Diagnose Hook Rate, Then Hold Rate
Know whether your video ad is dying at second three or at second fifteen.
Every video ad that stops converting gets the same panicked response. Kill it, write a new hook, try a new audience. That is a coin flip of a fix, because a dead ad can die in two completely different places. It can fail to stop the scroll in the first three seconds, or it can stop the scroll just fine and lose the viewer somewhere in the middle. Those are opposite problems with opposite repairs, and the one number most people watch, overall video completion, cannot tell you which one you have.
What to do: In Ads Manager, add Impressions, 3-Second Video Plays, and Video Plays at 50% as columns for every active video ad. Divide 3-second plays by impressions to get hook rate. Divide 50%-plays by 3-second plays to get hold rate. Do this for every ad before you rewrite a word of copy or touch a single audience.
Why it works: The two ratios isolate two different jobs a video ad has to do, stop the scroll and then hold the story, so a weak number in one and not the other points straight at which half is broken.
Example: Rafael Hernandez, CEO of the performance-marketing agency Great Marketing AI, runs this exact split across the Meta video campaigns his team manages. He treats a 38% hook rate paired with a 22% hold rate as proof the opening is doing its job while the body is losing the viewer's trust, a different diagnosis, and a different fix, than an ad stuck with a hook rate under 20%.
Walk it through
1. Pull the three raw numbers.
Open Ads Manager, go to the ad level, and customize the columns to show Impressions, 3-Second Video Plays, and Video Plays at 50%. These sit under the video metrics group. Do not rely on the default columns Meta shows you by default, they bury 3-second plays and 50%-plays behind extra clicks.
2. Do the division twice, per ad.
hook rate = 3-Second Video Plays ÷ Impressions
hold rate = Video Plays at 50% ÷ 3-Second Video Plays
Hook rate answers "did it stop the scroll." Hold rate answers "did the people who stopped scrolling keep watching." Keep them as two separate columns in a spreadsheet, one row per ad.
3. Read each ad against real benchmarks, not vibes.
Published Meta benchmarks put table-stakes hook rate around 25%, top performers at 30%+, and elite creative clearing 35 to 45%. On the hold side, a rate below roughly 25% signals a pacing problem, and anything above 45% is strong. Great Marketing AI's own floor for hook rate is 20%, below that the creative is not resonating at all, full stop, before the story even starts.
4. Sort every ad into one of four boxes.
| Hold rate weak | Hold rate strong | |
|---|---|---|
| Hook rate strong | Opener works, body fails. Rewrite the middle. This is the 38%/22% case. | Winner. Scale the spend, do not touch the creative. |
| Hook rate weak | Broken everywhere. Replace it, do not iterate on it. | Rare, but means the opener is under-selling a body that actually delivers. Punch up the first three seconds only. |
The box an ad lands in tells you exactly which three seconds of the timeline to open in your editing tool next.
The read
- Hook rate isolates the thumb-stop. It only measures whether the first three seconds earned attention, nothing about the pitch itself.
- Hold rate isolates the pitch. Once someone has stopped, this is the only number that says whether what follows kept them.
- The pairing is the diagnosis. Either number alone tells you an ad is underperforming. Only the pair tells you where in the timeline to go fix it.
Steal it
Run this split on every video ad currently spending, not just the losers. You will usually find a handful of ads with a strong hook and a weak hold, exactly the pattern Rafael Hernandez flags, and those are gifts. The opener already works. You do not need a new concept, a new actor, or a new hook idea. You need to sit with the 3-to-15-second window of that specific ad and fix the sagging part, tighten the proof, cut the dead air, move the ask earlier.
Defend against the opposite trap too. It is easy to punch up a hook until it is pure clickbait, a curiosity gap or a shock cut that stops the scroll for people who were never going to buy. That inflates hook rate while quietly cratering hold rate, because the promise in the first three seconds does not match anything in the next twelve. If your hold rate is falling across the board while hook rate climbs, the hook is the thing lying, not the body.
Gotchas
- Low-spend ads give you noise, not a diagnosis. An ad with a few hundred impressions can swing 15 points on either metric by chance. Wait for real volume before you trust the box it lands in.
- Benchmarks differ by placement. A hook rate that is strong in Feed can look weak in Reels, because the format trains different scroll speeds. Compare an ad against its own placement's baseline, not a single number pulled from a blog post.
- This tells you where, not what. Hook rate and hold rate point at a stretch of timeline, they do not tell you which line, which cut, or which proof point to change. You still have to watch the footage and make a call. Honest caution: treat this as triage, not a replacement for actually watching your own ads with sound on.