterns when naming a brand: 134–167-word self-contained paragraphs, 2.4× lift from comparison content, third-party mentions at r=0.737. Grounded in measurement.
How AI picks a brand for an answer looks like a black box, but large-scale measurement shows recurring patterns — the same patterns our engine turns into a five-dimension score. Here are the conditions for citation, in three layers.
First, the AI crawler has to be able to read the page. If your crawler is blocked or the page is a JavaScript-only SPA, the content might as well not exist. Half of our ten-brand cohort failed at this first gate. Everything else is moot until you clear it.
AI lifts text into answers paragraph by paragraph, so paragraph shape is decisive. Measured, cited paragraphs cluster in a clear band — roughly 134–167 words, self-contained, answering one question completely. Short copy ("hydration, elevated") isn't a complete answer; over-long paragraphs get cut.
Do the math on a simple case. If a product page has eight such 150-word paragraphs, you have eight citation candidates for eight queries like "gentle Korean cleanser for oily skin". Our sample case had a longest paragraph of 28 words — and a citability score of zero. Shape alone decides this dimension.
Content type matters too. In public measurement (Semrush), comparison content ("X vs Y", "best…") earned 2.4× the brand-mention rate of informational content, because most pre-purchase buyer queries are comparative.
Retrievable and well-shaped but still absent? That leaves trust. AI doesn't readily cite an unfamiliar domain. In a study of 75,000 brands, the strongest correlates of AI visibility were YouTube mentions (r=0.737) and diverse web mentions (0.66–0.71); classic backlinks trailed at 0.1–0.2. ChatGPT's most-cited sources are Reddit and Wikipedia. So trust work points at communities, video, and original data — not link buying.
The layers multiply, not add. If the crawler is blocked (Retrieval=0), the best content and trust in the world still net zero. So the point of a diagnostic isn't the score — it's finding which of the three is your bottleneck. The usual order to fix is Retrieval → Content → Trust.
The free scorecard runs these checks against your live pages and tells you which of the three pillars is your bottleneck — in two business days, no charge.