Word count gets you cited. View count does not.

The largest citation dataset says popularity does nothing.

Otterly tracked more than 100 million AI citation instances across a 30 day window and correlated each cited video's metadata against how often it was cited (Otterly).

Over 40 percent of cited videos have under 1,000 views.

And 35 percent of cited channels have fewer than 10,000 subscribers. Long-form video accounts for about 94 percent of YouTube citations, Shorts for 5.7 percent, and playlists, channel pages and livestreams combined for 0.3 percent.

This is the best news a small or specialized channel has had in a decade. It is also the whole argument for retrofitting a library rather than making more content. The footage that would get cited is usually already shot.

Read the caveat before you build a strategy on it.

The sample is videos that were already cited. So these numbers explain what earns repeat citation among videos already in the pool. They do not prove what gets a video into the pool in the first place.

That is a real limitation, and anyone quoting r = 0.31 at you without mentioning it has not read the study. We publish it here because a number you can trust is worth more than a number that sounds better.

Score your Video for AI Findability →

An engine is not running a popularity contest.

It is looking for the most legible, most specific, most attributable answer to the question in front of it. A video with 400 views and a 300 word description naming the exact product, the exact error and the exact fix is a better source than a video with 400,000 views and a campaign title.

Popularity is a signal about humans. Citation is a decision about text.

Chapters turn one video into several citable objects.

In the same dataset, 31 percent of cited videos carried timestamps, and 78 percent of those were cited repeatedly across two to five different chapters. A chapter is effectively a separate addressable object with its own topic.

Get the specification right or none of it happens. YouTube requires at least three timestamps in ascending order, the first starting at 00:00, and a minimum of 10 seconds per chapter (YouTube Help). Break any one of those and chapters silently do not render, which is the usual reason a team concludes chapters do not work.

Fact density is the lever that generalizes.

Princeton's GEO research tested content modifications against generative engine responses and found visibility gains of up to about 40 percent, with adding quotations and adding statistics the strongest levers and explicit source citation somewhat lower (Aggarwal et al., Princeton).

Applied to video, that means a description that states a number, names a source and quotes the speaker. "Watch to find out" gives an engine nothing to lift.

What to do first.

1. Write the description properly on the videos that answer a real question. It is the only field with a measured positive correlation and the cheapest one to change.
2. Add spec-valid chapters to anything over about eight minutes.
3. Correct the transcript, and publish it on a page you own. Auto-captions fail exactly where it matters, on product names and technical terms.
4. Ignore view count when you triage. Sort by whether the video answers a question your buyers ask.
5. Do not chase Shorts for citation. They are 5.7 percent of citations and their description links are not clickable.