What Theia learned, week 4: six AI TikToks, all past the first batch

Every week we publish what Theia, our AI content studio, learned from her own films. She runs our TikTok account on auto-post: 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, so every number below points back to one version.

This week covers the 6 films posted to TikTok from 6 to 12 November. There was no film on 10 November: every video provider refused the job that day, and the film was made the next day from the boards she had saved. Last week's post is What Theia learned, week 3. For the whole story from the first slideshows, read How Theia learned to make TikToks. The numbers are real and unedited, including the bad ones.

The week in short

	This week
Films posted to TikTok	6
Format	Spoken scenes, 28 seconds
Views at 6 hours	661 to 1,057
Films past the first batch (about 230)	6 of 6
Best 3-second hold	57% (The Charity Shop)
Likes per 100 views	3.1 to 7.0
Best Instagram skip rate	16.0% (The Charity Shop)
Tests that worked	1 of the 5 read so far

The films

"Held at 3 s" is the share of TikTok viewers still watching three seconds in. "Code" is the version of Theia's code that made the film. Views and likes are the newest read, on 12 and 13 November.

Film	Posted	First frame	Views at 6 h	Views	Likes	Held at 3 s	Finished	Avg watch	Code
The ring	6 Nov	Her crying face, large, at a jeweller's	672	686	40	42%	24%	9.5 s	b9954cd
The charity shop	7 Nov	Two-shot across a shop counter	1,057	1,259	39	57%	34%	13.5 s	289f9f9
The check-in	8 Nov	Two-shot at an airport desk	941	1,043	70	42%	34%	12.0 s	33c48f1
The chip	9 Nov	Two-shot at a vet's counter, her dog	661	673	43	42%	15%	8.3 s	f7c5d3f
The locks	11 Nov	Her front door, a locksmith	754	774	54	41%	26%	10.4 s	0808c22
The kids' table	12 Nov	A wedding table, her ex behind	904	942	57	44%	26%	10.7 s	67bccb9

Every film is a different code version, because Theia changes one thing after each result. The viewers were 94% to 96% women and about 94% to 97% in the UK, the same audience as before.

BeforeThis week

Median, 14 Oct to 4 Nov

232

The ring

672

The charity shop

1,057

The check-in

941

The chip

661

The locks

754

The kids' table

904

The first bar is the median of the 22 films from 14 October to 4 November, whose median stayed at TikTok's first batch.

TikTok's push is fast for these films too. Each had 638 to 897 views by 3 hours, and grew less than 20% after 6 hours. We measured this pattern on 38 films in How the TikTok algorithm hands out views, hour by hour.

What the six films share

For a month, every film on this account stopped at about 230 views, whatever the format. Last week, The Table (5 November) was the first to break that, with 644. Theia wrote down what it had that the films before it did not, and kept it as a fixed shape:

• Frame 0: she is crying in a public place, and her first line is spoken from the first frame.
• A stranger who works there takes her side: a shop assistant, a check-in agent, a vet nurse.
• A turn that sends the ex's own thing back at him: the ring he said was worth a fortune is glass, his "designer" shirt goes in the window at 50p.
• The last line hands his words back to him.

Every film this week kept that shape and changed one other thing. All six left the first batch, so the shape is the strongest result this account has had. It is still one account and six films, and each film changed more than the one thing it was testing.

One test per film

Theia runs other accounts too, and they share what they learn. This week, each film tried one finding from another account, with the aim written down before the film went out.

Film	What it tested	Aim	Result
The ring	Her face large in frame 0, the other person's hand already reaching in	Over 50% held at 3 s on TikTok	42%. Did not work
The charity shop	Both faces in frame 0, a two-shot	Over 50% held at 3 s on TikTok	57%. Worked
The check-in	A situation most viewers have lived: a holiday booked as a couple	10,000 Instagram views in a day, with more likes per view	4,655 views. Did not work
The chip	A first line that asks for something, with a stake: "Don't let him take my dog."	Instagram skip at or under 19%	24.3%. Did not work
The locks	A stranger at odds with her in frame 0, at her own front door	Instagram skip at or under 20%	32.4%. Did not work
The kids' table	The problem readable from the picture alone in second 0	Instagram skip at or under 22%	Not in yet. TikTok 3 s hold 44%, under its 50% aim

The one that worked changed the second that mattered most. The Ring held 81% at 1 second and 61% at 2 seconds, then fell to 42% at 3. The Charity Shop held 81% and 65%, and kept 57% at 3. The drop in second 3 shrank from 19 points to 8. The next film kept the two-shot, and its 3-second hold fell back to 42% with a different first line, so the two-shot alone is not the whole answer.

On Instagram, the push shrank

Every film also goes to Instagram Reels, and from 7 November to a Facebook Page. Instagram gives a skip rate (the share who swipe away at the start) and an average watch time.

Film	First frame	Skip rate	Avg watch	Views at about 1 day
The ring	Jeweller's counter	16.8%	28.9 s	15,319
The charity shop	Shop counter	16.0%	19.9 s	11,628
The check-in	Airport desk	24.9%	18.4 s	4,655
The chip	Vet's counter	24.3%	16.7 s	2,153
The locks	Her front door	32.4%	14.5 s	198
The kids' table	Wedding table	not in yet	not in yet	not in yet

Public placeHer front door

The ring

15,319

The charity shop

11,628

The check-in

4,655

The chip

2,153

The locks

198

Read at about 21 hours. The Kids' Table is not in yet.

The Ring's average watch is longer than the film, so people watched it more than once. It is the account's biggest Instagram post so far.

What the data supports

1. The shape got every film past the first batch on TikTok. Six of six films. The 22 films in the three weeks before The Table had a median of 232 and a best of about 300. The range was narrow: 661 to 1,057 at 6 hours.

2. On TikTok, the best hold got the most views. The Charity Shop held 57% at 3 seconds and had the most views. The other five all held 41% to 44%, and their views still ranged from 661 to 941, so the hold does not explain the rest.

3. On Instagram, the push followed the skip rate. The two films under 17% skip got 11,628 and 15,319 views in a day. The two at about 24% got 2,153 and 4,655. The one at 32% never left the first batch.

4. On Instagram, a repeated picture seems to wear out. Two different first lines (a statement, then a request) gave the same skip of about 24%. What they shared was a fifth and sixth go at the same picture: crying at a counter, a kind stranger. So Theia changed the place for The Locks, to her own front door, and that skipped worst of all. Her three front-door films so far (two of them last week) skipped 27% to 32%; the six that opened in public skipped 16% to 25%. Her rule for now: frame 0 stays in a public place. TikTok did not split the same way: The Locks still got 754 there.

What we cannot claim yet

• Why the shape works. Each film changed the place, the person and the story as well as the test. We can say the shape kept working, not which part of it does the work.
• That the two-shot is the TikTok lever. It worked once, then did not repeat with a different first line.
• Why the same film does so differently on the two platforms. The Locks was the worst on Instagram and mid-table on TikTok. Same file on both.
• The Kids' Table. Its Instagram read is not in yet.
• Shares, saves and comments. 4 shares, 2 saves and no comments across all six films. Too few to learn from.
• Facebook. The Page took 3 of the films and refused 2. We have no views from it yet.

Next week

Theia keeps the shape and a public place in frame 0, and goes on testing one change per film. The question for week 5 is what lifts TikTok's 3-second hold back over 50%, and what stops the Instagram push shrinking.

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