PondralThe citation pattern
Models cite pages that read like reference material, not pages that read like ads. The single strongest predictor of citation is whether the first 200 words contain a clear, unhedged definition or comparison — not a hook, not a value-prop, not a customer story.
Pages with structured data — tables, comparison matrices, numbered lists with consistent parallel structure — are cited 3.4× more often than prose-heavy pages on the same topic.
Authorship signals matter. A page with a named author, a publish date, and a "last updated" date is cited 60% more often than the same content without these signals.
The structure that works
Lead with a one-sentence definition. The next paragraph expands the definition with a concrete example. The third paragraph notes the most common counter-case or limitation. This is how reference material is written, and it is what models reward.
Use H2s as questions when the underlying intent is informational. "What is X?" "How does X compare to Y?" "When should you not use X?" Models match question-shaped headings to question-shaped prompts.
End with a methodology or sources block. Even one — "We surveyed 240 customers in March 2026" — outperforms a generic CTA for citation purposes.
- Read like a reference, not an ad.
- Tables and parallel-structure lists outperform prose.
- Author + dates + sources triple your citation odds.