A friend and I ran a slightly silly experiment two years ago. We took the same subject, shot separately but structured identically, with near-identical titles and comparable thumbnails, and published on our two channels within a day of each other.
Her channel had roughly a third of my subscribers at the time. Her version got about four times the views.
The reasons took a while to work out and they have shaped how I think about channels since.
Channel context is a real signal
Her channel was consistently about one subject. Mine covered several loosely related things.
The video sat naturally within her catalogue. It sat oddly within mine. Whatever mechanism decides who to show a new upload to appears to draw on what else the channel has done and how those performed with similar audiences.
Which means a video is not evaluated alone. It is evaluated as the next thing from a source with a history, and the history is doing work you cannot see.
This is not a mysterious algorithmic conspiracy. It is roughly what a human would do. If a channel has reliably produced good material on a subject, the next thing from it is a better bet than the same thing from somewhere unrelated.
The audience she had was the audience for the video
The subscriber count comparison was misleading because our subscribers were different populations.
Hers had subscribed because of that subject and were reliably interested in more of it. A large share of them watched within the first day, which appears to matter for whether something gets shown more widely.
A good proportion of mine had subscribed for something else entirely, several years earlier, and were indifferent. My early view rate was poor and the video never got pushed beyond that.
The lesson I took is that subscriber count is close to meaningless as a number and the composition is everything. Ten thousand people who want what you make next is worth more than a hundred thousand who wanted something you made in 2022.
The comments diverged immediately
Under hers: questions, additions, disagreements about specifics. Under mine: a lot of people saying they had not seen anything from me in a while.
That second category is worth noticing. It means a substantial part of my audience did not have a current relationship with the channel and was being reminded of an old one, which is a different and much weaker engagement.
What this changed about how I work
I stopped treating the channel as a container for whatever I make and started treating it as a claim about what I make.
Practically: I narrowed. Not to a single topic but to a describable one. Anything that would not sit comfortably next to the last five uploads either does not get made or goes somewhere else.
I have a second channel now for the things that do not fit, which grows slowly and is fine, and which stops the main one becoming incoherent.
The uncomfortable implication for older channels
If your catalogue is a mixture accumulated over years, you are carrying that mixture into every new upload.
There is no clean fix. Deleting old work is drastic and removes views you are still receiving. Unlisting things is less drastic and has a similar effect on how the channel reads.
What seems to work, slowly, is consistency going forward. After about eighteen months of narrower output my performance improved noticeably on the same kinds of video that had underperformed before. Whether that is the catalogue rebalancing or the audience shifting I cannot tell, and the effect was real.
The thing that was not the explanation
Worth ruling out what we checked and dismissed.
It was not the thumbnails. We swapped them a month later on both videos and nothing much happened.
It was not publishing time. We were a day apart and her weaker uploads at the same time of day did not benefit.
It was not the edit quality. Mine was, if anything, more carefully made, which was a slightly bruising thing to establish.
It was who the video reached in the first few hours, and that was determined before either of us pressed publish.
What I would tell someone starting now
Decide what the channel is about early and let that constrain you, even when it feels restrictive.
The freedom to make anything sounds valuable and mostly manifests as a channel that nothing recommends because nothing knows what it is.
And ignore subscriber count as a measure of anything. Look at how many people watch a new upload in the first day as a proportion of subscribers. That number tells you whether you have an audience or an accumulation.
The experiment we should have run
One flaw worth admitting. We tested this once, on one subject, which is not evidence of anything on its own.
What would have been better is swapping — her making something from my subject area and me making something from hers, published the same way.
My guess is that both would have underperformed, which would have separated the channel-fit explanation from the simpler possibility that her subject was more popular than mine.
We never got round to it, and I have watched several people describe the same pattern since without anyone testing it properly either.