A sketch of a puzzled robot whose answer cites two documents and a book

How do AI chatbots use sources?

AI chatbots act as gatekeepers in a new attention market and information ecosystem. This explorer audits who they choose to amplify, how they distribute credit, and more.

Questions with a scientific consensus, and public health.

Citation density

How often do chatbots cite external sources?

Share of answers with at least one link/citation.

Data

How many external sources do chatbots use on average?

Average citations in answers with at least one citation.

Data

How do chatbots’ citation patterns change over time?

Share of answers with a citation, each day.

0%25%50%75%100%Sep 11Sep 25Oct 9Oct 23GPT-5, Sep 11: 29% (n = 7)GPT-5, Sep 12: 29% (n = 7)GPT-5, Sep 13: 29% (n = 7)GPT-5, Sep 14: 29% (n = 7)GPT-5, Sep 15: 29% (n = 7)GPT-5, Sep 16: 14% (n = 7)GPT-5, Sep 17: 29% (n = 7)GPT-5, Sep 18: 29% (n = 7)GPT-5, Sep 19: 29% (n = 7)GPT-5, Sep 20: 29% (n = 7)GPT-5, Sep 21: 29% (n = 7)GPT-5, Sep 22: 29% (n = 7)GPT-5, Sep 23: 29% (n = 7)GPT-5, Sep 24: 14% (n = 7)GPT-5, Sep 25: 29% (n = 7)GPT-5, Sep 26: 29% (n = 7)GPT-5, Sep 27: 29% (n = 7)GPT-5, Sep 28: 29% (n = 7)GPT-5, Sep 29: 29% (n = 7)GPT-5, Sep 30: 14% (n = 7)GPT-5, Oct 1: 29% (n = 7)GPT-5, Oct 2: 29% (n = 7)GPT-5, Oct 3: 29% (n = 7)GPT-5, Oct 4: 14% (n = 7)GPT-5, Oct 5: 29% (n = 7)GPT-5, Oct 6: 29% (n = 7)GPT-5, Oct 7: 29% (n = 7)GPT-5, Oct 8: 29% (n = 7)GPT-5, Oct 9: 14% (n = 7)GPT-5, Oct 10: 14% (n = 7)GPT-5, Oct 11: 29% (n = 7)GPT-5, Oct 12: 29% (n = 7)GPT-5, Oct 13: 29% (n = 7)GPT-5, Oct 14: 14% (n = 7)GPT-5, Oct 15: 29% (n = 7)GPT-5, Oct 16: 29% (n = 7)GPT-5, Oct 17: 29% (n = 7)GPT-5, Oct 18: 29% (n = 7)GPT-5, Oct 19: 29% (n = 7)GPT-5, Oct 20: 29% (n = 7)GPT-5, Oct 21: 29% (n = 7)GPT-5, Oct 22: 29% (n = 7)GPT-5, Oct 23: 29% (n = 7)GPT-5, Oct 24: 14% (n = 7)GPT-5, Oct 25: 29% (n = 7)GPT-5, Oct 26: 29% (n = 7)GPT-5, Oct 27: 14% (n = 7)GPT-5, Oct 28: 29% (n = 7)GPT-5, Oct 29: 29% (n = 7)GPT-5, Oct 30: 43% (n = 7)Claude Sonnet 5, Sep 11: 57% (n = 7)Claude Sonnet 5, Sep 12: 57% (n = 7)Claude Sonnet 5, Sep 13: 43% (n = 7)Claude Sonnet 5, Sep 14: 57% (n = 7)Claude Sonnet 5, Sep 15: 57% (n = 7)Claude Sonnet 5, Sep 16: 57% (n = 7)Claude Sonnet 5, Sep 17: 