2026-05-19
The story of how 8-minute single episodes killed my YouTube channel, and long-form formats
Road to 2112 devlog. Today's topic is how I was pushed around by Gemini.
8-minute single episodes, not being distributed
At the beginning of May, I switched to an 8-minute single-episode format. The reason was clear: in the era of 58-minute 4-episode packs, the AVD (Average View Duration) was only 5% of the total. When I cut it down to 8 minutes, the AVD improved to 49%. People who watched it were actually listening to the end.
This decision was based on Gemini's deep research, which said,
'If you make it 8 minutes, even at 3 or 4 minutes, the AVD is 50%! That's the best kind of content!'
Well, once I actually changed it, the impressions themselves weren't being distributed. New videos had 0 or 2 views. The CTR dropped to 0.9%.
On the other hand, only the playlist (Bedtime Audiobook No.10) had 115 views with 47% coming from browsing. The bedtime title videos from the 58-minute era also recovered to a 2% CTR.
The data is saying that '8-minute standalone videos are not chosen by the YouTube algorithm.' It's not a problem with the quality of the content, but that 8 minutes as a distribution unit doesn't fit YouTube.
Is 8 minutes bad or not?
YouTube is a platform that wants to maximize viewer session time. Videos that keep viewers for a long time are rated higher. With 8-minute audio content where the video is a looped animation, there is no evidence for the algorithm to judge that 'this video can keep viewers for a long time.'
Viewers using it for sleep are especially like this; they don't choose videos that 'end in 8 minutes.' There is a fear that it will cut off in the middle. Even if you play them continuously in a playlist, there is a risk of buffering or ads appearing when the video switches.
So, how about as a podcast? On Apple Podcasts, all episodes have views spread evenly, and it has entered the top 74 in Fiction Japan.
I'm just distributing via RSS from Spotify, and without doing anything, the view count is steadily accumulating even now. 8 minutes works as a podcast.
It's just that the optimal format differs depending on the platform, but it's not working on the YouTube side.
So, here is
Gemini: 'If you make it 8 minutes, even at 3 or 4 minutes, the AVD is 50%! That's the best kind of content!'
Hey, Gemini. Your thinking is shallow. Are you really an AI from Google (the YouTube side)?
That is the conclusion.
Now, here are the finer details regarding that.
The fact that auto-dubbing has quietly become 120 minutes.
A discovery that is quietly the most painful.
The premise for shortening from 58 minutes in April was that "the limit for auto-dubbing (YouTube's automatic multilingual dubbing) is 60 minutes." I wanted to use this feature, which can automatically expand to 20 languages, so I set it to 60 minutes.
So, when I checked the official YouTube help again, the limit for auto-dubbing had become 120 minutes. Typical.
I was doing 120 minutes in the beginning, so it's quietly painful. But oh well, I'll just leave it at that.
The conclusion of a 3-channel system
Now, since the 8-minute figure isn't changing, I will recalculate from the data and restructure. The operation from June onwards is as follows. Since it has been changing constantly over the last three months and not going well, I want to solidify this and spend the summer with it.
Main Channel: Road to 2112 — Bedtime Tech Story
Going all-in on "sleep-aid project videos." Dedicated to theme-based collection videos within 120 minutes. I will freely compose them within the 120-minute limit, such as 7 episodes + 1 hour of ambient, 14 episodes of full reading, or 4 episodes + half ambient. I will also ensure value as video, such as changing loop animations for each episode. Auto-dubbing in 20 languages. Weekly uploads.
Until now, I have been uploading in the order of Kindle episode numbers, so the themes were scattered, but from June, I will group them by theme. "Stories of families and robots," "The future of work." The idea is to group them by theme and match the interests of the viewing audience through titles and description fields. The goal is to increase CTR with thumbnails and titles.
Podcast Channel: Road to 2112 — Real News, Future Story
Here, I will split the 8-minute videos I have been doing since May into a sub-channel. Furthermore, I will automatically distribute them to YouTube from the Spotify for Creators RSS. 8-minute single episodes. The change from the current state is that the Spotify cover art will become a still video as is, and will be the thumbnail. No BGM either.
Just by uploading an mp3 and cover art to Spotify, it will be automatically distributed to YouTube/Apple/Amazon/Pocket Casts. There is no need to open Premiere. Loop animations are also unnecessary. It is pure audio distribution, making production man-hours almost zero.
When I tested RSS ingestion on my own channel, the vertical poster from Spotify was displayed as a thumbnail, and it was properly recognized as a video. Since the poster art with lettering becomes the thumbnail as is, it looks good.
Vlog Channel: Jun Sasaki | Sassa Mahha
This is the same as before. A Japanese production process vlog. I was originally doing casual Shorts, so I started making casual Longs as well. That said, since it's casual, if I shoot with a smartphone, it's a Short, and if I shoot with a PC, it's a Long. To the English-speaking world via auto-dubbing.
Reason for not mixing them on the same channel
When I used GPT as a sounding board for the first time in a while, it told me about the "algorithm reclassification cost due to format changes." Because I changed the structure every month from March to April to May, YouTube cannot determine "what" the channel is.
Mixing 120-minute collection videos and 8-minute static image podcasts on the same channel causes videos with high and low AVD to coexist, sending mixed signals to the algorithm. That is why I am splitting the channels.
Since the current 8-minute videos aren't getting impressions, rather than waiting another two months for the algorithm to relearn, it is a better decision to split them into "Sleep Time" and "Gap Time" channels starting in June. Besides, since I was creating videos for YouTube after uploading to Spotify for the Gap Time side anyway, just reducing that double work is a plus.
The main channel will be exclusively for long-form content to unify signals. The podcast will be independent as audio content on a separate channel.
All existing material will be utilized.
The most saving grace is that all the 8-minute standalone episodes I've created over the past three months won't go to waste.
They will be distributed as-is on the podcast channel and will also serve as material for the collection videos on the main channel. They will be reborn as new content simply by being regrouped under different themes.
The same goes for the 100 episodes from volumes 6-9. They can be used as standalone episodes for the podcast and as material for the collections.
Gemini, your thinking is shallow.
Over these past three months, there have been several times where I've been pushed around by Gemini's suggestions.
"You can change the voice and expand horizontally to other channels as much as you want" -> Actually, there was a risk of violating YouTube's reused content policy.
"Fully automated distribution via RSS is fine" -> When I pressed for details, it admitted, "Actually, there are risks."
"The 'bedtime' title is causing you to self-destruct, remove it immediately" -> The 'bedtime' title in the playlist was the only lifeline that was actually working.
Usually, when I corner it in those situations, it says, "I am sorry for my shortcomings." There are serious and stubborn people like that among humans too, so I just spend my time thinking that's how it is.
This is true for GPT and Claude as well, so it's a matter of what you base your core on. Ultimately, the storyteller's core hasn't wavered, and since this is just about the means of distribution, I guess it's fine.
However, since Rt2112 is linked to a mid-to-long-term plan for a follow-up business and I want to scale it as quickly as possible, the stagnation over the past three months is painfully felt, even in terms of funding.
I'll write it off as a learning cost.
Next steps
By the end of May:
Publish the 7 private 58-minute videos (I had created the 58-minute video for Kindle Vol. 10, but had made it private to turn it into an 8-minute video)
Distribute on the podcast-exclusive YouTube channel
Publish Kindle Vol. 11 in 2 types across 7 languages
June:
Write Vol. 12
Distribute YouTube project videos
Add audio for volumes 1-9 to the podcast as needed (connect to YouTube as a separate archive channel)