What kinds of sources do LLMs cite most?
What kinds of sources ChatGPT, Gemini and Perplexity cite most, according to the studies: why Reddit, Wikipedia and YouTube dominate, and what to do about it.

If AI recommends brands based on what it reads on the web, the key question is: where does it get that information? Several studies in 2025 and 2026 analyzed millions of citations from ChatGPT, Gemini, Perplexity and AI Overviews, and the pattern is fairly clear.
According to the available studies, LLMs mostly cite community and encyclopedic sources: Reddit and Wikipedia consistently appear at the top, followed by YouTube. But there are two important nuances: no single source dominates absolutely (the distribution is long-tail, across thousands of domains), and the source that matters most to you depends on your industry, not the overall ranking.
What the data shows
Several independent analyses agree on the big picture:
- Reddit is usually the most-cited source. Peec AI's analysis of 30 million sources ranks it #1 or #2 across every model it tested.
- Wikipedia fights for the top spot. A Similarweb study of around 600,000 citation events found that Reddit and Wikipedia each account for roughly 12-13% of ChatGPT citations.
- YouTube is everywhere. It appears across platforms, and AI uses both the description and the auto-generated transcript of videos.
- But none dominates. Combining five studies, Contently notes that even the most-cited domain rarely exceeds 5% of total citations; the rest spreads across a long tail of thousands of domains.
Why AI trusts these sources
The pattern makes sense: AI favors content perceived as authentic and community-validated (Reddit) and broad encyclopedic coverage (Wikipedia) over marketing content. Reddit also has a question-and-answer structure similar to how people talk to a chatbot. In a single answer, AI often combines the "what it is" (Wikipedia) with the "what people think" (Reddit).
The differences between engines
They don't all cite the same way:
- ChatGPT leans heavily on Reddit and Wikipedia, plus authoritative editorial media.
- Google's AI Overviews and AI Mode favor Google-owned domains and social content (YouTube, LinkedIn, Yelp, Facebook).
- Perplexity, in B2B, surfaces sources like G2.
Wikipedia, for example, weighs heavily in ChatGPT and Perplexity but little in Google's platforms.
The nuance almost no one mentions: relevance > reach
Reddit leading in aggregate doesn't mean it's your best source. As Scrunch's analysis points out, Reddit looks dominant because it appears across many industries at once; but within a specific category, vertical specialists often stand out. The lesson: don't obsess over Reddit because it's trendy, find out which sources LLMs cite specifically in your industry.
What to do about it
- Earn genuine presence where AI trusts: participate authentically on Reddit and in your niche's communities (without spamming), and make sure you have a solid Wikipedia entry if your brand is notable.
- Use YouTube: clear descriptions and transcripts of your videos can end up cited.
- Identify your key sources by industry and earn mentions there.
For the general framework, see how to appear in ChatGPT and what GEO is.
A warning about the data
These figures move fast. In late 2025, for example, ChatGPT sharply cut its citations to Reddit and Wikipedia, though they remain its two main sources. Take these numbers as trends, not fixed truths, and measure what happens in your own category.
Frequently asked questions
What's the most-cited source by AI?
In most studies, Reddit, followed closely by Wikipedia.
Do I have to be on Reddit?
It helps, but what matters is being in the sources cited in your industry; Reddit isn't always the one that moves the needle for your brand.
Does YouTube help?
Yes. AI cites videos and uses their transcripts and descriptions as a source.
Do all models cite the same way?
No. ChatGPT and Perplexity use Wikipedia and Reddit heavily; Google favors its own domains and social content.
Is this data stable?
No, it changes often. Use it as a trend and measure what happens in your category.

Written by
Federico Ergang
Cliro cofounder & CEO
Federico Ergang is cofounder and CEO of Cliro, the AI visibility and GEO platform for Latin America.
Related articles
How to write content AI wants to cite
How to write content AI wants to cite: answer first, add data and statistics, clear structure and cite sources. What the evidence says actually works.
How to appear in Perplexity
Perplexity cites only a few sources per answer. How to make your brand one of them: crawlability, extractable content, freshness and authority.
How to measure your brand's share of voice in AI
What share of voice in AI is and how to measure it: how often models mention you versus your competitors, how it's calculated and how to improve it.
