Ask where AI engines get their answers and you get a short list that repeats across every major study: Reddit, Wikipedia, YouTube, LinkedIn, and a band of review sites and editorial publishers. Peec AI's analysis of 30 million cited sources found Reddit was the most-cited domain across ChatGPT, Google AI Mode, Gemini, Perplexity, and AI Overviews, with YouTube, LinkedIn, Wikipedia, and Forbes rounding out the top five. But the more useful finding sits one level deeper: each platform has a distinctly different source diet, the diets barely overlap, and they change fast enough that last year's map is already wrong.
This piece pulls together what the 2025 and 2026 citation studies actually show, platform by platform, and what it means for where you spend effort.
First, why every study reports a different number
Before quoting any percentage, it pays to know that "most cited" gets measured three different ways, and the three produce wildly different figures for the same platform.
Some studies measure the share of answers that cite a domain at least once. Others measure a domain's share of all citations. Others measure its share within a platform's top 10 sources only. Reddit illustrates the spread perfectly: it has been reported at roughly 40% (share of answers citing it, in a June 2025 study of 150,000+ citations), at 46.7% (share of Perplexity's top-10 sources, in 5W's synthesis of 680 million citations), and at under 5% (share of total citations, in Evertune's analysis of 200 million prompts, which found no single domain exceeds about 5% of all citations). None of those numbers is wrong. They are answering different questions.
The honest summary: a handful of domains appear constantly, but the citation ecosystem is a long tail. Analyze AI's August 2026 study found the top 10 cited domains cover only 11 to 13% of citations on each engine, with roughly 7,000 more domains sharing the rest. The outliers matter enormously for how AI describes categories, but most citations come from everywhere else.
Which sources does each platform cite most?
ChatGPT: the reference engine
ChatGPT leans on encyclopedic and editorial sources harder than any other platform. Wikipedia is its most-cited domain across essentially every study: 47.9% of ChatGPT's top-10 source share in 5W's synthesis, 16.3% mention share in Ahrefs' analysis of nearly a million prompts, and 7.8% of total citations in Profound's 680-million-citation dataset. Three lenses, one conclusion: Wikipedia matters more to ChatGPT than to anyone else. Reddit and editorial publishers like Forbes follow, and on branded questions, review platforms enter the picture.
ChatGPT is also the most volatile of the big platforms. Semrush tracked 230,000 prompts weekly through late 2025 and watched Reddit fall from being cited in roughly 60% of ChatGPT responses in early August to around 10% by mid September, with Wikipedia dropping from about 55% to under 20% in the same window. A source strategy built on one platform's preferences in one quarter can be stale by the next. Our guide to how ChatGPT chooses sources covers the retrieval mechanics behind this.
Perplexity: the community engine
Perplexity concentrates on community and professional platforms more than any rival. Reddit accounts for 46.7% of its top-10 source share, and by some counts around one in five of all its citations, the highest concentration of any domain on any platform. LinkedIn, G2-class review sites, and government and scientific sources fill out its diet, especially on B2B and technical questions. Perplexity also cites more sources per answer than anyone else, typically five to twelve, which spreads opportunity wider.
Google AI Overviews and AI Mode: the Google-ecosystem engines
Google's two AI surfaces eat differently from the chatbots and from each other. YouTube is the standout: one analysis puts it at 29.5% of AI Overviews citations, and Google-owned properties collectively absorb a large share of AI Mode's citations. Semrush's tracking found AI Mode consistently citing LinkedIn in nearly 15% of responses. Both surfaces draw on Google's index, but they diverge from classic rankings: Ahrefs' March 2026 study of 863,000 SERPs found only 38% of AI Overviews citations come from Google's top 10 organic results, a product of query fan-out breaking one question into many sub-searches. If you treat these as one surface, read our explainer on how they differ.
The overlap problem
Here is the number that should reshape strategy: only about 11% of domains are cited by both ChatGPT and Perplexity. The platforms are not sampling the same web differently. They are, in practice, reading different webs. Winning one engine's source diet tells you very little about the others, which is the same conclusion our engine comparison reached from the answer side.
What about your own website?
Less than you would hope, especially on branded questions. Omniscient Digital's study of 23,387 citations found that on branded queries, 57% of citations come from reviews and social proof, 17% from directories, and only about 4.5% from the brand's own pages. When someone asks an AI about you specifically, the answer is mostly assembled from what third parties say. Your site anchors the facts; it does not carry the verdict.
That is the strategic heart of this whole topic. AI visibility is earned in the sources each engine trusts, which means the work looks more like PR and presence-building than like traditional on-page optimization. It is why we treat source and citation work as a PR function as much as an SEO one.
What should you actually do with this?
Three moves follow directly from the data.
Map your own citation mix before copying anyone's percentages. The studies above describe averages across millions of generic prompts. Your category behaves differently: SaaS questions skew toward G2 and Reddit, health defers to institutions, finance rewards named authorities. The only citation data that should drive your plan is the citation data behind answers about your category and your brand.
Earn presence where your engines actually eat. If Perplexity matters for your buyers, Reddit and review platforms are not optional. If AI Overviews matter, YouTube and your Google footprint carry weight. If ChatGPT matters, Wikipedia eligibility and editorial coverage are the long game.
Expect the diet to change, and watch it. The Reddit and Wikipedia swings on ChatGPT happened in six weeks. Whatever mix is true today is a snapshot, not a constant.
Diploria tracks exactly this: every poll stores the full response and every citation behind it, classified by site type and page type, per engine. The Citations view shows which domains and pages are feeding answers about your brand and your competitors, so your source strategy runs on your data rather than industry averages. You can see it on your own brand in minutes with the free AI visibility check.