Hook rate is three-second video plays divided by impressions. Hold rate is fifteen-second ThruPlays divided by three-second plays, except in the tools that divide by impressions instead, which is why the same ad shows two different hold rates depending on where you read it. Both diagnose attention. Neither reliably predicts sales.
What exactly is hook rate?
The share of people served the ad who watched at least three seconds of it. Meta counts the numerator as three-second video plays, and the reporting tools generally surface that same field rather than recomputing it. Generally, not certainly: none of the major tools publishes a definitions page you can check, so treat cross-tool comparability here as likely rather than guaranteed.
hook rate = 3-second video plays ÷ impressions
Also called thumbstop rate.
It measures one thing well: whether the first frame and the first second stopped the scroll. That's a real job, it's the job most creative fails at, and a metric that isolates it is worth having. What the metric doesn't know is who stopped, why they stopped, or whether they had any interest in buying anything.
What exactly is hold rate, and why do two tools disagree about it?
Hold rate is meant to answer a second question: of the people who stopped, how many stayed. The common definition divides fifteen-second ThruPlays by three-second plays.
hold rate = 15-second ThruPlays ÷ 3-second video plays
The common definition. Retention measured among people who already stopped scrolling.
Other tools and agencies use impressions as the denominator instead. That's a different metric wearing the same name, and it always returns a smaller number, because impressions are always larger than three-second plays.
hold rate = 15-second ThruPlays ÷ impressions
The other definition in circulation. Retention measured across everyone served.
Neither is wrong. They answer different questions, and the problem is entirely that they share a label. A creative reported at 28% hold rate in one place and 7% in another hasn't changed. Nobody has made an error. The two systems divided by different things and neither said which.
Why does a better hook make your hold rate look worse?
Because on the common definition, the hook rate is inside the denominator. Improve the first second and more people enter the pool that hold rate is measured against, which drags the ratio down even when the number of people watching fifteen seconds hasn't moved at all.
Take an ad where fifteen-second ThruPlays are a steady 6% of impressions. Hold that fixed and vary only the hook.
The same retention, eight different hold rates
Hold rate over the 3-second denominator, plotted against hook rate, with 15-second ThruPlays held at 6% of impressions.
A line chart showing that when fifteen-second ThruPlays stay fixed at six percent of impressions, hold rate computed over three-second plays falls from forty percent to twelve percent as hook rate rises from fifteen percent to fifty percent. Nothing about how long people watch has changed.
Arithmetic, not measurement. Denominators as defined above.
The ad with the 50% hook rate reports a 12% hold rate and the ad with the 15% hook rate reports 40%. Judged on hold rate alone you'd kill the better ad. This is the single most common way these two metrics are misread, and it happens because they get put in adjacent columns and treated as independent.
The fix is to look at fifteen-second ThruPlays over impressions when you want to compare retention across creatives with different hooks. That figure moves only when retention actually moves.
Do hook rate and hold rate predict sales?
Not reliably, and the honest position is that nobody has published a study either way. What exists is a widely shared observation among people who buy media — a strong hook rate sitting beside flat sales is one of the most common patterns in a creative report — and a mechanism, below, that explains why you would expect exactly that. Neither is a correlation coefficient. If you want one for your own account, you already have the data: plot hook rate against cost per purchase across your last fifty creatives and look at the shape.
Hook rate measures whether the thumb stopped. It has no opinion about the wallet.
The mechanism isn't mysterious. A three-second play is triggered by anything arresting, and arresting is cheap. A loud noise, a face at an odd angle, a fast cut, a text overlay that promises something the product doesn't deliver. All of it raises hook rate. Some of it raises hook rate by attracting exactly the people who will never buy, which lowers conversion rate while the creative report gets greener.
There is a second reason the correlation is weak, and it is structural. Hook rate is measured over impressions Meta chose to serve, and Meta chooses who to serve based on its own conversion model. So the audience behind your hook rate isn't held constant between two creatives, which means the comparison you think you're making is confounded before you start.
What do these metrics actually diagnose?
Two specific things, both of them about attention and neither of them about money.
Hook rate tells you whether the opening frame is doing its job, which is worth knowing because a bad opening frame is fixable in an afternoon and a bad offer isn't. If two creatives share a body and differ only in their first three seconds, hook rate is the right way to pick between them.
