2026-04-14

After adjusting the target, the numbers dropped as expected, but I've come to terms with it. Rt2112 Development Log | April 7 - April 14, 2026

I am taking on the challenge of creating audio stories for a global audience on YouTube.
This is part of that process.

👇 My channel is here

👇 My BGM channel, just in case

While repeating trial and error, I applied the persona adjustment from the other day, and although the temporary numbers have started to drop, it's only the first month, so I'm taking it in stride. Here is that story. (It might overlap with the previous one.)

There has been a change in the algorithm

• My latest video hit a CTR of 4.1% and an AVD of 6:33. An all-time high.

• Impressions are low, but it is starting to be delivered pinpoint-style to the target audience.

• Transitioning from 'spray and pray' to 'sniper.' The algorithm is learning the cluster.

• Even if the view count is in the single digits, the retention rate and CTR are glowing green. This is the approach form before an explosion.

👉 This is also the reason why impressions dropped. Up until now, it was just browsing randomly (the stuff that gets scattered in the YouTube recommendation feed), which resulted in a low CTR (click-through rate) and low AVR (average view duration).

Conversely, when dealing with trending topics, if you create content with AVD in mind from the start, impressions will increase rapidly, but I'm not looking for that this time; I'm fine with a slow and steady increase.

The preamble aside, I believe the reason I got the best numbers ever is simply because 'it provided the video to the core audience first.' As a result, the next round of impression scattering didn't hit the mark, leading to an immediate drop. If this continues for a long time, it becomes a mechanism that rises in a staircase pattern.

Now, the previous entry was about verifying why it doesn't last, and this time I'll talk about how I further refined and organized the finer details.

The hypothesis was supported by country-specific data

• US: AVD 6:33 (Expressive Speech. It's hitting the mark properly)

• Japan: AVD 2:13 (Standard TTS. Immediate drop-off)

• Ukraine: AVD 1:07 (Standard TTS. Immediate drop-off)

• South Korea: AVD 0:46 (Standard TTS. Immediate drop-off)

• The standard TTS regions are dragging down the overall AVD of 2:47. No countermeasures available. Waiting for an increase in traffic from ES/PT/DE/FR regions

👉 This is information from last time. The point is that TTS audio is polluting the AVD. This time, I removed TTS languages from all videos. Including Japanese, of course. While I wish they would update it sooner, the major languages I'm creating on the Kindle side overlap with the Expressive Speech regions, so I'm prioritizing the policy of steadily improving quality there first. Also, I had originally intended to create Hindi and Indonesian on Kindle, but I stopped because Kindle didn't support them. I'd like to do them once Kindle supports them, and since I don't think there's much of a reading culture there, I hope they flow into podcasts instead.

I'm happy that I got one comment from Ukraine, but since 1 minute is tough, I excluded it. I expect that Arabic and others will likely become Expressive Speech in the next update.

Main Channel Data (As of 4/9, 28-day period)

• Views 1,714 / Impressions 34.1K / CTR 2.4% / AVD 2:47

• Subscribers 15 (+10) / Watch time 62.3h / Monthly audience 774

• The top 2 weekly videos account for 64% with 80 and 66 views respectively. The selection process has begun

Lo-fi 2112 Data (As of 4/8, 28-day period)

• Views 215 / Impressions 6,343 / CTR 2.0% / AVD 16:02

• Subscribers 4 / Watch time 72.5h (more than the main channel)

• The right end of the graph is surging. Reached 30 views/day recently

• The AVD for the 206 views excluding Japan is calculated to be over 17 minutes. Proof that it is being played as BGM

• The challenge is CTR. The content (AVD) is acceptable. It's just a problem with the entrance (thumbnail)

👉 The challenges are clear. For the main side, it's AVD; for the sub, it's CTR. However, regarding Lo-fi, it stays at 2.0% no matter what I do, and I think that's just how it is relative to the population. I did a lot of BGM research and crawling, so similar-sized successful BGM channels appear in recommendations, but since what I'm doing is no different and the format is similar, I think a chance will come if I keep going like this.

TuneCore distribution started

• First album "Lo-fi2112 / Distant Workshop, Craftsman Dusk" distribution completed across all services

• Artist name: Sassa Mahha

• Content ID off only, all others checked

• Decided on the policy of pinning one album-specific Linkcore URL in the fixed comment

• No need to direct traffic from the main channel to Lo-fi via pinned comments (risk of cluster contamination)

👉 This is simply a project I'm doing because I want to.

