We asked four AI search engines the same 100 questions. Only the one Google owns kept reaching for the platform Google owns.
Ask an AI assistant a question and it hands you an answer with a tidy list of sources, which makes the answer feel neutral and settled. It is not neutral. We put the same 100 everyday questions to the four AI search engines that show their sources, traced every citation back to the real site behind it, and found that one engine keeps reaching for a website its own parent company owns. Google’s Gemini cites YouTube far more than any rival does. The others cite it almost never.
We asked four AI search engines the same 100 everyday questions, the kind people actually type: best noise cancelling headphones, how to lower cholesterol, best budgeting apps. We used only the engines that retrieve and show real web sources, because those are the citations you can actually check: Google’s Gemini (with search grounding), Microsoft’s Copilot (Bing), OpenAI’s ChatGPT search, and Perplexity. Then we traced every cited link to the real domain behind it. Gemini in particular hands back Google redirect tokens rather than plain links, so we followed each one to the site it really points to.
The gap is not subtle. Across those 100 questions Gemini cited YouTube 40 times. Every other engine combined cited it once. YouTube was the single most-cited website in Gemini’s answers, and it does not appear near the top of any rival’s list.
Put as a share: 6.7% of everything Gemini cited was a YouTube link, rising to 7.9% once you add Google’s own pages. Gemini reached for YouTube on 24 of the 100 questions. For ChatGPT search and Copilot that figure was zero. For Perplexity it was one link out of more than a thousand. Google owns Gemini, and Google owns YouTube, and Google’s AI cites Google’s video platform at a rate its competitors round to nothing.
The lean is not uniform, and where it disappears says a lot. On health questions Gemini’s YouTube share dropped to about 1%. Asked how to lower cholesterol or reduce inflammation, it deferred to Mayo Clinic, Harvard Health and the NHS like everyone else. The YouTube habit showed up on commercial and how-to questions instead: consumer electronics (11% of its citations there), personal finance (10%), marketing (10%), home and lifestyle. In the categories where a review channel or a demo is plausible, Gemini reaches for Google’s platform. In the categories where authority matters most, it does not.
This is the part that decides whether to trust the result. We ran the same ownership check on all four engines, and only Google’s showed it. Microsoft owns Copilot, yet Copilot sent just 0.3% of its citations to Microsoft-owned sites. OpenAI and Perplexity do not own a content platform to favour, and their self-citation was effectively zero. If our method simply flattered whoever owned the engine, Copilot would lean on MSN and LinkedIn the way Gemini leans on YouTube. It did not. The pattern is specific to Google.
An AI engine’s list of sources looks like a neutral read of the web. It is partly a read of its owner’s interests. That has a direct consequence for anyone trying to be found in AI answers: there is no single "AI visibility." Being cited by Gemini and being cited by ChatGPT are different games, played on different webs, and in Gemini’s case the home advantage runs through YouTube. If your audience leans on Google’s AI, a real presence on YouTube is part of your visibility there in a way it simply is not for ChatGPT or Perplexity.
It also explains why a single blended "AI visibility score", the number most tools sell, cannot be trusted. You cannot average an engine that favours its owner’s video platform with one that has no platform to favour and get a figure that means anything. The only honest unit is per engine, with the receipt attached: which assistant cited you, in which answer, linking to what. That is why a single score is a guess, and it is the gap Unsourced exists to close.
AI answers arrive looking impartial. Under the hood, at least one of the big engines keeps a quiet home-field advantage for a site its parent owns. We are not claiming intent, only reporting what the citations show. But it is a reminder that the sources an AI shows you are a product of who built it, not a neutral survey of the web, and that the only way to know where you really stand is to measure each engine directly and keep the evidence.
In our study, yes, and by a wide margin. Across 100 identical questions Gemini cited YouTube 40 times, 6.7% of all its citations and its single most-cited domain, while ChatGPT search and Microsoft Copilot cited it zero times and Perplexity once. Google owns both Gemini and YouTube.
We measured the behaviour, we did not audit the cause, so we do not claim intent. What is factual: Google owns YouTube, Gemini’s grounding runs on Google’s own search stack, and video is woven through Google surfaces. Whatever the mechanism, the outcome is that Google’s AI routes a meaningful share of its citations to Google’s own platform.
No, not in our data. Microsoft’s Copilot sent only 0.3% of its citations to Microsoft-owned sites. OpenAI and Perplexity own no content platform to favour and self-cited effectively zero. The lean toward an owner’s property was specific to Google, which is part of why we are confident it is a real pattern and not a quirk of the method.
That there is no single AI visibility to chase. If your audience uses Google’s AI, a genuine YouTube presence can matter there in a way it does not for ChatGPT or Perplexity. The practical move is to measure each engine on its own and keep the receipt, rather than trusting one blended score.
We sent 100 neutral everyday questions to the four AI search engines that return real sources (Gemini with grounding, Copilot/Bing, ChatGPT search, Perplexity), captured every cited URL, followed redirects to the true domain, and classified each by owner using only wholly-owned, publicly known properties. ChatGPT search was sampled on 40 of the 100 questions due to its rate limit; the others ran all 100. Snapshot, September 2026; every figure measured, not modelled.
The engines already read different webs. This is the same story with an owner’s thumb on the scale.
Read it →You cannot average an engine that favours its owner’s platform with one that has none.
Read it →See which engines actually cite your domain, per engine, with the answer attached.
Read it →Unsourced captures which AI assistants cite you, proves which crawlers really fetched your pages, and re-checks after you act — evidence, not a score.
© Unsourced, the evidence layer for AI search. Our data and findings are free to quote and cite — please attribute to Unsourced and link to unsourced.app.