Retention over impressions tells you where in the video attention is dying. Watching three-second, fifteen-second and ThruPlay figures together sketches a decay curve, and the shape of the drop points at a section. A cliff between three and fifteen seconds is a promise-and-payoff problem, where the hook set up something the next ten seconds did not deliver. A slow slide across the whole video is a pacing problem.
Your hook rate is high and sales are flat. What now?
Don't touch the hook. A high hook rate with weak sales means the creative already did the thing hook rate measures, so the leak is downstream of it. There are four places to look and they are worth checking in this order.
The offer. Cheapest to change and most often the answer. If the ad promises something the product page does not repeat and honour within a scroll, the attention you bought evaporates on arrival.
The landing page. Baymard's synthesis of 50 studies puts average cart abandonment at 70.22%. In its separate checkout survey, which excludes people who were only browsing, the largest stated reason is extra costs — shipping, tax and fees — being too high, at 40%. A creative can't fix a shipping charge that appears on the final step.
70.22%
Average documented online shopping cart abandonment rate
40%
Largest stated reason for abandoning checkout: extra costs too high
15%
Meta under-reporting for DTC-only brands measuring on click-only attribution
The audience. Broad delivery plus a hook that appeals to a general audience is a reliable way to buy attention from people with no purchase intent. The symptom is a strong hook rate, an ordinary click rate and a poor conversion rate on the clicks that do arrive.
The measurement. Before concluding the creative didn't sell, note that the dashboard reporting it understates Meta's contribution. Across 640 incrementality experiments Haus found Meta under-reporting its own incremental effect by about 15% on average — for DTC-only brands measuring on click-only attribution, which is not Meta's default setting — so a creative sitting just under your break-even may be above it in reality. There is more on that gap in MER, ROAS and what to steer on.
Is frequency the right way to spot fatigue?
Frequency is a lagging indicator dressed as a leading one. It rises after the audience has been saturated, by which point cost per purchase has usually already moved and the decision has been made for you.
The better signal is the share of a day's impressions going to somebody who has never seen the ad before, usually called first-time-impression ratio. It falls before performance does, because it measures the fresh audience remaining rather than the exposure already spent. The bands people quote for it — healthy prospecting somewhere around 65% to 80%, exhausted below about 50% — are unverified. They circulate without a published method behind them and should be treated as folklore until somebody sources them.
The mechanism is sound even if the thresholds aren't. Track the ratio on your own account for a quarter, note where it sat when a creative started decaying, and you've got a threshold that's yours and defensible rather than borrowed.
What should you actually report on a creative?
Four columns, and an explicit note of what each one cannot answer. The failure mode with video metrics isn't that they're useless. It's that they're precise, cheap to compute and adjacent to the metric people care about, which makes them very easy to promote into a decision they cannot support.
| Metric | The question it answers | The question it cannot answer |
|---|---|---|
| Hook rate | Did the first frame stop the scroll | Whether the person who stopped wanted the product |
| Hold rate over 3-second plays | Of the people who stopped, how many stayed fifteen seconds | Anything comparable to a tool using the other denominator |
| 15-second ThruPlays over impressions | How many of everyone served stayed fifteen seconds | Where in the video they left, without the intermediate figures |
| Cost per purchase | What this creative costs to convert at current delivery | Whether the conversion was incremental |
| First-time-impression ratio | How much fresh audience is left in front of the ad | Exactly when it will break, since the published bands are unverified |
Which of these numbers should you act on?
Use hook rate to choose between two openings of the same ad. Use retention over impressions to find where the video loses people. Use neither to forecast revenue, and never compare a hold rate to a benchmark whose denominator you haven't confirmed.
If you want the metrics that do move with outcomes, the honest ones are cost per purchase against your own break-even and the share of spend your best creatives attract. The arithmetic of creative volume covers the second one, and reading a competitor ad library covers the outside-in version of the same signal, where an ad's survival is the only performance evidence available. For the wider pattern of metrics that sound predictive and are not, the ecommerce numbers that do not survive checking collects several more.
Sources
- Motion, Creative Benchmarks 2026, from $1.29B of Meta spend across 578,750 creatives and 6,015 accounts. April 2026
- Haus, The Meta Report, lessons from 640 incrementality experiments. July 2025
- Baymard Institute, Cart abandonment rate statistics, synthesised across 50 studies. Updated September 2025