Five TikTok Formats, One Ceiling: What 2.5 Seconds Told Me

For two weeks in late October and early November, OutfitMaker posted short videos on TikTok and Instagram almost every evening. The goal was simple: find a format that holds attention long enough for the algorithm to push it past its first test batch. I changed one thing at a time and wrote every number down. This is what the data said. The setup OutfitMaker turns photos of the clothes you already own into outfits. The TikTok account targets the UK, where most viewers already were. Every video followed the same rule: the exact search phrase people type (from TikTok's "content gap" insights) on screen in the first second and in the caption. The benchmark was the best video before the test: a 7 second "which outfit?" duel that held viewers for 3.6 seconds on average, with 6.1% watching to the end. Five formats, side by side Format Length Views Avg. watch time Watched to the end 1. AI duel, two looks side by side (benchmark) 7 s about 770 3.6 s 6.1% 2. Same duel, longer, with a "comment 1 or 2" hook 10 s 744 2.5 s 1.96% 3. Same idea in a bright "game show" style 8.5 s 768 2.4 s 2.28% 4. Real screen recording of the app, "POV" hook 9.8 s 734 2.4 s 0.69% 5. Finished look full screen at 0:00, with a voice 10.2 s 277 (day 1) 2.5 s 1.43% Three things stand out. The audience was never the problem. Across the tests, women made up 66 to 81% of viewers and 69 to 87% were in the UK. On the videos where TikTok showed the full age split, 83 to 93% of viewers were 35 or older. That is exactly who I wanted to reach. The ceiling did not move. The first four formats landed between 734 and about 770 views, roughly the size of TikTok's first test batch for a small account, and the fifth was at 277 after one day. None earned a second push. Viewers left at the same moment. Whatever I changed (length, colours, a real app recording, a voice, the result shown first), most people swiped away at about two seconds. Engagement stayed near zero: one like on the last video, no comments, no shares. The "comment 1 or 2 and we will style the winner tomorrow" mechanic got zero comments on 744 views, so I dropped it. What social did not bring From 20 October to 9 November, organic TikTok sent one visit to the site and Instagram none, against 8,880 from Google. Every paying customer and trial I could trace since September came from Google search. A small paid TikTok test brought 93 visits for 21 € and no sign-ups at all. Part of that was a bug: TikTok's in-app browser blocks Google sign-in, and that button was first on the page. I wrote about that one separately. Fixing it did not change the bigger picture. Search was quietly doing the work: 10,031 clicks from Google in 28 days, up 27% on the previous period. The lesson Short video is a slot machine with a two-second handle. The formats were reasonable, the targeting was right, and the data was consistent: watching an AI dress someone else does not stop the thumb. So I stopped producing. The account keeps a light presence (one or two posts a week, no heavy production), and the time goes where customers actually come from: pages that answer what people search for, and what happens after they land. The one variable I never tested is a real person talking to camera. If I try anything else on TikTok, it will be that, and it costs a phone and ten seconds. Originally published on Meridian Build.