57% (n = 7)Claude Sonnet 5, Sep 18: 57% (n = 7)Claude Sonnet 5, Sep 19: 71% (n = 7)Claude Sonnet 5, Sep 20: 57% (n = 7)Claude Sonnet 5, Sep 21: 57% (n = 7)Claude Sonnet 5, Sep 22: 57% (n = 7)Claude Sonnet 5, Sep 23: 57% (n = 7)Claude Sonnet 5, Sep 24: 57% (n = 7)Claude Sonnet 5, Sep 25: 57% (n = 7)Claude Sonnet 5, Sep 26: 57% (n = 7)Claude Sonnet 5, Sep 27: 57% (n = 7)Claude Sonnet 5, Sep 28: 57% (n = 7)Claude Sonnet 5, Sep 29: 57% (n = 7)Claude Sonnet 5, Sep 30: 57% (n = 7)Claude Sonnet 5, Oct 1: 57% (n = 7)Claude Sonnet 5, Oct 2: 57% (n = 7)Claude Sonnet 5, Oct 3: 57% (n = 7)Claude Sonnet 5, Oct 4: 57% (n = 7)Claude Sonnet 5, Oct 5: 43% (n = 7)Claude Sonnet 5, Oct 6: 57% (n = 7)Claude Sonnet 5, Oct 7: 57% (n = 7)Claude Sonnet 5, Oct 8: 43% (n = 7)Claude Sonnet 5, Oct 9: 57% (n = 7)Claude Sonnet 5, Oct 10: 57% (n = 7)Claude Sonnet 5, Oct 11: 57% (n = 7)Claude Sonnet 5, Oct 12: 57% (n = 7)Claude Sonnet 5, Oct 13: 57% (n = 7)Claude Sonnet 5, Oct 14: 57% (n = 7)Claude Sonnet 5, Oct 15: 57% (n = 7)Claude Sonnet 5, Oct 16: 57% (n = 7)Claude Sonnet 5, Oct 17: 71% (n = 7)Claude Sonnet 5, Oct 18: 57% (n = 7)Claude Sonnet 5, Oct 19: 57% (n = 7)Claude Sonnet 5, Oct 20: 71% (n = 7)Claude Sonnet 5, Oct 21: 57% (n = 7)Claude Sonnet 5, Oct 22: 57% (n = 7)Claude Sonnet 5, Oct 23: 57% (n = 7)Claude Sonnet 5, Oct 24: 57% (n = 7)Claude Sonnet 5, Oct 25: 57% (n = 7)Claude Sonnet 5, Oct 26: 43% (n = 7)Claude Sonnet 5, Oct 27: 57% (n = 7)Claude Sonnet 5, Oct 28: 57% (n = 7)Claude Sonnet 5, Oct 29: 43% (n = 7)Claude Sonnet 5, Oct 30: 57% (n = 7)Gemini 3.5 Flash, Sep 11: 86% (n = 7)Gemini 3.5 Flash, Sep 12: 100% (n = 7)Gemini 3.5 Flash, Sep 13: 100% (n = 7)Gemini 3.5 Flash, Sep 14: 86% (n = 7)Gemini 3.5 Flash, Sep 15: 100% (n = 7)Gemini 3.5 Flash, Sep 16: 86% (n = 7)Gemini 3.5 Flash, Sep 17: 86% (n = 7)Gemini 3.5 Flash, Sep 18: 86% (n = 7)Gemini 3.5 Flash, Sep 19: 86% (n = 7)Gemini 3.5 Flash, Sep 20: 86% (n = 7)Gemini 3.5 Flash, Sep 21: 71% (n = 7)Gemini 3.5 Flash, Sep 22: 86% (n = 7)Gemini 3.5 Flash, Sep 23: 86% (n = 7)Gemini 3.5 Flash, Sep 24: 100% (n = 7)Gemini 3.5 Flash, Sep 25: 86% (n = 7)Gemini 3.5 Flash, Sep 26: 86% (n = 7)Gemini 3.5 Flash, Sep 27: 86% (n = 7)Gemini 3.5 Flash, Sep 28: 100% (n = 7)Gemini 3.5 Flash, Sep 29: 