Skip to content
AI Search Index

What AI assistants actually cite

Updated daily. The same prompts run against five assistants every day, so a change on one is visible against the other four.

254,995 citations
25,890 source domains
28,397 answers
Rolling 90 days

The latest change in how an assistant sources answers

reddit.com fell from 5.7% to 2.8% of Perplexity citations, a move the other assistants did not match.

First seen Jul 20, 2026. Seen through Oct 3, 2026.

reddit.com's share of citations, by assistant

Share of all cited sources, 7-day rolling average

0%2.0%4.0%6.0%8.0%Jul 11Aug 25Oct 84.6%4.5%0.4%0.2%0.2%0%

Claude stops where we no longer see enough daily activity to report a share. The line ends rather than continuing at zero, which would claim a measurement we do not have.

How each assistant answers

ChatGPT
4.9
sources per answer
Cites something
98%
Uses what it reads
38%
Searched the web
99%
Perplexity
9.1
sources per answer
Cites something
100%
Uses what it reads
54%
Searched the web
100%
Gemini
10.5
sources per answer
Cites something
96%
Uses what it reads
100%
Searched the web
96%
Claude
13.0
sources per answer
Cites something
85%
Uses what it reads
Not reported
Searched the web
86%
Google AI Mode
11.8
sources per answer
Cites something
95%
Uses what it reads
Not reported
Searched the web
100%
Grok
9.1
sources per answer
Cites something
99%
Uses what it reads
Not reported
Searched the web
99%

Rates over every answer in the range. ChatGPT did not say whether it searched on 1.5% of answers; Perplexity did not say whether it searched on 0.1% of answers; Gemini said it did not search on 4.0% of answers; Gemini did not say whether it searched on 0.3% of answers; Claude did not say whether it searched on 14% of answers; Grok did not say whether it searched on 0.6% of answers.

How old the pages are
ChatGPT
114 days
median age of a cited page

Only ChatGPT reports when a cited page was published; 32% of its citations carry a date.

What it searched for behind the question
ChatGPT
2.3
searches per answer

reports its searches on 95% of answers, 19% use a site: filter

Gemini
2.0
searches per answer

reports its searches on 95% of answers, 0.0% use a site: filter

Claude
1
searches per answer

reports its searches on 1.1% of answers, 0% use a site: filter

Only ChatGPT, Gemini, and Claude report the searches they ran.

What kinds of sources get cited

ChatGPT

Editorial
3.1%
Community
12.2%
Reference
10.6%
Institutional
16.1%
Company website
31.5%
Other source
26.5%

Perplexity

Editorial
12.1%
Community
9.4%
Reference
2.0%
Institutional
6.0%
Company website
38.4%
Other source
32.1%

Gemini

Editorial
7.7%
Community
16.4%
Reference
2.6%
Institutional
4.5%
Company website
36.7%
Other source
32.1%

Google AI Mode

Editorial
5.6%
Community
22.8%
Reference
1.9%
Institutional
6.0%
Company website
47.9%
Other source
15.8%

Grok

Editorial
3.2%
Community
7.0%
Reference
4.2%
Institutional
2.5%
Company website
45.7%
Other source
37.4%

Measured over the past 7 days, with every share averaged across the websites we track, so a change in what is being measured cannot move a number. The same kinds sit in the same order on every card, so a row compares across assistants. Claude did not have enough activity this week to break down.

See how the assistants cite your brand

Run the same questions for your own website. Free. Email verification required; no credit card needed.

See SurfacedBy in actionWatch overview52 seconds