01, The opening matters more than the rest.
The first frame of a vertical video carries roughly 40% of the entire attention budget the spot will ever earn. By the time the brand reveal arrives at second 4, you are not addressing the audience you started with, you are addressing the fragment of it that decided not to flick upward.
This is not a stylistic preference. It is a scroll-mechanic. The thumb is already mid-motion when your ad enters frame; the opening either interrupts that motion or it does not. Everything downstream, the brand, the offer, the CTA, is conditional on that interruption happening.
"If the first two seconds do not earn the next two, the rest of the spot is talking to nobody. The most expensive frames in advertising are the ones that came after the viewer already left."
02, The hook windows, by surface.
Not every surface has the same window. The skip button, autoplay behaviour, and the audience's expectation of length all shift the budget you have to land the promise. Below, the windows we use when scoring Beat the Skip, with the penalty applied when the promise lands after the window closes.
| Surface | Window | Median scroll/skip | Penalty if missed |
|---|---|---|---|
| TikTok For You | 0 – 2.0s | 1.9s | −18 pts |
| Instagram Reels | 0 – 2.5s | 2.3s | −14 pts |
| YouTube Shorts | 0 – 3.0s | 2.7s | −10 pts |
| YouTube In-Stream | 0 – 5.0s | 5.0s (skip) | −9 pts |
| Meta Feed | 0 – 3.0s | 2.6s | −11 pts |
Median scroll/skip times are derived from public platform disclosures and our own held-out cohort of 1,200+ ads with engagement outcomes. Penalty figures are the Beat the Skip deduction applied when the spoken or visual promise arrives after the window closes.
03, Ten openings that won.
Each of the patterns below is a composite, described by mechanic, not by brand. We see each one repeatedly across our cohort. Score shown is the median Beat the Skip across spots using that mechanic in the first 2.0 seconds. Strong is ≥ 70.
What ties these together is not aesthetics, three of the ten openings are deliberately ugly. What they share is that frame 1 carries a promise, a stake, or an unresolved question. Something has to be answered, and the only way to get the answer is to keep watching.
04, Ten openings that lost.
Same rubric, same window. Below are the ten composites that consistently land in the weak band, Beat the Skip under 50. Each of them costs the spot roughly half the attention budget before the brand even arrives.
"A brand sting in frame 1 is the most expensive mistake in vertical video. It spends the most valuable attention seconds of the whole spot communicating a thing the audience has not yet asked to know."
05, Four moves that consistently scored.
Across the ten winners, and across our held-out cohort of 1,200+ ads, four mechanics keep recurring. None of them is expensive to execute. None of them depends on a celebrity, a studio shoot, or a media plan above five figures.
1. Visual stakes in frame 1
Something is at risk, mid-motion, or unresolved. The hand is reaching, the liquid is mid-pour, the face is mid-laugh. Frame 1 carries a question the viewer wants closed.
2. Spoken promise before brand
The audio carries the value proposition before the brand identifier arrives. The order is promise → demonstration → attribution, not the reverse.
3. Pattern-break audio
A drop, a cut, an off-rhythm beat, a hard accent, any audio event that does not match the surrounding feed. The ear catches what the eye is still parsing.
4. Caption-first composition
The shot is composed so that an on-screen caption can sit cleanly above the face or hand without occlusion. Sound-off readability is built in, not retro-fitted.
06, Four moves that consistently failed.
The losers are not random. They cluster into four anti-patterns. If your opening matches any of them, the prediction is a Beat the Skip in the low 40s before the rubric even reaches the rest of the spot.
1. Brand sting first
Logo card, audio mnemonic, or animated mark before any value claim. Burns the most-watched seconds on attribution the viewer has no reason to want yet.
2. Slow establishing shot
Landscape, drone push, ambient lifestyle, beautiful, but it tells the scroll-mechanic nothing is happening. The thumb keeps moving.
3. Talking-head intro
"Hi, I'm…" / "Today I want to talk about…" The opening defers the promise by an entire sentence. By the time the promise lands, the window is closed.
4. Generic category opener
Studio product on white, generic gradient, stock B-roll. Indistinguishable from any other ad in the category. Pattern-match without pattern-break.
07, Audit your opening in 60 seconds.
You do not need our pipeline to do this part. Scrub your spot to exactly 2.0 seconds and pause. Ask a colleague who has never seen the brief the questions below. If any answer is "no", re-edit before you spend a euro on media.
- Can a stranger state the promise from what is on screen and what they just heard?
- Can a stranger state the stakes, what is at risk, what is being resolved?
- Can a stranger guess (within a category) what the brand is, or what it will be?
- Is there at least one pattern-break in audio or visual, something that does not match the surrounding feed?
- Is the composition sound-off readable, caption-first, face or hand prominent?
- Is there no brand sting in the first 2.0 seconds?
Six checkboxes. If you cannot tick all six, the opening is the bottleneck, not the offer, not the CTA, not the media plan.
08, The honest caveat.
Beat the Skip is a model prediction. It is not measured eye-tracking, and it is not a sales or ROAS forecast. The validation we run is against public engagement and click-intent signals, likes, shares, comments, view counts, CTR percentile bands, across a held-out cohort of 1,200+ ads. Held-out OOS Spearman ρ as of May 2026: +0.31 on TikTok engagement (n=700), +0.30 on TikTok CTR (n=691), +0.32 on YouTube view counts (n=403).
That correlation is strong enough to rank-order openings reliably, top-quintile vs bottom-quintile lift is 6.5× on our cohort, and weak enough that you should treat individual scores as directional. The four winning moves and four losing moves above are the durable patterns, not the exact numbers.
We do not validate against in-market business outcomes, sales lift, attributed conversion, ROAS, brand recall. Those constructs require survey panels and media attribution data we do not collect. If anybody in the category tells you they predict those from a 90-second scoring run, ask to see the held-out chart.
Keep reading.
Frequently asked.
How long is the actual scroll-stop window on TikTok?
Median scroll-stop on TikTok sits around 1.9 seconds, the viewer has effectively decided whether to keep watching before frame 60. Reels skews slightly longer at roughly 2.5s, YouTube Shorts at roughly 3.0s, and YouTube in-stream pre-roll extends to about 5.0s because the skip button is delayed.
What is the single most common reason an opening loses?
Leading with a brand sting or logo card. It uses the highest-attention frames of the entire spot to communicate something the viewer is not yet asking about. The promise must arrive first; the brand attribution comes inside the same shot or immediately after.
Does Beat the Skip predict sales?
No. Beat the Skip is a model prediction validated against public engagement and click-intent signals, likes, shares, comments, view counts, CTR percentile bands. Held-out OOS Spearman ρ is +0.30 to +0.32. It is not measured eye-tracking, and it is not a sales or ROAS forecast.
Can I audit my own opening without running it through SaliencyLab?
Yes, pause your spot at exactly 2.0 seconds and ask three questions. Can a stranger state the promise from what is on screen? Can they state the stakes? Can they identify (or guess) the brand? If any answer is no, re-edit the opening before you spend on media.