The YouTube Video Shaping AI’s Answer About Your Product Has 565 Views

I went looking for the sources Perplexity trusts most when people research software. One of the most-cited ones I found was a YouTube video with about 560 views.

Let me back up.

I ran 75 searches about AI meeting note tools, changed the source instructions each time, and logged every citation that came back — 926 in total. When I sorted them, YouTube was one of the biggest layers: about 14% of all citations, and roughly a third of a plain, no-instruction answer. On questions about a specific product it was bigger still, around 57% of the sources for "Fathom AI review."

That alone surprised me. I expected official pages and review sites. I did not expect video to be carrying a third of the answer.

But the real surprise was where inside YouTube the weight sat.

Key findings#

• YouTube was 131 of the 926 sources Perplexity cited (~14%), and roughly a third of a default, no-instruction answer — around 57% on a branded query like "Fathom AI review."
• All 131 YouTube citations came from just 12 videos. Two of them were cited 30 times each — together, nearly half of every YouTube citation in the test.
• Those two most-cited videos had only ~560 and ~607 views when I checked. About 1,200 views between them.
• Both were affiliate comparison videos using the same template: the same affiliate link, the same word-for-word disclaimer, the same throwaway business email.
• View count and AI influence are unrelated. The clips almost no human watched were the ones the model treated as a fair account of how the products compare.

How much of an AI answer comes from YouTube?#

More than most people expect. Sorted by source type, YouTube was about 14% of all 926 citations and roughly a third of a default answer. And it climbed the closer a question got to a buying decision:

So when someone asks the AI about your product by name, a third to a half of what it says back can come from video. That is exactly the moment you want to be represented well.

Twelve videos — then really, two#

Those 131 citations did not come from 131 videos. They came from 12. That is concentrated enough on its own. But it gets narrower.

Two of those twelve were cited 30 times each. Between them, those two clips accounted for almost half of every YouTube citation in the whole test. So I went and watched them.

The first carries an affiliate link and notes, in the fine print, that the creator may earn a commission. The second uses the same template: the same affiliate link to the same product, the same word-for-word disclaimer, the same kind of throwaway business email in the description.

Two videos. About 1,200 views between them — which is to say almost no one has watched either. And they shaped close to half of what the AI had to say on video about these tools.

Why does AI cite YouTube videos almost no one has watched?#

This is the part I cannot stop thinking about.

By public attention metrics, these videos barely registered. About 560 views is a rounding error, and by any normal measure they are not the standout review on the topic. But inside Perplexity's citation layer, they mattered a lot — cited 30 times each, treated as a fair account of how two products compare.

The two rankings point in opposite directions. The thing people ignored is the thing the model leaned on. AI visibility is not the same as human popularity: a 560-view clip can outweigh a polished video with a hundred thousand views, simply because the model picked it up and kept citing it.

Were the most-cited videos even neutral?#

It would be one thing if these were careful independent reviews that just happened to be unpopular. They were not.

Both are affiliate videos. The format exists to route you toward a product the creator earns a commission on. The matching template across the two suggests a playbook — the video version of the cheap comparison pages that flood search results. So the model did not just trust tiny videos. It trusted tiny videos built to sell, and presented their verdict as if it were an even-handed comparison.

If that sounds familiar, it is the same pattern I kept finding in the text sources: the layer that looks neutral usually is not. Video just makes it easier to see, because the view count is right there.

What low-view AI citations mean if you run a brand#

Most of us think about video, when we think about it at all, in terms of reach: land a popular creator, rack up views, win the audience.

For AI visibility, reach is not the game. The question you want answered is not "which videos are popular," it is "which videos does the AI actually pull from when someone asks about my category." Those can be two completely different lists.

That is uncomfortable, because it is not a list you can buy your way onto with an influencer budget. But it is a list you can study:

• Find the handful of videos the model leans on for your category.
• Watch them. Are they accurate about you?
• Are they made by someone with a reason to push a competitor?
• Is there a better video for the model to find instead?

You do not get to pick which videos AI uses. But you can make sure the ones it is likely to reach for actually represent you, instead of leaving the most-cited layer of your category to whoever happened to film a cheap affiliate clip first.

A few honest limits#

This is one engine (Perplexity), one category (AI meeting note tools), and 75 searches over a short window. The two videos I singled out are the two most-cited in this test, not proof that AI everywhere prefers tiny affiliate clips. The view counts are what I saw when I checked, and they will drift over time.

But the shape of it is hard to unsee. The most-cited video layer in my test was not the most-watched, the most-authoritative, or the most-neutral. It was small, monetized, and almost invisible to humans — and the AI treated it as a fair witness. You can trace every YouTube citation, and the search it came from, in the full 926-citation dataset.

FAQ#

Does AI cite YouTube videos?#

Yes, and more than most people expect. In this 75-search test, YouTube videos were about 14% of every source Perplexity cited and roughly a third of a default, no-instruction answer — climbing to around 57% of the sources on a branded query like "Fathom AI review."

Why does AI cite low-view YouTube videos?#

Because the model does not weight by view count. It reaches for videos that topically match the question — a product review or an "X vs Y" comparison — regardless of how many people watched them. In this test the two most-cited videos had only about 560 and 607 views, yet each was cited 30 times.

Are the YouTube videos AI cites neutral?#

Often no. The two most-cited videos here were affiliate comparisons built to route viewers toward a product the creator earns a commission on, using a near-identical template. AI presented their verdict as if it were an even-handed comparison, which is the same "looks neutral, isn't" pattern that shows up in text sources too.

How do I find which YouTube videos AI cites about my brand?#

Stop optimizing for reach and start auditing influence: ask the AI engines questions about your category and product, note which videos they cite, then watch that handful. Check whether each is accurate about you, whether it is made by someone with a reason to push a competitor, and whether a better video exists for the model to find instead.