What sources back AI brand recommendations? We traced 6 million

When an AI recommends a brand, the recommendation is built on sources the brand almost never controls. We traced 6.2 million sources sitting behind AI brand mentions on ChatGPT and Google AI Overviews. Two open platforms, Reddit and YouTube, shape more brands than anything else; the sources a brand actually owns barely register; and which sources do the work depends on which AI you ask. The evidence behind a recommendation is a different map than the one most brands optimize.

The recommendation is built on sources you don't own

Parse stores, for every brand an AI names, the citation sources attached to that mention, which is the evidence layer beneath the recommendation. Across ChatGPT and Google AI Overviews between October 19, 2025 and April 24, 2026, that came to 6,207,023 brand-supporting source references spanning 14,605 prompts, 58,056 brands, and 139,499 distinct domains. More than 99% sat behind a brand being recommended for a need, not merely compared or dismissed (the gap between being mentioned and being the pick is its own measure). The top of that list is defined by what is absent: the seven most common sources behind brand recommendations are YouTube, Reddit, Medium, Wikipedia, Forbes, Facebook, and LinkedIn, every one a third-party platform. The first source a brand could call its own, an Amazon listing, ranks eighth. This matches what others measured from the outside: Idea Grove found only 9% of AI brand mentions come from a brand's own website (Idea Grove). Our first-party view of the evidence behind recommendations says the same thing, one layer down.

How we measured it

We read Parse's source-provenance records, the links its evidence pipeline draws between a reviewed AI brand mention and the citation sources behind it. Each record ties one brand mention, on one answer, to one cited domain. We scoped this to ChatGPT and Google AI Overviews, the two surfaces with a complete October 2025 to April 2026 window, and kept the 6.2 million links that resolved to a named domain. A brand-supporting source here is a domain Parse attached to a recommendation, counted as repeated references rather than unique sessions, the same definition we use in our other citation studies. About 67% of these links sit on fully approved brand observations and the rest on observations still in review; the rankings below are identical on the approved-only subset, so the mix does not move them. This is structural analysis of a Parse mirror slice, not a live-state snapshot.

Reddit and YouTube shape more brands than anything else

By reach, two platforms tower over the rest. Reddit is the source behind AI brand recommendations for 24,388 distinct brands, which is 42% of every brand in the set, across 9,642 prompts, or 66% of them. YouTube backs 19,877 brands. Each is cited roughly five times more often than the third-ranked source, Medium. That said, no single source dominates the whole surface: the citations spread across 139,499 domains, and even the top 10 account for only about 13% of all brand-supporting references. The headline is concentration at the very top over a very long tail. The two biggest are not review sites or publishers; they are an open forum and a video platform, which is where buyers argue and demonstrate in public. We dug into that video lead separately in YouTube versus Reddit as the top cited source.

Read the brand column, not just the reference column. A source that backs recommendations for thousands of different brands is shaping the category; a source with high references but few brands is concentrated somewhere specific.

If you want to see which sources shape AI answers about your brand, run a free brand check — it takes a minute.

Some sources shape every category, others own one

The reach numbers hide a second pattern: depth without breadth. NerdWallet drew 26,621 brand-supporting references but spread them across only 1,685 brands, almost all in personal finance. Bankrate showed the same shape, and FoxSports backed 5,320 references across just 139 brands, at 38 references per brand, entirely in sports. Compare that to Reddit's 9.7 references per brand across 24,388 brands. The broad platforms shape who gets recommended almost everywhere; the vertical sources shape one category intensely and nothing else. For most brands, the practical takeaway is that your category almost certainly has a NerdWallet of its own, a single source doing outsized work behind your recommendations, and it is rarely the source you would have guessed.

ChatGPT and Google build recommendations from different sources

The same brand recommendation rests on different evidence depending on the engine. Wikipedia supplied 2.5% of ChatGPT's brand-supporting sources but only 0.13% of Google AI Overviews', a roughly 20× difference in the engine's reliance on it. YouTube ran the other way: 6.0% of Google AI Overviews' brand-supporting sources against 0.2% on ChatGPT, a roughly 28× gap. Reddit and the rest of the user-generated set also weighed more than twice as heavily on Google. This is not a single-surface fluke. In the May-June 2026 window, ChatGPT Search cited YouTube just 744 times while Google AI Mode cited it 86,583 times, so the pattern holds across both of each company's surfaces. We measured the citation overlap between these two engines separately in our study of what sources shape AI answers; this is the layer beneath it, the evidence behind the recommendations themselves, and it diverges just as sharply as the brand shortlists do.

The plan that follows is not one plan. Winning ChatGPT recommendations leans on Wikipedia-grade entity authority and publisher coverage; winning Google AI Overviews leans on video and community presence. Optimizing for "AI" as one channel funds the wrong source list for at least one of them.

When the source is neutral, AI still sounds like it is recommending

On the smaller slice where Parse analyzed the cited source's own stance, about 260,000 links, a gap opens between what the source says and how the AI frames it. The AI described the brand positively in 79.7% of those mentions. But the underlying source was only neutral in 42.5% of those positive framings, and outright negative in another 3.3%. In other words, AI rounds up: it routinely turns a neutral source passage into a positive recommendation. The same disconnect appears at the level of the final verdict: when AI crowns a head-to-head winner, its own cited sources often favor someone else. Treat this as directional, not settled, because it covers only the 2.7% of links where source content was analyzed and that subset is not a random sample. The operating implication still holds. A neutral mention on a source the AI already reads can become a positive recommendation, which is why being named on the cited source matters even before you worry about sentiment, the same dynamic behind ghost citations.

What this means for your AI visibility plan

Four moves follow from the evidence. First, fund the sources AI reads, not only the site you control: the recommendation is built almost entirely from third-party platforms, so owned-media polish alone will not put you in the answer. Second, find your category's vertical source, the NerdWallet or FoxSports doing concentrated work behind your recommendations, and earn presence there before chasing broad reach. Third, split the plan by engine, because Wikipedia-grade entity work moves ChatGPT while video and community presence move Google AI Overviews, and the same effort does not serve both. Fourth, get named inside the sources already cited for your category, even the neutral ones, since AI tends to frame a named brand favorably once it appears in the evidence at all. None of this is fast, and none of it is a single placement; it is steady presence in the specific sources the model already trusts.

How Parse maps this to your sources

Parse tracks AI visibility across ChatGPT and Google AI Overviews, covering a public index of more than 4.7 million AI responses, 603,000 brands, and 57 million citations. The rankings in this study are the industry-wide pattern; they are not yours. The sources behind recommendations for a personal-finance app are not the ones behind a developer tool or a hospitality brand, and the split between ChatGPT and Google AI Overviews shifts again on top of that. Parse's Citations view surfaces the exact sources AI draws on when it names your brand and your competitors on each engine, ranks them by how often they back a recommendation, and shows where a competitor is named in a source and you are not. For the wider source map beyond the evidence behind recommendations, our breakdown of which domains AI models cite most and our community-to-citation pipeline cover the rest of the surface. See which sources cite your category.