We spend a lot of time watching what Ai assistants actually say when people ask them for a recommendation. One thing surprised us early on: you'd assume ChatGPT, Perplexity, Gemini and Grok mostly pull from the same handful of sources. They don't. They have remarkably different taste — and that changes how you'd try to show up in any one of them.
It's easy to think of “getting cited by Ai” as one job. It's really four jobs, because the assistants don't shop at the same stores.
The same short list keeps showing up
Short version: Across the studies, a few kinds of sites do most of the citing — and almost none of them are a company's own website.
Beyond Reddit, the names that keep appearing are YouTube, LinkedIn, Wikipedia, and established publications. The thread running through all of them is simple: they're places where other people talk about a business, not places the business controls. That's the first thing worth sitting with — being referenced tends to matter more than being polished. (We dug into that idea in why Ai recommends some businesses and not others.)
Each assistant has its own taste
Short version: These are tendencies from public studies, not promises — but they're consistent enough to plan around.
ChatGPT leans toward authoritative, encyclopedic sources — Wikipedia-grade references and well-known publications — plus Reddit. Interestingly, most of what it cites reportedly comes from outside Google's top results, so “trusted” beats “ranks first.” Perplexity is the Reddit-heavy one, and it runs a fresh web search on every question, so recent, genuinely useful pages tend to surface. Google's Ai Overviews and Gemini reach for YouTube and for clear pages they can lift a self-contained passage from — less about your old ranking than you'd think. And Grok reads X in real time and quotes posts there directly.
There isn't one door into Ai answers. There are four, and they open with different keys.
What this means for you
Short version: If you're missing from just one assistant, the fix is usually specific to a source you might not guess.
Because the sources differ, a move that helps in one place can do little in another. A real, helpful presence in community discussions tends to track with Perplexity. Authoritative third-party mentions track with ChatGPT. A clear, well-structured video or page tends to track with Google's Ai answers. So the smart play isn't to do everything everywhere — it's to find which assistant is leaving you out, and work the source that one actually trusts.
Where these numbers come from: the figures here are from public 2026 research that aggregated Ai citations at scale — the AI Platform Citation Source Index (about 680 million citations, August 2024 to April 2026), Semrush's most-cited-domains analysis, and reporting from Search Engine Land. Studies disagree on the exact percentages and methods differ, so we treat them as directional, not gospel — the pattern (sources differ sharply by assistant) is the durable part.