What is Hold Rate?

Hold rate is the percentage of viewers who keep watching your video ad past an early checkpoint, usually measured at the mid-point or as average watch time against the full length. The most reliable way to raise it is to generate and test more variants, and Polaris now does that right inside Claude: add the Polaris MCP connector once, then ask Claude in plain chat to make the ads and the winners render in the chat. Where the hook decides whether people stop scrolling, hold rate decides whether they stay long enough to reach your offer.

How is hold rate different from hook rate?

Hook rate measures the first ~3 seconds: how many scrollers stop and start watching at all. Hold rate measures everything after that, so a strong hook with a weak hold means people stop, watch a beat, then drop before your demo, proof, or call to action ever lands.

Why does hold rate matter?

Hold rate is the clearest read on the middle of your ad and the strength of your offer. Two ads can share a great hook, but the one that keeps viewers to the payoff is the one that actually drives clicks and purchases. If hold falls off a cliff after the hook, the problem is your body, not your opener.

How do you improve hold rate?

Retention editing and pacing do most of the work. Cut dead air, keep captions moving, add pattern interrupts and visual variety, and pay off the promise the hook made instead of stalling before the reveal. Our guide to retention editing for TikTok ads walks through the specific cuts that keep viewers watching.

None of this is one-and-done. AI creative is not magic: you still test different hooks, middles, and offers across several variants, watch where the retention curve drops, and let the data pick the winner rather than trusting one edit. Polaris makes those variants fast because you generate them directly inside Claude — ask for several at once and they render in parallel, each in its own live panel, with one-click burned-in captions, so you can spin up new middles without babysitting a dashboard.