New: Coeus, our open Video Jev model. It watches and listens to a video and answers questions about it, with a probability you can trust. Meet Coeus.

The log

6 min read

Theia's first 100,000 views: four AI accounts, one week, one format that broke out

From 26 October to 1 November, Theia ran four AI accounts on Instagram Reels and TikTok: 42 posts and about 112,000 views. Every Reel past 4,000 views was a spoken scene of 26 to 30 seconds with a skip rate near 20%. TikTok did not move.

  • theia
  • analytics
  • instagram
  • tiktok

Entry · first-100000-views

Theia is our AI content studio. She picks the idea, makes the film, posts it, reads the analytics and changes her own code. Each film records the exact version of the code that made it.

Until last week she ran one account. From 26 October she ran four, each a test account with its own character and its own daily film. This post covers the first week of all four: 42 posts from 26 October to 1 November. Between them, they passed 100,000 views for the first time.

For the story so far, read How Theia learned to make TikToks. Last week's post is What Theia learned, week 2. The numbers are real and unedited, including the bad ones.

The week in short

26 Oct to 1 Nov
Accounts4
Posts42 (28 Instagram Reels, 14 TikToks)
Viewsabout 112,000
Views on Instagram109,066 (97%)
Views on TikTok2,883
Biggest film40,846 views
Reels past 4,000 views5

Views are each platform's own count, at the newest read, 1 to 7 days after posting. Instagram and TikTok count a view differently, so compare within a platform, not across.

The four accounts

AccountWhat it postsReelsInstagram viewsBest Reel
CleoGlossy spoken scenes in London695,43040,846
NoorA 90s romcom sitcom, from 1 Nov65,7254,986
Small ProblemsClay animation about daily annoyances66,4381,928
The first accountHeartbreak, one person on camera101,473174

One account made 85% of the views. That is the first thing to be honest about: this is one breakout and a few early signs, not four accounts growing.

What broke out

Five Reels passed 4,000 views. All five are the same kind of video: a spoken scene between people, 26 to 30 seconds long, with a line of conflict said in the first second and a clear turn at the end.

FilmAccountLengthSkip rateAvg watchViews
Be a loveCleo30 s19%26.4 s40,846
Fat in thatCleo30 s19%23.0 s30,106
Members onlyCleo30 s14%26.1 s12,335
Table for oneCleo30 s16%23.7 s11,927
Mum's earlyNoor26 s22%16.5 s4,986

Skip rate is Instagram's share of viewers who scroll past in the first seconds.

Cleo's first two films were a 13-second morning vlog and a 14-second train trip home. They got 81 and 135 views. On 29 October she moved to spoken scenes, and her first one got 30,106.

Noor made five films in other formats from 27 to 31 October, at 57 to 247 views. On 1 November she moved to a 90s sitcom with a laugh track, and that film got 4,986.

Skip rate split the winners from the rest

Every Reel past 4,000 views had a skip rate of 22% or less. Every Reel under 2,000 views had a skip rate of 30% or more. There was no overlap between the two groups.

Skip rate against views, every Reel this week

Past 4,000 viewsUnder 2,000

Members only
14.3%
Table for one
16.4%
Be a love
19.3%
Fat in that
19.3%
Mum's early
22.3%
The chilli
29.7%
The tab
29.8%
The system
33.1%
How are you
38.9%
First account, best
33.3%
First account, worst
64.0%
The Paris trip
67.9%

Lower is better. The five Reels past 4,000 views all sit at 22% or less.

Average watch time points the same way. The five breakouts held people for 16.5 to 26.4 seconds on average. The first account's Reels held them for 4.9 to 7.3 seconds.

Length alone did not do it

It is easy to read the table above as "make 30-second films". The week says no. Noor's Paris trip ran 36 seconds, had a skip rate of 68% and got 57 views, the lowest of any Reel on the account.

Small Problems is the in-between case. Its first spoken episode, the tab, ran 29 seconds and doubled its average watch to 16.7 seconds. But its skip rate stayed at 30%, and its views stayed at 1,865, in line with the account's earlier best. Its first seconds still lose too many people.

So the length gave room for a story. What earned the views was that people did not scroll past at the start.

TikTok did not move

The same films also went to TikTok on two of the accounts. TikTok gave them 93 to 341 views each, whatever the format.

FilmInstagramTikTok
The chilli1,898118
The system1,928123
The tab1,86593

The TikTok account for Small Problems is new and has 4 followers, so this may be the account and not the film. The first account's TikToks stayed at 225 to 318 views, the band it has been in since August. Neither of the two accounts that broke out on Instagram has posted to TikTok yet, so no film that won on Instagram has been tested there.

Views are not followers yet

Cleo's 95,000 views brought her from 883 to 970 followers. That is less than one new follower per 1,000 views. Almost all of her views came from people who did not follow her. Likes ran at about 1 per 100 views on the biggest films.

People watched to the end. Few of them came back for more.

What we cannot claim yet

  • That it repeats. Cleo's last two films got about 12,000 each, down from 40,846. One week is not a trend.
  • Why TikTok is flat. We have a guess, the account and not the film, and no test of it yet.
  • That skip rate causes views. It separated the two groups cleanly this week. Instagram may show more people a film because they stay, or people may stay because Instagram showed it to the right people. We cannot tell the two apart from this data.
  • That the format is the reason. Each account changed more than one thing when it changed format.

Next week

The first account moves to the spoken format from 2 November, on both Instagram and TikTok. That is the first test of a spoken scene on TikTok: if it still stops at about 250, the film is not the problem. Small Problems works on its first seconds, where its skip rate is stuck at about 30%.

If you want the same loop on your brand's account, start with a pilot month. For how a loop like this works, and how it can go wrong, read Recursive self-improvement in practice: lessons from the STOP paper.

Next entry

What Theia learned, week 2: 9 AI TikToks, 7 formats, one view ceiling

Week 2 of our AI TikTok account in numbers: 9 films from 22 to 30 October, with views, likes, engagement rate, 2-second hold and code version. Seven formats all stopped at about 240 views. Likes and retention did not.

Read it