Hook rate vs thumb-stop, why the platforms measure different things.

TikTok measures a watched view. Meta measures a stopped scroll.

TikTok's hook rate, in its native reporting and in the Creative Center, is computed against six-second video views, the platform treats six seconds as the threshold where a user has chosen to watch, not merely failed to swipe past. Meta's thumb-stop ratio is the share of impressions where the user stayed for three seconds, a much weaker bar, often satisfied by a single legible frame and a beat of motion.

The denominators also differ. TikTok's denominator behaves like impressions in a continuous-play feed, where the platform is auto-playing video in the user's face. Meta's denominator is built on a scroll surface where the user can never actually see the ad, they swipe past the frame before three seconds elapse. The asymmetry matters: a thumb-stop is a victory over scroll. A six-second view is a victory over the next video.

The numbers do not move together.

We scored 2,000+ ads against public engagement signals across TikTok, Meta and YouTube between January 2025 and May 2026. The median scroll-stop time, the window in which most viewers either commit or abandon, varies sharply by surface.

Read the penalty column as: how much our model docks Beat the Skip when the creative places its first meaningful payload, face, stakes, or spoken claim, after the surface's median scroll-stop. TikTok punishes lateness hardest because the platform's whole grammar is built on the next video being one swipe away.

Winning the first second does not save the next five.

A consistent pattern across the sample: ads that score above 70 on hook can still collapse on retention. In one cluster of 84 TikTok beauty ads, mean hook was 71 and mean retention was 38. The first frame did its job, a face, a clean stake, a pattern-break, but seconds two through six contained no escalating reason to keep watching.

The implication for edit decisions is uncomfortable: hook and retention are not the same problem and rarely share a fix. A stronger hook will not patch a flat middle. A tighter middle will not save a creative whose first frame fails to stop the thumb.

Four edit moves that actually move hook.

From the structural patterns in the highest-scoring 200 ads against Beat the Skip:

• Visual stakes in frame 1. A face mid-reaction, a thing about to fall, a hand mid-pour. Static product photography in frame 1 underperforms by roughly 30 percent in predicted hook on TikTok.
• Spoken promise before brand. Voice-over that names the user's problem within 1.2 seconds correlates with a +14 lift in predicted hook. Brand mention before promise correlates with a small negative.
• Pattern-break audio. A sharp non-musical opening sound, a snap, a beep, a voice from off-screen, triggers attention re-allocation even when the visual is mid-strength.
• Caption-first composition. Roughly 78 percent of TikTok viewing in our sample is in a context where the user can see captions but may not hear audio cleanly. Frame 1 with a legible caption hook outperforms voice-only by a meaningful margin.

It depends on whether you are editing the creative or buying the media.

If you are creative-led, pre-spend

Optimise structurally, for the things that produce thumb-stop irrespective of platform. The four moves above are platform-agnostic precisely because they encode how human attention works on vertical short-form, not how a specific ad platform happens to count it this quarter. Pre-spend, you should not be reading hook rate at all, there is no hook rate yet.

If you are media-led, post-spend

Use the platform's own metric as feedback, but read it with the right denominator in mind. A 1.5x lift in TikTok hook rate and a 1.5x lift in Meta thumb-stop ratio are not the same achievement. The TikTok lift is harder because the bar is higher and the next-video pressure is constant.

What this guide is not.

SaliencyLab predicts engagement and click-intent signals against public outcome data. The Spearman correlations we publish, ρ +0.30 to +0.32 out-of-sample, held out, are calibrated against TikTok engagement, TikTok CTR percentile, and YouTube view counts. They are not calibrated against attributed sales, ROAS, or measured brand recall, and we do not claim they are.

Heatmaps in the product are predicted attention, not measured eye-tracking. Score values are model predictions, not measurements. The right way to read a Beat the Skip score of 72 is "this creative shares structural properties with ads that historically held the first six seconds on TikTok," not "72% of viewers will watch six seconds."

The other five guides in this hub.

Questions we get from creative strategists.