How often does AI cite comparison pages?

Pages with an explicit comparison term in their address supplied 352,550 of 7,434,246 cited-page links, or 4.7%. Those links covered 123,295 cleaned pages on 46,324 websites.

Ahrefs found that best-X blog lists were 43.8% of page types in a manually classified sample of 26,283 source URLs from 750 shopping-research prompts. Semrush found that comparative content produced 2.4 times more brand mentions in a different 115-prompt study. Those studies measure selected prompt types and brand mentions. This full-corpus address rate refines rather than contradicts them. Track comparison-page share beside other page types instead of treating all cited content as comparison content.

At least one comparison-labeled page appeared in 170,788 of 701,353 cited answers, or 24.4%, and across 15,339 of 18,107 organic prompts, or 84.7%.

A comparison page may be available somewhere in a prompt's history without appearing in the next answer. Measure answer-level reach separately from prompt-level reach.

Vs supplied 181,821 links, or 51.6% of comparison-labeled links. Alternative or alternatives supplied 106,019, or 30.1%. Compare or comparison supplied 62,956, or 17.9%, and versus supplied 1,754, or 0.50%.

Keep several explicit comparison labels in page audits. A single compare filter misses most of the observed set.

Comparison-labeled pages supplied 280,261 of 5,770,388 Google AI Mode links, or 4.9%, and 72,289 of 1,663,858 ChatGPT Search links, or 4.3%. The gap was 0.51 percentage points. At least one appeared in 33.6% of cited Google AI Mode answers and 14.2% of cited ChatGPT Search answers.

Per-link share and answer reach answer different questions. Google AI Mode records more page links per answer in this window, so report both measures instead of comparing raw counts.

Comparison-labeled pages supplied 6,705 of 66,788 links from developer websites, or 10.0%. They supplied 29,152 of 390,186 editorial links, or 7.5%, and 13,619 of 220,943 finance links, or 6.2%. Documentation, government, and academic website rates were 1.6%, 0.53%, and 0.46%.

Compare page-type share within a source category. These website labels are broad publisher categories, not a judgment about the content or quality of one page.

Reddit supplied 12,287 comparison-labeled links, Medium 3,713, Dash0 2,655, G2 2,414, and LinkedIn 2,323. The top ten websites together supplied 32,958 of 352,550 links, or 9.3%.

Use the leaderboard to find pages and publishers to inspect, not as a complete placement list. A comparison-content plan should cover owned pages, specialist publishers, directories, and communities.

Amplitude's AI visibility tools page supplied 922 cited-page links across 58 prompts. IR's observability tools comparison supplied 700 across 79 prompts. HackerNoon's usage-based billing page supplied 442 across 40 prompts. Cribl's observability pipeline page supplied 380 across 56 prompts, and a Reddit project-management comparison supplied 354 across 67 prompts.

Audit the exact page and buyer question. A domain-level citation total cannot show which comparison earns repeated use or which need it serves.

Software recorded 28,626 comparison-labeled links among 320,181 cited-page links, or 8.9%. Data and Analytics recorded 8.4%, Blockchain and Cryptocurrency 8.3%, and Payments 8.1%. Manufacturing recorded 903 of 58,433, or 1.5%. Software's rate was almost six times as high.

Use an industry baseline before setting a comparison-content target. The differences describe the observed prompt corpus and do not show that industry caused a citation.

The headline token rule returned 4.7%. Requiring a complete path segment returned 0.58%. Excluding alternative terms returned 3.3%. Expanding the address rule with ambiguous best and top tokens returned 23.2%. The headline pages were 5.0% of all cleaned pages.

The public headline uses the conservative middle rule. It misses comparison pages whose address has no comparison label and can include non-brand comparisons. It does not replicate manual content classification, infer page quality, measure traffic, or show that a label caused a citation.

Comparison-labeled pages supplied 352,550 of 7,434,246 cited-page links and appeared in 24.4% of cited answers. The top ten websites supplied only 9.3% of those links.

Track exact cited pages, not domains alone. Label owned comparison pages by buyer need and audit vs, compare, and alternative formats separately. Check priority prompts on both engines. Review the cited page before inferring quality or influence. Rerun the same fixed rule next quarter before calling any change movement.