18.2×
Google AI Mode: top-3 vs 21-50 read rate
13%
of ChatGPT's citations rank in Google's top 50
62%
read rate for pages ranking on 4+ questions
81
analyses with Google results

The rank curve, model by model

Google AI Mode: read 17.4% of top-3 pages and 1% of 21-50 (18.2×). Google AIO: read 38% of top-3 pages and 2.4% of 21-50 (16.1×). Gemini: read 25.2% of top-3 pages and 2.9% of 21-50 (8.6×). ChatGPT: read 9.1% of top-3 pages and 1.1% of 21-50 (8.2×). Claude: read 21.9% of top-3 pages and 3.7% of 21-50 (5.9×).

Figure 1Share of Google-ranked pages each model reads, by rank
Share of Google-ranked pages each model reads, by rankGoogle AI Mode Google AI Mode Google AIO Google AIO Gemini Gemini ChatGPT ChatGPT
On prompts where the model searched. Read = opened or cited.
Data table
Series1-34-1011-2021-50
Google AI Mode17.4%6.5%2.4%1%
Google AIO38%20.3%5.9%2.4%
Gemini25.2%13.4%6.9%2.9%
ChatGPT9.1%3.5%1.8%1.1%
Claude21.9%12.6%7.4%3.7%

How much of what each model cites Google ranks

Google AI Mode: 41.2% in the prompt's top 50. Google AIO: 59.9% in the prompt's top 50. Gemini: 45.3% in the prompt's top 50, 67.7% counting its own searches. ChatGPT: 13.3% in the prompt's top 50, 17.8% counting its own searches. Claude: 36.4% in the prompt's top 50, 47% counting its own searches.

Figure 2Cited pages that rank in Google's top 50 for the prompt
Cited pages that rank in Google's top 50 for the prompt Google AIO: 59.9% Google AIO 59.9% Gemini: 45.3% Gemini 45.3% Google AI Mode: 41.2% Google AI Mode 41.2% Claude: 36.4% Claude 36.4% ChatGPT: 13.3% ChatGPT 13.3%
Share of each model's cited pages.
Data table
Seriestop 50
Google AIO59.9%
Gemini45.3%
Google AI Mode41.2%
Claude36.4%
ChatGPT13.3%

Breadth beats a single rank

Pages ranking for 1 prompt: read 11.8% of the time (59716 pages). Pages ranking for 2-3 prompts: read 34.9% of the time (4884 pages). Pages ranking for 4+ prompts: read 61.6% of the time (695 pages).

Figure 3Read rate by how many of the questions a page ranks for
Read rate by how many of the questions a page ranks for 1 prompt: 11.8% 1 prompt 11.8% 2-3 prompts: 34.9% 2-3 prompts 34.9% 4+ prompts: 61.6% 4+ prompts 61.6%
Read rate by how many of the questions a page ranks for
Data table
Seriesread
1 prompt11.8%
2-3 prompts34.9%
4+ prompts61.6%

Method & limitations

  • Cumulative sample to 2026-10-01: 81 analyses across 80 brand categories, skewed toward the markets and categories our clients measure.
  • One Google snapshot per prompt, read close to the analysis. This is correlation; a page that ranks is also more relevant to the question.
  • Breadth read rates count a page as read if any model read it on any question, so a page ranking for more questions has more chances.
  • Too little evidence to report: ChatGPT Web, Grok, Perplexity.

How every figure on this page is counted, floored and reviewed: our methodology.