Production update 2022-12-30 (Patreon)

Hello everyone!

I hope you're having great holidays. I spent my week visiting family, eating too much, playing board games and overall enjoying some time off. Unfortunately, right now I have a pretty mean cold and I've been feeling like trash for 48 hours.

Thumbnails and CTR

When the supercut was released, it initially did quite poorly. I think the reason is twofold: first, the thumbnail I gave it was a bit uninspired, and second, more importantly, it was initially shown to my most fervent viewers, i.e. everyone who had already watched the series in full, which ironically was the worst group of people to show it to.

After a couple of days, I had a sudden idea for a better thumbnail. I made it in the morning and showed my girlfriend, and she liked it better, but thought there was room for improvement. In hindsight, she was absolutely right.

I implemented some of her suggestions, I came up with my own improvements, and I ended up with a thumbnail that I am overall much happier with.

Here is the CTR graph (Click-through rate, i.e. the % of people who saw your thumbnail that actually clicked on it). Marked with a red arrow is when I uploaded the new thumbnail. Ignore the sharp dip at the end, that's just incomplete wonky data.

"But Bismuth! That's terrible! CTR went way down after you uploaded the new thumbnail!" You would think that, but the reality is more complex. When you upload a new video, the initial CTR is typically between 5% and 10%, sometimes even higher for a few hours. It takes a pretty big drop on the second day, and another big drop on the third. From there, it largely depends on if it's getting recommended a lot. Lots of recommendations = lower CTR on average. For comparison, here are the first 14 days of another video of mine, the Zelda 1 video.

Notice just how much higher the peak is. Sub-4% CTR on the first day is straight up terrible. But then, if you look after the first two days, the Zelda 1 video had a CTR under 2.5%, even under 2% once it got past the initial release bump and it started getting recommendations to all sorts of people. Meanwhile, the ABC History supercut barely dipped and is still doing pretty well despite increased visibility, hovering around 2.5%, although it is getting slightly less impressions per day compared to Zelda 1 in the same period. All this to say, I think the thumbnail change was a huge benefit for the video.

This led to a discussion in my Discord server about clickbait. There's a very good Veritasium video about it that I think is valuable to watch even if you don't post content online. https://www.youtube.com/watch?v=S2xHZPH5Sng

I basically agree with everything he says. The part about Type I vs. Type II clickbait is especially relevant. What I want to make is Type I clickbait, or titles and thumbnails that get you to click, but you watch the video and the question asked is answered, or whatever curiosity you had is fulfilled. Titles like "How is this speedrun possible?" drive a lot of people to click on the video, but the question is literally the entire point of the video: explaining how the speedrun was made possible. 

This is what I attempted with this thumbnail: it's very YouTubey with the Mario face, but it communicates the point of the video very clearly and succinctly.

In short, my strategy is simple: I don't enjoy the thumbnail clickbait game and I don't like to play it. But it's an undeniable aspect of getting your channel out there and I will do it to a certain extent. What I will never do is go for sensationalized thumbnails and titles that have no relevance with the content. They might be phrased a little sneaky, but they will never be misleading. Having great content is nice to keep your viewer base interested, but to get new people into your channel, you need something that piques their interest. If they don't click, they don't watch, they don't know what they're missing.

With that said, this is what happened after I changed the thumbnail:

You can clearly see the line go from below typical to above typical. The effect was instant and very noticeable. It really works!

The next project: Minecraft any% set seed TAS

I also recently started the initial work on my next project: breaking down the Minecraft any% TAS in 20 seconds. I messaged the author, DylanDC14, and he is on board. He showed me a 0.1x speed version that I studied and that I plan on using for the breakdown. I came up with a rough structure and a set of questions for things I didn't understand right away, and he gave me pretty detailed answers to nearly all of them the same day. I'm pretty excited for this project!

It's gonna be a departure from the ABC stuff. I think in terms of how complex the concepts are gonna be, this might be one of my simplest videos yet. I plan on making it super accessible for anyone that's at least remotely familiar with Minecraft, even if all you know is that it's a sandbox 3D blocks game. I'm gonna explain the goal, the route, and everything that needs to be done and why. The TAS is 20 seconds so I can definitely afford to spend extra time on adding more context. I don't want to pad for length so the video might end up only 15 to 20 minutes long, but I hope it will be a good introduction to Minecraft speedruns and to TASes for casual viewers.

I will get on the actual script as soon as January 2nd or 3rd, and I hope to complete it within the week and start recording on Friday. We'll see how that works out.

Thank you for your support, and I wish you all a happy New Year!