Median likelihood of being cited, pages with the feature against pages without, among the pages a model actually opened: structured data 1.29x (129 audits); updated in last 12 months 1.21x (111 audits); named author 1.06x (133 audits); specifications 1.06x (111 audits); visible pricing 1.05x (129 audits); direct answer in first 100 words 1.01x (133 audits).
A second, weaker measure asks how hard an already-cited page works, citations per cited page, with the feature against without: structured data 1.29x (153 audits); updated in last 12 months 1.29x (120 audits); named author 1.11x (161 audits); visible pricing 1.10x (154 audits); specifications 1.08x (134 audits); faq schema 1.08x (20 audits). Read it lightly: 84% of cited pages earn exactly one citation, so there is little for that ratio to explain.
Among pages that were cited at all, citations per page with the feature divided by without. A weaker measure: 84% of cited pages earn exactly one citation.Data table
Share of AI-cited pages carrying each trait (median across audits): structured data 71%; direct answer in first 100 words 47%; named author 46%; visible publish date 42%; updated in last 12 months 36%; specific statistics 28%.
Terms most common in third-party cited page titles: "2026" (18.6% of titles); "uk" (16.6% of titles); "best" (16.4% of titles); "how" (11.3% of titles); "guide" (8.3% of titles); "how to" (7.4% of titles); "what" (6.3% of titles); "insurance" (5.0% of titles).
The passages citations trace to run a median 131 words; 67% contain at least one number, and 22% were matched to the AI answer verbatim (across 43 attribution-depth audits).
The median cited page answers 4.0 questions outright, and 90% of cited pages answer at least one question directly.
A median 8% of profiled cited pages are cited by more than one model, and those consensus pages earn 2.1x the citations of single-model pages.
At domain level: 146,725 domains cited by one model only carry 12.4% of citation volume; 64,942 domains cited by 2-4 models carry 26.8% of citation volume; 10,718 domains cited by 5+ models carry 60.8% of citation volume.
Across 112 audited projects, the median brand holds 15% of the citations it could control, while third parties earn 86% of all citations, the earned layer, not the owned one, decides most AI visibility.
Figure 10How much likelier a page is to be cited, by feature
Share of pages cited WITH the feature, divided by the share cited without, among pages an AI model actually opened. 1.0 = no difference.Data table
Series
median lift
Structured data
1.3
Updated in last 12 months
1.2
Named author
1.1
Specifications
1.1
Visible pricing
1.1
Direct answer in first 100 words
1
FAQ schema
1
Article schema
1
Original data or research
1
Comparison table
1
Visible publish date
0.9
Specific statistics
0.9
Cites its own sources
0.6
Method & limitations
Benchmarks are medians over 183 completed content audits (112 projects), cumulative evidence to date, not a monthly movement.
Feature lifts and prevalence figures are correlational; the audits observe pages as they are, they do not run controlled experiments.
Title terms and schema types come exclusively from public third-party pages, and only where they recur across many pages and domains.
Every audit ran the same profiler, feature detection and denominators, so comparisons are like for like.
How every figure on this page is counted, floored and
reviewed: our methodology.
