Company websites win half of all AI citations

Across 160,240 citations from five B2B SaaS brands, company websites won half of everything AI cited (50.7%). One domain still beat them all. Reddit was the single most-cited source in the dataset at 11,237 citations, more than twice the next domain. So corporate content wins the category, and a community forum wins the leaderboard. Community sources also turn a fetch into a citation far more often, which is one reason AI search behaves like a dark funnel channel.

That tension is the whole story. Below is what we measured, every table behind it, and what you should change about your own AEO work because of it.

Finding 1: company websites win half of everything

Corporate content is the most-cited source type by a wide margin. Across all three engines, owned-and-operated company sites took 50.7% of citations. The next type, user-generated content, sat at 13.2%. Nothing else cleared 12%.

This held in every single project. Corporate ranged from 41.6% to 66.9% across the five brands, and it was the top type in all of them. The ordering below Corporate shifted (UGC and Editorial traded places depending on the client), but the headline never moved. If an AI answer cites a single source about your category, the odds favor a company website saying it.

That is the good news for anyone who owns a website. You are not fighting Wikipedia for every slot. You are competing inside the source type that already wins. The bad news is that a cited page can still carry a stale or mis-attributed fact about your company.

Finding 2: Reddit is the single most-cited domain, and it is a ChatGPT habit

Source type is one lens. Individual domains are another, and they tell a sharper story. Corporate wins as a category because thousands of company sites add up. No single corporate domain dominates. Reddit does.

Reddit took 7.0% of all citations on its own. The second-place domain, toku.com, took 3.6%. So Reddit out-cited the next domain by more than 2x.

Look at the "top engine" column and the pattern jumps out. Reddit's dominance is 80.5% driven by ChatGPT specifically. Of Reddit's 11,237 citations, 9,050 came from ChatGPT, and ChatGPT cited Reddit at a rate of 2.64 citations per retrieval. Perplexity barely touched it, contributing 335 citations against 1,252 retrievals. Same forum, completely different treatment depending on which engine answers.

Finding 3: the three engines have different personalities

If you optimize for "AI search" as one thing, you are averaging across machines that disagree. Each engine leans on a different source mix.

ChatGPT leans hardest on UGC at 15.0%, and most of that is Reddit. It also pulls Reference content more than the others (10.3%). Perplexity goes the other way: it is the most editorial engine at 13.6% and the lightest on UGC at 8.4%. Google AI Overviews is the most corporate-heavy at 53.4% and pulls the least Reference at 3.5%.

The Corporate-first rule held for every engine in every project, and ChatGPT's UGC tilt was consistent across clients. So the floor is the same everywhere (own your category page), but the edges reward different moves per engine.

Finding 4: owned pages get cited hardest when they get pulled

Citation share tells you how often a source shows up. It does not tell you how efficiently a page converts attention into a citation. For that we use citation rate, which is citations per retrieval: when an engine pulls a page into its working set, how reliably does it actually cite it.

Owned pages lead at a 2.04 citation rate, far ahead of every other type. When an engine reaches a well-structured owned page, it cites it roughly twice per retrieval. Competitor pages (0.95) and UGC (0.90) come next. Corporate-at-large sits lower at 0.72, which makes sense: the corporate bucket includes a lot of pages that get crawled and ignored.

This shows up in real ranks, not just averages. Three of our five brands are a top-2 cited source in their own category. Toku ranks #1 of 1,425 domains in its space. LoudFace ranks #2 of 2,160. The anonymized B2B research client ranks #2 of 1,302. Owning a domain that AI trusts is not theoretical. It is happening for most of the brands we measured.

Finding 5: being cited is not the same as being retrieved

The last column of that table is the one most people miss. Every source type was cited more than it was retrieved, which means citation and retrieval are not the same signal.

UGC has the highest retrieval rate of any type (0.413), so engines pull community content into context constantly. But its citation rate (0.90) is lower than owned pages, so a lot of what gets retrieved never makes the answer. Owned pages flip that: a smaller retrieval rate (0.775) and a much higher citation rate (2.04). Engines reach for them less often, but when they do, the page earns the link.

The practical read: chasing retrieval (getting crawled, getting pulled) is a different job than chasing citation (getting quoted). A page can be retrieved all day and cited rarely. You want both, and the two levers are not the same page edits.

What this means for your AEO

Three moves come straight out of the data.

First, get into the Reddit conversation if you care about ChatGPT. ChatGPT drives 80.5% of Reddit's citations and cites the forum at 2.64 per retrieval. You cannot fake your way into that with a marketing account, but you can make sure your category's real Reddit threads are accurate, current, and mention you where it is honest to. For ChatGPT specifically, the community layer is part of the answer. Our guide on how to become a trusted LLM source goes deeper on building that off-site trust.

Second, structure owned pages to be quoted, not just crawled. Owned pages cite at 2.04 per retrieval, the highest of any type, but only if the engine can lift a clean, self-contained answer off the page. Short definitional blocks, direct claims, and clear headers do this. See how to structure content for AI extraction for the format that earns the quote.

Third, optimize per engine instead of for "AI" as one blob. Perplexity wants editorial-grade sourcing (13.6% editorial), ChatGPT wants community and reference signals, Google AI Overviews wants authoritative corporate pages (53.4%). A page tuned for one is not automatically tuned for the others. The full method is in our answer engine optimization guide.

Run this for your brand

We measure exactly this for the clients we run AEO for, per engine, per source type, per competitor, every month. If you want to see where your domain ranks in your own category and which engines are ignoring you, book an AEO audit. You can also read how we got Toku to a top-cited B2B pipeline or see which agencies show up in ChatGPT and Perplexity citations. Or see the buyer questions no agency is winning in AI search. And if you want a free look at your other channels, Search Console now tracks connected Instagram, TikTok, X, and YouTube posts as platform properties.