HomeResourcesCitation data

Citation data

The Citation Fingerprint: Where ChatGPT, Perplexity, and Google AI Overviews Get Their Sources

Ask three AI engines the same question and they will lean on very different corners of the web. We broke down 680 million citations to show each engine’s source mix.

By MentionMethod 6 min read
7.8%of ChatGPT citations go to Wikipedia
6.6%of Perplexity citations go to Reddit
83%of the time, AI Overviews and AI Mode cite different domains
Key takeaways
  • Each AI engine has a distinct source “fingerprint.” Wikipedia leads ChatGPT’s citations, Reddit leads Perplexity’s, and Google AI Overviews spread across Reddit, YouTube, Quora, and LinkedIn.
  • Even the most-cited sites are a small slice: an engine’s top 10 domains add up to only 10% to 16% of its citations. Most citations come from a long tail of publishers.
  • Engines also differ in how much they cite. ChatGPT cites in 96% of responses with about five links; Claude cites in 55% but averages 13 sources when it does.
  • Even two Google surfaces disagree: AI Overviews and AI Mode cite different domains 83% of the time.

One question, three different source lists

It is tempting to talk about “AI search” as one channel. The citation data says otherwise. When ChatGPT, Perplexity, and Google’s AI Overviews answer a question, they lean on noticeably different parts of the web, and those preferences are consistent enough to describe as a fingerprint.

The largest public dataset on this comes from Profound, which analyzed 680 million citations collected between August 2024 and June 2025.1 The figure below shows each engine’s ten most-cited domains as a share of all of its citations. Switch between engines to see how the mix changes.

Figure 1Top 10 most-cited domains, by AI engineShare of each engine’s total citations. Colors show the type of source.
Reference Community and social Video Publishers and news Reviews and research
Top 10 total: 16.3% of ChatGPT’s citations. Hover the legend to highlight a source type.
Source: Profound, AI Platform Citation Patterns (June 2025), 680M citations, August 2024 to June 2025. Source-type labels are MentionMethod’s classification.
View the data
RankChatGPTGoogle AI OverviewsPerplexity
1Wikipedia 7.8%Reddit 2.2%Reddit 6.6%
2Reddit 1.8%YouTube 1.9%YouTube 2.0%
3Forbes 1.1%Quora 1.5%Gartner 1.0%
4G2 1.1%LinkedIn 1.3%Yelp 0.8%
5TechRadar 0.9%Gartner 0.7%LinkedIn 0.8%
6NerdWallet 0.8%NerdWallet 0.6%Forbes 0.7%
7Business Insider 0.8%Forbes 0.6%NerdWallet 0.6%
8New York Post 0.7%Wikipedia 0.6%TripAdvisor 0.6%
9Toxigon 0.7%Business Insider 0.5%G2 0.6%
10Reuters 0.6%Medium 0.4%PCMag 0.5%

Reading the three fingerprints

ChatGPT: the reference librarian

ChatGPT’s single largest source is Wikipedia, at 7.8% of all citations, more than four times its next source. The rest of its top 10 is dominated by established publishers such as Forbes, Business Insider, and Reuters, plus review and comparison sites like G2 and NerdWallet. It favors sources that read like reference material or reporting.

Perplexity: the community researcher

Perplexity leans hardest on people talking to people. Reddit alone accounts for 6.6% of its citations, and YouTube adds another 2.0%. Its list is also full of review and research platforms where real users and analysts weigh in: Gartner, Yelp, TripAdvisor, G2.

Google AI Overviews: the generalist

Google’s AI Overviews have the flattest profile. No single domain passes 2.2%, and the top of the list is a mix of user-generated platforms: Reddit, YouTube, Quora, and LinkedIn. Together its top 10 make up only 10.3% of citations, the lowest concentration of the three.

84 to 90%

of each engine’s citations come from outside its top 10 domains. The long tail of publishers, niche sites, and individual pages is where most AI answers are actually sourced.Calculated from Profound’s top-10 shares: 16.3%, 14.2%, and 10.3%.

That last point is easy to miss. The big names get the attention, but even for ChatGPT, the most concentrated engine, the top 10 domains supply only about one in six citations. A brand does not need to be on Wikipedia or Reddit to be cited. It needs to be present across the many specific pages that engines pull from for questions in its category.

