We pulled 5,676 real AI citations. Here's who AI actually cites

Most "what AI cites" posts quote someone else's study. This one is our own data, with the caveats in plain sight. We pulled every citation our scanners recorded over a recent nine-day window - June 16–24, 2026 - across the brands and questions our users actually track. That's 2,209 answers, 5,676 citations, 1,616 distinct domains. Here's what it actually looks like under the hood.

Finding 1: most AI answers cite nothing

Only 39% of answers carried a single citation. The other ~60% named brands, made comparisons and gave recommendations with no visible source at all - answering from memory, not the live web. If your mental model is "AI shows its work," most of the time it doesn't. That alone reframes the channel: you can't earn your way into a majority of answers by being citeable, because those answers were never going to cite anyone. It's also why mentions and citations have to be tracked as two different things.

Finding 2: whether you get cited is the engine's decision, not yours

The single clearest pattern is how violently the citation rate splits by engine:

• Perplexity - 100% of its answers cited at least one source (and ~9 citations per answer). It is a pure search-and-cite engine; it essentially never answers without showing receipts.
• ChatGPT - ~90% cited sources, averaging ~6 per sourced answer. When it searches, it cites generously.
• Claude - ~67% cited sources. More selective about when it reaches for the web.
• Gemini and Grok - 0% in this dataset. Important honesty note: that's a configuration artifact - those engines ran without live web search in our default scans, so they answered purely from memory. It is not evidence that those engines "never" cite; it's evidence that an engine with web search off cites nothing, which is exactly the point.

So "can I get cited?" has no single answer. On Perplexity, almost always; on a memory-only engine, never. The same brand, the same question, a completely different game depending on who's answering.

Finding 3: the recognisable head is exactly what you'd guess

Filter to the cross-category domains anyone would recognise, and the leaderboard is unsurprising in a reassuring way - these are the surfaces worth being present on:

• Tech-review publishers - TechRadar (the single most-cited domain at ~2.4%), Android Central, Tom's Guide, MakeUseOf, TechTarget.
• Community & video - YouTube (~2.1%), Reddit (~1.3%), Instagram, Facebook.
• Reference - Wikipedia (~2.0%).
• Stores & aggregators - Google Play, the App Store, AlternativeTo.

Nothing exotic: established review sites, the big community platforms, the encyclopedia, the app stores. If you want to influence the citeable answers, those are the doors - and almost none of them are your own website.

Finding 4: the long tail is enormous

Here's the stat that surprised us most. 5,676 citations spread across 1,616 domains - and the most-cited domain in the entire set accounts for just 2.4% of citations. There is no monopoly. After the recognisable head, citations fan out into hundreds of niche, industry-specific and regional sites, each appearing a handful of times. For a specialised category, the page that wins the citation is often a small site that happens to answer the exact question well - which is genuinely good news if you're not a big brand.

What we'd do with this

Stop obsessing over getting your homepage cited; engines cite third parties far more than first parties. Instead: make sure your presence on review, reference and community sites is accurate and current, then own the long-tail questions in your category with pages built to be quoted. And track it per engine - because as finding 2 shows, an average hides the only number that matters.

Want this view for your own brand instead of our aggregate? A free Zene audit shows exactly which domains the engines cite when they answer questions about you - and which of them belong to your competitors.