Finalizing the DJ character composition

• Switching Lo-fi 2112 thumbnails to a front-facing DJ booth composition

• Fixed: Front-facing DJ booth composition, Lo-fi 2112 logo position

• Variable: Player (cat robot, dog robot, owl, rabbit...), background (Distant theme)

• Differentiating by using horizontal composition for the main channel (story-driven) and front-facing DJ for Lo-fi (music-driven)

• The Lo-fi side will follow a "go all out and be silly" policy. A contrast to the dignity of the main channel

• First DJ character: Scarf cat robot. An ennui atmosphere

👉 Since it's a sub-channel, I thought I'd have some fun with it. I sometimes listen to BGM channels that feature DJs, and with that vibe in mind, I thought, 'Let's have a robot animal be the DJ.' I tried it, and as expected, the CTR dropped to 1.0% or so—I really messed up.
(Claude and Gemini both said, 'It'll definitely go up!', so they're irresponsible.)
However, it might just be a coincidence, but I got a subscriber from a Spanish-speaking country who even left a comment. There is some reaction, so then
The last two videos mix scenes and DJing. (The idea is a bit lazy)

Still, the reaction is good, and the AVD is consistently high.
Ultimately, this is the optimal solution.

BGM start-up issue

• Starlight Dome Nights has an AVD of 0:28. Immediate drop-off on a 56-minute video

• Cause: The BGM starts too slowly. Ambient-style fade-in

• Countermeasure: From now on, place the catchiest part of the melody at the beginning. Make it an immediate-impact type

Maybe I touched on this last time? Regardless, if background music for work doesn't have a steady beat from the start, it's a no-go.
It would be different if it were specifically for sleep, but it depends on the genre.
My content is Lo-fi centered on near-future stories, so it's not meant to put people to sleep, but if they do fall asleep, that's fine too—it's just relaxing Lo-fi.

Changes to Tag Strategy and Title/Description

• 3 fixed tags: #CozySciFi #GentleFuture #FutureSociety

• 2 variable tags: Selected from 16 categories based on the news source

• Tags section completely revamped for the tech-savvy audience. Sleep and healing-related tags removed.

• Changed the beginning of the title to "🎧📖 4 Bedtime Audio Stories…"

• Description updated to match the title.

👉 As I mentioned last time, I think the fact that the algorithm is currently in a state of settling down is one of the reasons for the drop in impressions.
The title and description are minor details, but I am thoroughly presenting them as audiobooks.
I added 🎧📖 to the beginning. AI advice suggests putting [Audio] at the start for SEO, but I prioritized visibility with emojis.
I'm using the word "Audio" in the title text instead as a countermeasure.
There was a common phrase "from gentle future" after "Stories," but I've moved that to the description. Since it's covered by tags, I've trimmed it from the title to fit as much of the story title as possible, up to the 100-character limit. The intention is to broaden the search reach.

For example, if "solarpanel" is in the title, I'll consciously include it.
It's all about tech, after all.

Podcast Maintenance

The numbers were dropping quite significantly, so I fed the current data into Gemini's Deep Research to get some countermeasures.
As a result, the most important thing to do is the podcast.
It might seem odd that I wasn't doing it despite it being an audiobook, but since I was just porting the YouTube playlist as-is, I started by organizing the playlist.
The playlist is created to match the Kindle books, so the intention is naturally to drive traffic there as well.

Listening with ears or reading with eyes

I also added minor things like prompts for the Hipe feature into the description—I've done what I can. I don't know if it will be effective.

Refining Content Creation

This is also the main content of this note update.

I mentioned last time (I think) that I'm shifting from a Kindle-first to a YouTube-first story creation process, combining 4 episodes x 7 to publish 28 episodes on Kindle, but
I've also decided to generate the 4-episode batches within the same category.
Same category = the 16 types in the tag strategy.

I'm using Claude to create a dedicated story team for brainstorming story plots.
I apply one of the 16 categories to it to have it fetch news.
Previously, I was mixing them up, like episodes 1-4 being education-themed and 5-8 being space-themed. I don't think it changes anything for the reader experience.
The reason is ease of management and ease of expansion.
Since I manage tags for each episode in JSON, I can bind books by tag. (Like the collected works of classics you see on Kindle.)

For example,
if I release a collection like
Road to 2112: Space Edition,
I can use a single Python command to convert episodes tagged "space" in JSON from epub to kpf, and the Kindle book is finished.

Above all, since I make videos in 4-episode units,
YouTube also ends up with consolidated videos like Space, EdTech, etc.
As a result, I end up creating playlists by genre.

The audience can narrow down to fields they are interested in, which improves the viewing experience.
With that in mind, I've also changed the very foundation of how stories are generated.

Next steps

Either way, the accumulation remains the same.
The main focus is to thoroughly build up the shift in strategy. I will clear out the previous stock within April. The goal of this is algorithm correction and maintaining input.
The videos currently being prepared will go through an auto-dubbing period, and videos based on the current policy will start airing from May, so I will begin the accumulation of the 30th to 50th videos from there.
The 50th video will be the first point of review. The target date is the end of June. If there are no signs of progress at all, I will look for a compromise.

The sub-project is going perfectly. I enjoy music production, so I will continue with it, but on the other hand, since Lofi2112 is an IP based on the main story, there will also be a review of its relationship with the main project.

That's about it.
For now, I will keep logging the progress as I go, and from here until the end of June, the tasks will remain unchanged. It will be a series of monotonous and uneventful logs.