Gemini and Perplexity lean heavily on YouTube as a source, far more than ChatGPT or Claude. There’s a structural reason, and it’s showing up in Google’s AI Overviews too.

Growth Feature in our dashboard

Something I keep noticing about where different AI assistants pull their answers from, and I think there's a clean mechanistic explanation worth discussing.

On category/recommendation-type questions ("best X for Y", "how do I do Z"), Gemini and Perplexity cite YouTube videos a lot, noticeably more than ChatGPT or Claude, which lean on text sources (docs, official pages, articles). It's a pretty lopsided split by engine on the same prompts.

Why it likely happens:

• Gemini is a Google product, and Google owns YouTube. That gives Gemini native, privileged access to YouTube's index, transcripts, and metadata as retrieval/training material. Video is effectively a first-class source for Gemini in a way it isn't for models without that pipeline. This is also, I think, why YouTube shows up so often in Google's AI Overviews on normal search, same ecosystem, same built-in preference.
• Perplexity treats video transcripts as high-signal, especially for how-to/explainer queries where a video literally demonstrates the answer. Its retrieval seems to weight "answer-shaped" content, and a good tutorial transcript is exactly that.
• ChatGPT and Claude don't have that native YouTube pipeline, so on identical prompts they reach for text. Hence the split.

Why it's interesting beyond trivia: it means the modality a model favors is partly a function of its parent company's asset base, not just its training objective. Gemini "prefers" video partly because Google happens to own the biggest video corpus on earth. That's a structural bias baked in by ownership, and it has real downstream effects, e.g. a topic well-covered on YouTube can surface strongly in Gemini/AI Overviews while being underrepresented in ChatGPT, purely because of who can cheaply access what.

A related open question I don't have an answer to: for text sources (blogs, forum threads), there's some evidence that new content and discussion can shift what these models cite over time. I haven't seen anything on whether YouTube engagement (comments, etc.) feeds back into what Gemini/Perplexity surface. Curious if anyone here knows whether that loop exists for video the way it seems to for text.

Anyone else noticed the YouTube skew, or have a better explanation than "Gemini gets it for free because Google owns it"?