The Algorithm Is the Reaction, Not the Starting Point

By the end of this piece you’ll have five specific questions to ask when a video flops, instead of the one question every creator asks and none of us can answer: why didn’t the algorithm push it?

I spent several days at VidCon 2026 sitting in YouTube growth sessions and talking to creators between them. Three sessions in particular stuck with me: Todd Beaupré from YouTube’s Growth and Discovery team on how recommendations actually work, Houston Matter’s “Anatomy of a YouTube Video,” and Jon Youshaei’s masterclass. Jon is ex-YouTube, ex-Instagram, and now a creator himself, so he’s seen the machine from both sides.

I don’t have YouTube figured out. What made these three talks valuable was that they weren’t giving the same advice. They were describing different links in the same chain, and once I saw the chain, I couldn’t unsee it.

The chain

Here’s the core of it. By the time YouTube is deciding whether to recommend your video more widely, the viewer has already answered a series of questions:

Did the idea make sense as a video? Did the packaging set the right expectation? Did the first 30 seconds confirm the click? Did the video deliver what it promised?

The recommendation system is reacting to those answers. It’s downstream of the viewer, not upstream of you.

So “the algorithm didn’t push it” might be technically true, but it tells you almost nothing about what broke. It’s like saying “the scoreboard didn’t like us” after losing a game. Accurate, and useless.

And this matters more right now than it did two years ago. The cost of making and uploading content keeps falling, which means the pile of videos competing for the same viewers keeps growing. “Make good content” is too vague to be actionable advice anymore. The creators pulling ahead aren’t necessarily working harder. They’re diagnosing better.

Each of the three talks covered a different link in that chain. Here’s what each one taught me.

Before you film: not every good idea is a good video

Youshaei’s session opened with a distinction I hadn’t heard framed this way. He quoted a line from Johnny Harris: if there are a million good stories worth reporting, only 900 should be told as videos.

His test for whether an idea belongs on YouTube was simple. Imagine explaining it to a friend over coffee. Would you need to show them something? A chart, a before and after, a physical object, a comparison, a real example? If yes, the idea has visual anchors, and it can carry a video.

If no, it might still be a great idea. It might just be a better essay, newsletter, Reddit post, or conversation than a video. Forcing a non-visual idea into a visual medium is how you get technically fine videos that nobody finishes.

This one hurt. A lot of small-channel videos, some of mine included, are basically spoken essays with footage underneath. The idea was good. The medium was wrong. No amount of thumbnail work fixes a medium mismatch.

After the click: your retention graph is a replay, not a report card

Houston Matter’s talk changed how I read analytics. Most creators, me included, treat the retention graph like a grade. Line goes down fast, video bad. Line holds, video good.

He framed it differently: the retention graph is a replay of viewer behavior. Every dip is the exact moment a viewer changed their mind. That moment has a cause, and the cause is visible if you go watch that timestamp the way a stranger would.

Maybe the intro dragged. Maybe the title promised one thing and the opening started somewhere else. Maybe the good part came too late. Maybe the first 30 seconds explained the video instead of proving the video was worth watching.

That last failure mode is the most common one I now notice everywhere, including in my own uploads. The viewer doesn’t know the good part is coming. You know, because you made it. They only know what’s on screen right now. The first 30 seconds have one job: confirm the click. The viewer should immediately feel that this is what they came for.

After viewers react: there is no single algorithm judging you

The session with YouTube’s own Growth and Discovery team was the most myth-clearing hour of the conference. They pushed back, politely but firmly, on the mental model most creators carry around: one giant algorithm forming an opinion about your channel.

What’s actually happening is personalized recommendation, viewer by viewer. Some consequences of that:

One bad upload doesn’t permanently poison your channel. Taking a break doesn’t make YouTube punish you. Subscribers skipping some of your uploads is normal, not a death sentence.

But the flip side is just as important. The system needs evidence that viewers were satisfied. If people click and leave, that’s evidence. If they watch one video and never want another, that’s evidence too. You can’t argue with the system, because the system is just an aggregate of what real viewers did.

The bonus reframe: niche is the expectation, not the topic

One idea from the conference floor that tied everything together for me: niche isn’t your topic. It’s the expectation a viewer forms about what your next video will feel like.

“Cooking channel” is not a niche. It’s a category. Cheap meals, lazy weeknight meals, chaotic cooking, restaurant recreations, tiny food. Those are niches, because a viewer who liked one video knows exactly what the next one offers. The clearer the expectation, the easier it is for a viewer to want another one. And if the viewer’s decision is easy, YouTube’s job is easy.

The five questions

So now, when a video underperforms, these are the questions I ask instead of blaming the weather:

1. Was the idea visual enough to be a video at all?
2. Did the title and thumbnail create the right expectation?
3. Did the first 30 seconds confirm the click, or just explain the video?
4. Did the payoff deliver what the packaging promised?
5. Did the video give the viewer a reason to watch the next one?

Every one of those questions is answerable with things you already have: your own footage, your own retention graph, and the willingness to watch your video like a stranger who owes you nothing.

That’s less comforting than “the algorithm ignored me.” It’s also the only version of the problem you can actually fix.