100% (n = 7)Gemini 3.5 Flash, Sep 30: 86% (n = 7)Gemini 3.5 Flash, Oct 1: 86% (n = 7)Gemini 3.5 Flash, Oct 2: 100% (n = 7)Gemini 3.5 Flash, Oct 3: 86% (n = 7)Gemini 3.5 Flash, Oct 4: 100% (n = 7)Gemini 3.5 Flash, Oct 5: 86% (n = 7)Gemini 3.5 Flash, Oct 6: 100% (n = 7)Gemini 3.5 Flash, Oct 7: 86% (n = 7)Gemini 3.5 Flash, Oct 8: 86% (n = 7)Gemini 3.5 Flash, Oct 9: 86% (n = 7)Gemini 3.5 Flash, Oct 10: 100% (n = 7)Gemini 3.5 Flash, Oct 11: 100% (n = 7)Gemini 3.5 Flash, Oct 12: 86% (n = 7)Gemini 3.5 Flash, Oct 13: 86% (n = 7)Gemini 3.5 Flash, Oct 14: 100% (n = 7)Gemini 3.5 Flash, Oct 15: 100% (n = 7)Gemini 3.5 Flash, Oct 16: 100% (n = 7)Gemini 3.5 Flash, Oct 17: 100% (n = 7)Gemini 3.5 Flash, Oct 18: 100% (n = 7)Gemini 3.5 Flash, Oct 19: 100% (n = 7)Gemini 3.5 Flash, Oct 20: 100% (n = 7)Gemini 3.5 Flash, Oct 21: 100% (n = 7)Gemini 3.5 Flash, Oct 22: 100% (n = 7)Gemini 3.5 Flash, Oct 23: 100% (n = 7)Gemini 3.5 Flash, Oct 24: 100% (n = 7)Gemini 3.5 Flash, Oct 25: 100% (n = 7)Gemini 3.5 Flash, Oct 26: 100% (n = 7)Gemini 3.5 Flash, Oct 27: 100% (n = 7)Gemini 3.5 Flash, Oct 28: 100% (n = 7)Gemini 3.5 Flash, Oct 29: 100% (n = 7)Gemini 3.5 Flash, Oct 30: 100% (n = 7)Grok 4.5, Sep 11: 86% (n = 7)Grok 4.5, Sep 12: 71% (n = 7)Grok 4.5, Sep 13: 100% (n = 7)Grok 4.5, Sep 14: 86% (n = 7)Grok 4.5, Sep 15: 86% (n = 7)Grok 4.5, Sep 16: 100% (n = 7)Grok 4.5, Sep 17: 57% (n = 7)Grok 4.5, Sep 18: 86% (n = 7)Grok 4.5, Sep 19: 100% (n = 7)Grok 4.5, Sep 20: 71% (n = 7)Grok 4.5, Sep 21: 100% (n = 7)Grok 4.5, Sep 22: 86% (n = 7)Grok 4.5, Sep 23: 100% (n = 7)Grok 4.5, Sep 24: 100% (n = 7)Grok 4.5, Sep 25: 100% (n = 7)Grok 4.5, Sep 26: 100% (n = 7)Grok 4.5, Sep 27: 86% (n = 7)Grok 4.5, Sep 28: 100% (n = 7)Grok 4.5, Sep 29: 100% (n = 7)Grok 4.5, Sep 30: 100% (n = 7)Grok 4.5, Oct 1: 71% (n = 7)Grok 4.5, Oct 2: 100% (n = 7)Grok 4.5, Oct 3: 86% (n = 7)Grok 4.5, Oct 4: 86% (n = 7)Grok 4.5, Oct 5: 86% (n = 7)Grok 4.5, Oct 6: 86% (n = 7)Grok 4.5, Oct 7: 86% (n = 7)Grok 4.5, Oct 8: 86% (n = 7)Grok 4.5, Oct 9: 100% (n = 7)Grok 4.5, Oct 10: 100% (n = 7)Grok 4.5, Oct 11: 86% (n = 7)Grok 4.5, Oct 12: 100% (n = 7)Grok 4.5, Oct 13: 100% (n = 7)Grok 4.5, Oct 14: 100% (n = 7)Grok 4.5, Oct 15: 100% (n = 7)Grok 4.5, Oct 16: 100% (n = 7)Grok 4.5, Oct 17: 100% (n = 7)Grok 4.5, Oct 18: 100% (n = 7)Grok 4.5, Oct 19: 100% (n = 7)Grok 4.5, Oct 20: 100% (n = 7)Grok 4.5, Oct 21: 86% (n = 7)Grok 4.5, Oct 22: 86% (n = 7)Grok 4.5, Oct 23: 100% (n = 7)Grok 4.5, Oct 24: 86% (n = 7)Grok 4.5, Oct 25: 100% (n = 7)Grok 4.5, Oct 26: 100% (n = 7)Grok 4.5, Oct 27: 100% (n = 7)Grok 4.5, Oct 28: 100% (n = 7)Grok 4.5, Oct 29: 100% (n = 7)Grok 4.5, Oct 30: 100% (n = 7)Muse Spark 1.2, Sep 11: 100% (n = 6)Muse Spark 1.2, Sep 12: 100% (n = 6)Muse Spark 1.2, Sep 13: 100% (n = 6)Muse Spark 1.2, Sep 14: 100% (n = 6)Muse Spark 1.2, Sep 15: 100% (n = 6)Muse Spark 1.2, Sep 16: 100% (n = 6)Muse Spark 1.2, Sep 17: 100% (n = 6)Muse Spark 1.2, Sep 18: 100% (n = 6)Muse Spark 1.2, Sep 19: 100% (n = 6)Muse Spark 1.2, Sep 20: 100% (n = 6)Muse Spark 1.2, Sep 21: 100% (n = 6)Muse Spark 1.2, Sep 22: 100% (n = 6)Muse Spark 1.2, Sep 23: 100% (n = 6)Muse Spark 1.2, Sep 24: 100% (n = 6)Muse Spark 1.2, Sep 25: 83% (n = 6)Muse Spark 1.2, Sep 26: 83% (n = 6)Muse Spark 1.2, Sep 27: 100% (n = 6)Muse Spark 1.2, Sep 28: 100% (n = 6)Muse Spark 1.2, Sep 29: 100% (n = 6)Muse Spark 1.2, Sep 30: 100% (n = 6)Muse Spark 1.2, Oct 1: 100% (n = 6)Muse Spark 1.2, Oct 2: 83% (n = 6)Muse Spark 1.2, Oct 3: 100% (n = 6)Muse Spark 1.2, Oct 4: 100% (n = 6)Muse Spark 1.2, Oct 5: 100% (n = 6)Muse Spark 1.2, Oct 6: 100% (n = 6)Muse Spark 1.2, Oct 7: 100% (n = 6)Muse Spark 1.2, Oct 8: 100% (n = 6)Muse Spark 1.2, Oct 9: 100% (n = 6)Muse Spark 1.2, Oct 10: 100% (n = 6)Muse Spark 1.2, Oct 11: 100% (n = 6)Muse Spark 1.2, Oct 12: 100% (n = 6)Muse Spark 1.2, Oct 13: 100% (n = 6)Muse Spark 1.2, Oct 14: 100% (n = 6)Muse Spark 1.2, Oct 15: 100% (n = 6)Muse Spark 1.2, Oct 16: 83% (n = 6)Muse Spark 1.2, Oct 17: 100% (n = 6)Muse Spark 1.2, Oct 18: 100% (n = 6)Muse Spark 1.2, Oct 19: 100% (n = 6)Muse Spark 1.2, Oct 20: 100% (n = 6)Muse Spark 1.2, Oct 21: 100% (n = 6)Muse Spark 1.2, Oct 22: 100% (n = 6)Muse Spark 1.2, Oct 23: 100% (n = 6)Muse Spark 1.2, Oct 24: 100% (n = 6)Muse Spark 1.2, Oct 25: 100% (n = 6)Muse Spark 1.2, Oct 26: 100% (n = 6)Muse Spark 1.2, Oct 27: 100% (n = 6)Muse Spark 1.2, Oct 28: 100% (n = 6)Muse Spark 1.2, Oct 29: 100% (n = 6)Muse Spark 1.2, Oct 30: 100% (n = 6)
GPT-5Claude Sonnet 5Gemini 3.5 FlashGrok 4.5Muse Spark 1.2
Data