How often, and how much, each engine cites

Source mix is only half the fingerprint. Engines also differ in how often they cite sources at all, and how many they include. Muck Rack’s May 2026 What Is AI Reading? report measured this across more than 25 million links.2

Figure 2Citation habits by engineRing: share of responses that include citations. Dots: average sources per response that cites.
ChatGPT
of responses cite sources
about 5 sources each
Gemini
of responses cite sources
about 8 sources each
Claude
of responses cite sources
about 13 sources each
Source: Muck Rack, What Is AI Reading? (May 2026), 25M+ links across 17 industries.
View the data
EngineResponses with citationsAvg. sources when citing
ChatGPT96%5
Gemini82%8
Claude55%13

ChatGPT cites almost every time but keeps lists short, so each slot is competitive. Claude cites less often but casts a wide net when it does, which gives more sources a chance to appear. These habits shape how hard it is to break into an engine’s answers, not just where to show up.

Even Google disagrees with Google

If any two AI surfaces should agree, it would be two built by the same company on the same index. SISTRIX tracked Google’s AI Overviews and AI Mode across 82,619 prompts for 17 weeks and found the two cite different domains 83% of the time, with an overlap score (Jaccard index) of just 0.17.3

Figure 3Domain overlap: AI Overviews vs. AI ModeTwo Google AI surfaces, mostly different sources.
AI Overviews AI Mode 17% overlap

In 83% of cases the two surfaces drew on different domains. Optimizing for one Google AI experience does not guarantee visibility in the other.

Source: SISTRIX, AI Citation drift (May 2026), 82,619 prompts, 17 weeks, six countries. Circle overlap is illustrative of the 0.17 Jaccard index.

A note on how fast this changes

Citation fingerprints are not permanent. Engines change their retrieval systems and partnerships, and source shares can swing quickly as a result. Profound’s figures cover August 2024 to June 2025, and newer datasets may show different leaders. Treat the specific percentages as a snapshot and the overall pattern, that engines favor different source types, as the durable lesson.

What brands and agencies should do with this
  1. Find out which engines your buyers use. A B2B audience researching in ChatGPT and a consumer audience asking Perplexity for local recommendations need different source strategies.
  2. Match your presence to each fingerprint. For ChatGPT, prioritize reference-style coverage and reputable publishers. For Perplexity and AI Overviews, invest in community, video, and review platforms.
  3. Do not chase only the giants. With 84% to 90% of citations coming from outside the top 10 domains, coverage on the specific niche sites your category relies on can matter as much as one big placement.
  4. Audit each engine separately. Since even AI Overviews and AI Mode cite different domains 83% of the time, measure visibility per engine instead of assuming one result represents all of them.

Frequently asked questions

Which AI engine cites Reddit the most?

In Profound’s dataset, Perplexity, where Reddit made up 6.6% of all citations, its single largest source. Reddit was also the top source for Google AI Overviews at 2.2%.

Why does ChatGPT cite Wikipedia so often?

ChatGPT tends to favor reference-style and editorial sources. Wikipedia accounted for 7.8% of its citations, more than four times any other single domain in the same dataset.

How many sources does an AI answer usually cite?

It depends on the engine. Muck Rack found ChatGPT averages about 5 citations per response, Gemini about 8, and Claude about 13 when it cites at all.

Do Google AI Overviews and AI Mode cite the same websites?

Mostly not. SISTRIX found the two surfaces cite different domains 83% of the time.

Sources

  1. Profound, “AI Platform Citation Patterns,” June 5, 2025. tryprofound.com
  2. Muck Rack, What Is AI Reading? report, May 2026. Press release
  3. Johannes Beus, “AI Citation drift: How stable are sources in AI search results?” SISTRIX, May 2026. sistrix.com
Written byMentionMethodMentionMethod helps brands understand and grow how they show up in AI answers. We publish data-driven research on AI visibility, citations, and brand mentions.
Learn more about MentionMethod

Convert ad spend to mentions in AI answers.

Built by ex-Facebook, LinkedIn, and IBM engineers
Influence answers across ChatGPTGeminiPerplexityClaudeCopilot

© 2026 MentionMethod. A product of OptalX, LLC. All rights reserved.