What 4,365 AI Citations RevealContent Patterns That Get Recommended
Structural analysis of GPR experiment citation URLs. Six patterns common to AI-cited pages and characteristics that reduce citation likelihood.
4,365
Citation URLs analyzed
440
Unique domains
6
Structural patterns found
109
Queries analyzed
Dataset Overview
We analyzed 4,365 citation URLs across 440 unique domains from the 109-question GPR experiment to identify structural patterns common to AI-cited content.
The previous article identified 5 conditions for getting cited. This article goes deeper, quantitatively examining the structural characteristics of the cited pages themselves.
Six Structural Patterns of AI-Cited Content
1. Self-contained answer structure
87% of frequently cited pages present a direct answer in the first paragraph under each h2. AI prefers structures where "citing this one paragraph is sufficient."
Action: Place a 1-2 sentence summary at the start of each h2 section. Self-contained blocks of 134-167 words are most frequently cited.
2. Data tables
42% of cited pages contained HTML tables (comparison, spec, pricing). Pages without tables were cited roughly 1/3 as often.
Action: Structure comparison and specification data as HTML tables. Even data that "could be explained in text" gets cited more when tabulated.
3. Clear heading hierarchy (h2 → h3 → h4)
93% of cited pages had 2+ heading levels. Average of 6.2 h2 headings per page. Flat structures (h2-only or no headings) showed significantly lower citation rates.
Action: Use at least 4 h2 sections per article with 2-3 h3 subsections each. Headings should accurately summarize their section.
4. Source attribution
78% of top-20% cited pages included external source references (links, study names, data attributions). The "claim → evidence → source" structure boosts AI trust signals.
Action: Attach sources to all statistics, numbers, and claims. Embed source links inline (more effective than footnotes).
5. Optimal word count (2,000-5,000 words)
Citation rate peaks in the 2,000-5,000 word range. Under 1,000 words: "insufficient depth." Over 8,000 words: "diffused focus."
Action: Target 2,000-5,000 words per article. If longer, split into self-contained pieces.
6. Author and update date visibility
65% of top-cited pages displayed author names; 71% showed last-updated dates. These function as E-E-A-T (Japanese) signals.
Action: Display author name and last-updated date on every article. Link with Schema.org Person/Organization markup.
Content Characteristics That Reduce Citation Likelihood
Under 500 words: Too thin for AI to consider citable
List-only articles ("Top 10 X"): Parallel items without deep analysis of each
Unsourced claims: General statements without data or evidence
Stale content (no update date): Undated pages or those not updated in 2+ years
Content Template for AI Citation Optimization
A structural template integrating all 6 patterns:
h1: [Title containing the target question]
meta: Author name, published date, last updated
h2: [Direct answer to the question]
134-167 word self-contained answer block
h2: [Background and context]
Why this question matters, prerequisites
h2: [Detailed analysis / comparison / steps]
h3 subsections, data tables, inline source links
h2: [Common questions / mistakes]
FAQ format for related questions
h2: [Next steps / related resources]
Internal links to related articles
Target: 2,000-5,000 words / 4-8 h2 sections / 1+ table
Check Your Site's Technical Readiness
Enter your domain to check your site's technical readiness to be read by AI (crawler access, structured data, initial HTML, and more). Whether AI answers cite you is measured with the paid plan.
Try free diagnosis