Did the algorithm change, or did something else?

What people mean when they say it changed

The phrase gets used for several different events, and they have almost nothing in common except the feeling that arrives with them.

Sometimes it means a video underperformed. Sometimes it means a channel has been flat for two months. Sometimes it means a whole category of creators saw the same thing in the same week. The first is normal variance, the second is usually a slow drift in the work, and only the third is the kind of event the phrase is supposed to describe. Treating all three as the same thing is what makes the anxiety so expensive: the response to one of them is to wait, and the response to another is to change something specific.

It helps to be precise about what the system is. YouTube describes its recommendations as driven by what each viewer clicks and keeps watching, with more of the same offered to that person. Nothing in that description contains a judgement about you. So when your views fall, the sentence that describes it is about viewers: the people it had been showing your videos to stopped choosing them, or stopped staying.

That sounds like a distinction without a difference and it is the most useful one in this post, because the second sentence points at something you can look at.

The check that settles it in ten minutes

Open YouTube Studio, go to Analytics, and pick a window that spans the drop with a few good weeks before it. You are looking for where the change entered, and there are only a few doors.

Now do the same thing for your older videos, the ones already published before the drop. This is the step almost nobody takes and it is the one that answers the question.

If your back catalogue kept its usual traffic and only the new uploads underperformed, the platform did not change. Your work did, in some way worth finding. If the whole channel fell at once, including videos that had been quietly earning views for a year, something outside the videos moved, and that is the case where the word algorithm is doing honest work.

Four things that get blamed on the algorithm

Each of these produces exactly the pattern people describe, and none of them is a platform change.

A small drift in the work. Very few creators change format deliberately. They change it by degrees: slightly longer intros, topics that interest them more than they interest the audience, a thumbnail style that evolved over four months. Each step felt like nothing. The audience who subscribed for the earlier version experiences the sum of them.

The peak you are comparing against. One video did unusually well, the channel's average moved up around it, and everything since has been measured against a number that was never the baseline. Work out your median video rather than your best one, and a lot of apparent decline turns out to be a return to normal.

Season and calendar. Audiences move. Exam periods, holidays, summer, a major event that occupies everyone's attention for a fortnight. If your channel has a year of history, look at the same weeks last year before concluding anything.

The end of a spike. A video that was being recommended heavily often tails off as the viewers it suits have seen it. The traffic it contributed to your channel total goes with it, and the channel can feel as if it fell even when every other video is behaving normally.

What a real change actually looks like

Platform changes do happen, and they leave a recognisable shape.

They tend to arrive on a date, not over a month. They hit a whole surface, so the effect shows up in one traffic source across many channels at once. Above all they are visible in other people's analytics, which is why the useful question to ask a peer is never whether they are seeing a drop. It is which traffic source, and from which date.

Be careful with the evidence you collect, because this is where creators talk each other into things. A thread full of people agreeing that views are down is not a measurement. The people whose views went up are not posting in it, and the people whose views are normal have no reason to be there at all. If you want to know whether something changed across your niche, you need dates and traffic sources from a handful of channels, and you need the ones who are doing fine to answer too.

Why the response is the same either way

Suppose you do the check and it was a platform change. There is nothing to appeal to and no setting to adjust, and understanding the mechanism in detail would not recover a single view. What you can do is find which of your videos still works under the new conditions and make more of those. Suppose instead it was a drift in your own work. Then you find which of your videos still works and make more of those.

The diagnosis changes what you tell yourself. It does not change the next move. That is worth knowing in advance, because the search for an explanation can eat weeks that the next three videos would have answered faster.

What the diagnosis does change is what you should not do. If the fall is platform-wide, resist rebuilding a channel that was working, because you would be responding to conditions rather than to a flaw, and you may end up changing the thing that was carrying you. If the fall is your own drift, resist waiting it out, because it will not recover on its own.

The habits that make this question rarer

Much of the worry comes from judging single videos. One video tells you little, because variance between videos is large and normal. Look at the last ten instead, and the noise flattens into something you can read.

Keep a short written record of what you changed and when. Thumbnail style, length, intro structure, the day you post. When views move three weeks later, a record turns guesswork into a list of candidates, and it is the cheapest analytics you will ever build.

Decide how long a test runs before you start it. A change judged on one video is judged on noise, and creators who change something after every disappointment never hold anything still long enough to learn from it. The channel ends up with a format that has been revised twenty times and tested none.

What blaming the algorithm costs

Blaming the algorithm moves the explanation outside you, and an explanation outside you is one you cannot act on. Creators who settle on it can stop testing, because testing only makes sense if the outcome depends on what you do.

A drop that hits only your new uploads has a cause inside the work. That is the unpleasant version to hear, and the one you can do something about this week.

So run the check. Ten minutes of your own analytics will tell you more than a week of reading what happened to strangers.

Common questions