The last post went up at the end of September, and then a long weekend of public holidays arrived and I mostly let it. I went to festivals. I still ended up building things, just from my phone, in between, through Claude Code and a lot of Codex, where I spent some time trying out OpenAI’s new Astra model. It’s a strange and slightly addictive way to update a website.

Then I came back to my desk and scheduled three Instagram videos for the week.

The numbers

The first was a talking head video, about a minute long. It got 95 views.

The second was a meme. You know the one: someone thinking “I should really start a business” while doing absolutely nothing and staring off into the distance. Thirty seconds, about twenty minutes to make, 47 views and around ten likes.

The third was one of the experimental videos I’ve been making, where I draw something, animate it with AI, then take it into an editor for captions and music. This one was about reaching 1,500 followers and the celebratory cake being cancelled due to budget cuts. It got 27.

The three videos the week before got roughly 2,000 views each. That’s a lot more than a slow week, so I went looking for why.

What I checked before deciding it was me

Most of what you find when you search “Instagram reach dropped” is people saying yes, something is broken, and other people with big accounts saying everything’s fine for them. Both can be true at once, which doesn’t help much.

So here’s what seemed worth actually checking, if you’re in the same spot.

Account Status first. It’s in your Instagram settings, and for professional accounts it tells you whether your content can be recommended to people who don’t follow you. If it says you can’t be recommended right now, that’s your answer, and it shows you a sample of the posts behind it so you can edit, remove or appeal them. If it says you’re fine, the drop is probably about the posts themselves, or about Instagram in general, not about a penalty on your account. That’s less dramatic than a shadowban, and a bit more annoying.

Then the general picture, which is messier than either side says. Metricool looked at about a million accounts across 2025 and found Reels reach down 35% year on year, with regular posts down 31%. But their Instagram-only study in June this year found Reels reach going up in early 2026, by about 12%, with single image posts taking the hit instead. So depending on which report you read, everyone’s reach is collapsing or Reels are doing fine. I think the honest version is that nobody outside Instagram knows what happened to your three videos on Tuesday.

Then the AI question, which is the one I was actually worried about. My drawing videos start as an image, and I animate that image with AI. I had a theory that Meta was flagging that and quietly turning my account down.

What Instagram has said publicly is more specific than my theory. In May it started testing an optional “AI creator” label and said ranking and recommendations wouldn’t change. Then at the end of August it renamed it “AI-generated profile” and said profiles built around an AI-generated person, without the label, would get less reach in recommendations. Creators who use AI for edits, captions or graphics apparently don’t need it.

Which leaves me somewhere in the middle. The character in my videos is a drawing of me, and the drawing is mine, but the movement isn’t. I genuinely don’t know which side of that line Instagram thinks I’m on, and I suspect their detection doesn’t know either. Account Status is supposed to tell you if it’s decided, which is another reason to check that before anything else.

The part I keep coming back to

Honestly, the numbers are only part of it. I’ve been saying for a few months that Instagram isn’t really doing much for Creative Mind Habits or any of the projects, and this week made that easier to see. I’m not deleting it. I just don’t think I’ll be planning around it anymore.

X has been the opposite. It’s become the place where I try out half-formed ideas and share the building as it happens, and people actually reply.

When I posted about hitting 1,500 followers and the cake being cut from the budget, someone commented asking where the link to the cake was. There wasn’t one. It was a joke in a drawing.

So I made one. creativemindhabits.com/cake is a page where the cake downloads to 99% and gets stuck there, “finance is reviewing the frosting”, until you give it a nudge, and then you get a little GIF of a slice landing on my empty plate. Then people asked for other flavours, so now there’s chocolate, buttercream, strawberry, and lemon pound cake, which if you know the Afroman video, you know.

None of it is polished. It’s very makeshift and scrappy, and that seems to be part of why people like it. People have been saying they really like this version of the website, which I didn’t expect.

If there’s anything useful in this week, I think it’s that the comment was a better brief than anything I’d planned. Someone told me exactly what they wanted, in public, and it took one evening to make. The meme took twenty minutes, and nobody had asked for it.

The app, and the $99 question

The bigger thing I’m building in public is FriendFriendFriends, an app for meeting people through the friends you already have. You add three to ten friends, they add theirs, and slowly there’s a group of real people who are a couple of steps away from each other. It’s deployed, I’ve tested it, and it works.

The next step is the app stores, and that’s where I stalled. Apple’s developer program is 99 USD a year (it varies a little by region). Google Play is a one-off $25, but new personal accounts have to run a closed test with at least 12 people opted in for 14 days in a row before they can publish to production.

Which is funny, for an app whose whole premise is getting a handful of real people to show up together. The test is sort of the product.

I’ve had doubts about it this week, whether it’s a good project, whether anyone will use it. I think some of that is just the week. Between public holidays, a 9 to 5, and a half marathon in seven days, it didn’t get much of my attention, and the projects I’m not touching tend to look worse than the ones I am. I still want it in the stores. I think it’s a good project, and it doubles as a really practical case study for CMH.

The ten-hour video

The other thing that ate the week was trying to turn my drawing videos into a Claude Code skill, so the process is repeatable instead of starting from zero every time.

We started with Hugging Face models and hit the daily quota. We tried ComfyUI. We ended up on Replicate, which worked out cheapest for what I needed, and then Claude Code edited the clips together directly, which was simpler than I expected. ElevenLabs did the music: something with a bit of a Severance feel, the cheerful celebration intro, which then turns eerie when the budget cuts land.

It took about ten hours, and very little of that was the tooling. Most of it was me going back and forth on how the story should play out and what the visuals should look like. At one point we switched to a whiteboard animation style, a hand placing things into the frame, and then I wasn’t sure about that either.

I’m still not happy with the result. I think the vision could be better. But I think the useful thing I learned is where the ten hours actually went. If I make another video skill, I’ll decide the story on paper before I open a terminal.

What I’m still excited about

Tarot for Builders, mostly. You print the coloring sheets, color in your own deck, and then play a game with it. A lot of us are pretty overstimulated right now, and something you make with your hands and then play away from a screen feels like the right shape for that. I’ve also been giving the online pull more personality this week. It now remembers what you drew last time and asks how it went.

The half marathon is next week, so the next post might just be about that.

References