Likes from logged in views do count, the like to view ratio is higher due to that.
A German YouTuber tested it with an unlisted video, I can’t find it as of now but I do find a lot of deleted videos in my liked ones…
Basic theory as of the time of the video is that they want to go against bot views. Short views did not decline, only long form video, and of those the people with mostly desktop views are hit the most.
Edit: found it!
I find it crazy that the client would call an API endpoint to register a view. Then we could just call that end point in a script to arbitrarily inflate the view count, can we not? How do they prevent this exploit? Wouldn’t it be better if the server registered the view after a certain amount of the video was delivered?
Why is that crazy?
You don’t want to register a view just because the backend sent a 200 response for the video.
You want to register a view when the video has loaded, it has begun to play, and quite likely after some arbitrary percentage of the video.
This is all front end information. How much of the video was made available to the front end to play isn’t really relevant.
Alright, not “crazy”, but this doesn’t explain how they combat exploiting this by just calling the end point.
That’s more or less how it works, but that’s still an additional call. If Google is not tying it directly into segment download requests, then it could potentially be blocked without disrupting playback.
I have no insight into the details of the inner workings. If I download a video with yt-dlp, does it increase the view count? If not, then it’s a broken system, yeah?
I’m very curious to know how they register a view, yeah. The only true measurement is the video reaching someone’s retina or ears, to measure attention, which is impossible at a large scale as of yet. 🥲
They’ve been using AI to detect things like that for a long time. (AI in general not LLMs as far as I know.