How do chatbots’ citation patterns differ by topic category?

Share of answers with a citation by question category.

Data

What sources do chatbots cite?

What kinds of sources do chatbots use?

Share of each chatbot’s citations by type of publisher.

20%
63%
18%
30%
11%
24%
10%
18%
28%
18%
12%
12%
23%
32%
19%
13%
13%
56%
9%
News and mediaEncyclopedias and referenceAcademicCommercial contentGovernmentNonprofits and think tanksCompaniesSocial media and videoForums and Q&ABlogs and newslettersData and toolsLow provenanceNot yet classified
Data

Editorial, institutional or user-generated?

Citations grouped into the earned, owned and social split used in search studies.

25%
71%
51%
39%
10%
32%
51%
14%
37%
52%
11%
29%
59%
12%
EditorialInstitutionalUser-generatedOther
Data

How credible are the cited news sites?

Average Lin et al. (2023) domain-quality score of cited news-like sites, from 0 (low) to 1 (high). Only rated domains count.

GPT-50.91
40 rated citations
449 rated citations
377 rated citations
416 rated citations
88 rated citations
Data

What are the most-cited websites?

The 15 most-cited domains overall, and each chatbot’s share of its own citations that go to them.

GPT-5Claude Sonnet 5Gemini 3.5 FlashGrok 4.5Muse Spark 1.2
nasa.gov
–
2%
6%
6%
26%
cdc.gov
17%
<1%
4%
3%
3%
tfah.org
2%
8%
3%
1%
–
who.int
4%
<1%
3%
4%
12%
ipcc.ch
–
<1%
3%
6%
<1%
theicct.org
–
3%
<1%
4%
<1%
gao.gov
15%
<1%
1%
2%
–
climate.gov
–
2%
2%
3%
4%
nature.com
1%
5%
–
1%
1%
wikipedia.org
–
2%
3%
1%
1%
bmj.com
6%
–
1%
2%
2%
pmc.ncbi.nlm.nih.gov
7%
1%
–
2%
1%
mit.edu
–
1%
2%
1%
1%
ncbi.nlm.nih.gov
<1%
5%
–
<1%
–
sciencedirect.com
<1%
1%
–
3%
–
Data

Do different chatbots cite the same websites?

Overlap between the sets of domains two chatbots cite (Jaccard: shared domains over all domains either cites).

GPT-5Claude Sonnet 5Gemini 3.5 FlashGrok 4.5Muse Spark 1.2
GPT-5
–
8%
7%
9%
7%
Claude Sonnet 5
8%
–
19%
21%
15%
Gemini 3.5 Flash
7%
19%
–
17%
10%
Grok 4.5
9%
21%
17%
–
14%
Muse Spark 1.2
7%
15%
10%
14%
–
Data

How many unique websites do chatbots cite?

Distinct domains cited in the selection.

Data

How do chatbots use what they cite?

What are citations used for?

Each citation coded by the job it does in the answer.

77%
11%
55%
31%
13%
58%
14%
15%
13%
46%
25%
11%
15%
52%
17%
19%
12%
Evidence for a claimA statisticWho holds a viewBackgroundFurther reading
Data

What kind of a page gets cited?

A specific article, or a home page that cannot back a specific claim.

49%
38%
69%
20%
ArticleHome or section pageData pageReference entryOfficial documentSocial or video postOther
Data

How often do chatbots say that the question is contested?

Share of answers that say experts or groups disagree.

Data

How often do chatbots cite sources that disagree?

Share of answers with two or more citations whose sources give conflicting facts or conclusions.

Data

How do chatbots handle yes-or-no questions?

Average position: +1 is a clear yes, −1 a clear no, 0 both sides. Tap a cell for the answers.

Data

Special reports