On this pageJump to a section
When a video ad underperforms, the instinct is to rewrite everything. New script, new editor, new music, new offer. That is how a good middle section gets thrown out because the opening was weak.
A video ad can fail in exactly three places. The first seconds failed to stop anyone. The middle lost the people who stopped. Or the end did not give anyone a reason to click. Three ratios will tell you which, and each one points to a different fix.
What do the video metrics in Ads Manager actually count?
Know the raw counts before you build ratios on them.
Per Affect Group's summary of Meta's video metrics, a 3-second video play counts when a video plays for at least 3 seconds, or nearly its full length if shorter, with replays within the same impression excluded. ThruPlays are plays that reach completion or run at least 15 seconds. Video average play time includes replay time, so it can exceed the length of the video. Plays at 25%, 50%, 75%, 95% and 100% count plays that reached that point, including plays that skipped ahead.1
Hook rate and hold rate are conventions
The raw counts are Meta's. The ratios are the community's. Affect describes hook as based on early-view metrics and hold in terms of completion ratios without fixing a single formula.1 That is the honest state of things, so here are the definitions I use. If you choose different ones, fine. Just pick them, write them down and stay consistent.
| Ratio | How I calculate it |
|---|---|
| Hook rate | 3-second video plays divided by impressions |
| Hold rate | Plays at 50% divided by 3-second plays |
| Click rate | Link clicks divided by impressions |
| Cost to act | Spend divided by the result you care about |
Where is the ad failing?
Read the three ratios against each other, not alone.
- Hook3-second plays divided by impressions
- HoldPlays at 50% divided by 3-second plays
- ClickLink clicks divided by impressions
- FixChange only the first section that fails
A low hook rate: the opening failed
Few people stopped. The first seconds did not earn attention, and everything after them is irrelevant because almost no one saw it. Nielsen Norman Group's research on web pages found that users often leave within 10 to 20 seconds and that the first 10 seconds are critical to the decision to stay.2 That study is about web pages, not video ads, but the principle carries: people triage fast, and the opening is where the triage happens.
What to change: the first frame and the first line. Start with the most surprising or relevant moment, not a logo or a greeting. Put text on screen for people with the sound off. Test two or three different openings on the same body, on the cadence in Creative Is the New Targeting.
Dara Denney made a related point in a video she posted on August 28, 2025, about the hooks her teams had tested. The video includes a Storyblocks sponsorship, so read it with that in mind.4
VideoDara Denney
I tested over 1000 ads. Here are the hooks ACTUALLY making MONEY
Ten tested hook types in under twelve minutes, and the visual hook matters more than the words.
Essentially, the best hooks show, they don't tell.
That is why I say first frame before first line. If the picture does not stop someone, the words never get a chance.
A healthy hook but a low hold rate: the middle failed
People stopped, then left. The opening promised something the middle did not deliver, or the pace dropped, or the point got buried.
What to change: the structure after the opening. Cut anything that does not move the story forward. Keep the promise the hook made and pay it off quickly. Add a visual change or a new idea every few seconds so the ad does not go flat.
Healthy hook and hold but a low click rate: the ending failed
People watched and still did not act. The ask was unclear, the offer was weak or the video made the product feel interesting but not necessary.
What to change: the call to action and the offer framing. Say exactly what to do and what happens next. Put the benefit in the same breath as the ask. Check that the landing page matches what the ad promised.
What else can distort these numbers?
A few things, and they matter.
Placement and sound
An ad that auto-plays silently in a feed behaves differently from one on a sound-on placement. Always compare ads on the same placement, or break results out by placement before you decide anything.
Replays and average play time
Meta's 3-second plays exclude replays within the same impression, but video average play time includes them.1 If the average play time looks too good, check for replays before you celebrate.
Small samples
Three hundred impressions is a coin flip. Wait for enough volume that a difference between two ads is bigger than random wobble, and do not kill an ad on one rough afternoon.
How does this fit with how Meta delivers ads?
I cannot see inside the model, so I keep this modest.
Meta's engineering team describes its large ads model, GEM, as learning from ad content and engagement data, including long sequences of user behavior, and says it transfers that learning to the models that rank ads across Facebook and Instagram.3 I do not claim to know how any single signal, like a three-second play, is weighted. What I do know is that creative which earns attention and then earns the action is the creative you want the system to find, and that the three ratios are the cheapest way I know to tell you whether yours does.
How do you run a clean test?
- Choose the failing section from the ratios. Name it in writing.
- Make two or three versions of that section only. Keep everything else identical.
- Run them in the same ad set conditions, same placement and same audience.
- Wait for enough volume to see a clear gap, not a hint.
- Keep the winner, then move to the next section.
- Log what you changed and what happened in your tracking sheet.
Keep a library
Every test teaches you something about your audience. Save the hooks that worked, the structures that held and the endings that converted, in a creative bank. Over time you stop guessing at what to make and start building from evidence.
What to watch on Monday
The most reliable creative improvement I know is boring: diagnose before you rewrite. Find the section that failed, change that section and leave the rest alone. I have run Meta ads since 2014, and the client results are public.
Most weak ads do not need a new idea. They need the right part of the old one fixed.
Keep building.Ekai
Questions people ask
What is hook rate in Meta video ads?
Hook rate is a common community metric, not an official Meta column. It is usually calculated as 3-second video plays divided by impressions, showing how many people stopped long enough for the opening to register.
How do you calculate hold rate?
There is no single official formula. One practical version divides plays that reached 50% of the video by 3-second plays, showing how many of the people who watched the opening stayed through the middle. Whatever you choose, define it once and use it consistently.
What counts as a 3-second video play in Meta?
Meta counts a play when the video runs for 3 seconds, or nearly its full length if it is shorter. Replays within the same impression are excluded.
What is a good hook rate?
There is no reliable universal benchmark, so compare against your own recent ads on the same placement and audience. A relative drop or gain between ads tells you more than a number from a blog.
Sources
- 1
- 2
- 3
- 4
