Hold rate is the video metric I actually watch, and my definition has two conditions: more than half the video viewed, and more than three seconds watched. If somebody cleared both, the opening did its job, which is why I do not measure the opening separately.
That single substitution removes a metric with no sourceable benchmark and loses nothing.
This is for you if you are tracking several video metrics and are not sure which one should change a decision.
What hold rate measures
Whether people stayed, rather than whether they started.
A video view is a very low bar. Autoplay, an accidental pause while scrolling, and a moment of confusion all produce one. Counting those tells you about the feed rather than about your advert.
Requiring both conditions filters most of that out. Three seconds removes the accidental. Half the video removes the briefly curious.
What remains is people who chose to keep watching, which is the population worth knowing about. That is a much smaller number than your view count, and it is a considerably more honest one.
Why hold rate replaces the hook measurement
Because you cannot reach the second condition without having satisfied the first.
Somebody who watched past three seconds and beyond the halfway point was hooked. That is not an inference, it is arithmetic. The hook worked or they would have gone.
So measuring hook rate alongside gives you a partial view of something already fully captured. Another number, another row in the report, no additional decision.
There is a sharper reason too. Optimising towards a hook measurement specifically rewards openings that stop people without delivering, because the metric is satisfied at three seconds and does not care what happens next. Pointing an incentive at that is asking for bait.
Hold rate cannot be gamed the same way. A misleading opening produces a strong start and a collapse, and the collapse is what gets measured.
There is no hold rate benchmark either
Worth stating explicitly, because the substitution does not solve the sourcing problem.
No published benchmark with a disclosed methodology exists for hold rate, exactly as with hook rate. Every figure circulating comes from the same layer of vendor content citing other vendor content.
So the metric is better and the reference point is equally absent. Use it to compare your own creatives, on the same audience and placement, and never against a published target.
That is a smaller claim than most articles on this subject make and it is the accurate one. A metric with no external standard can rank your options. It cannot tell you whether you are doing well in absolute terms, and nothing in this space currently can.
What the shape of the curve tells you
The single number is less useful than where retention falls away.
If hold rate is poor and the drop is immediate, the opening is not landing. If it is poor and the drop happens consistently at a specific point, something at that point is losing people, and you can go and look at what.
That diagnosis is the real value here, and it is invisible if you only ever read the summary figure.
It also tells you how long the video should be. When retention collapses at nine seconds, the material after nine seconds is doing nothing, and cutting it costs you nothing. Meta’s own documented placement requirements accommodate five to fifteen seconds across every placement, so shortening rarely creates a compliance problem.
Where hold rate stops mattering
It is a diagnostic, and diagnostics lose to outcomes.
A creative with a modest hold rate and a good cost per result is working. I have watched accounts pause profitable adverts because an upstream video metric looked weak against a benchmark nobody could source, and that is a straightforwardly expensive mistake.
The hierarchy is worth stating plainly. Cost per result decides. The three Meta rankings explain. Video metrics explain the explanation.
So hold rate earns its place when you need to know why something is not working, and it should never override the fact that something is.
What hold rate cannot tell you at all
Two limits worth knowing before building reporting around it.
It has no meaning for static creative, and plenty of accounts run mostly statics. Reporting that assumes every advert is a video produces empty columns and, worse, encourages people to make videos to fill them.
And it says nothing about who stayed. A high hold rate among people who will never buy is not progress, and the metric cannot distinguish that from the alternative. Only the conversion side can.
That second limit is why hold rate belongs next to the conversion rate ranking rather than on its own. Attention from the wrong audience is still attention, and it is worth nothing.
Reading it honestly across creatives
Compare on matched conditions or do not compare.
Same audience, same placement, same period. Placement matters more than people expect here, because a video in a feed and the same video in a story behave completely differently, and averaging across them produces a figure describing neither.
Then look for large differences rather than small ones. At most account volumes, a few percentage points between two creatives is noise, and the same power problem applies here as everywhere else in testing.
When one creative holds people substantially better than another and the cost per result agrees, you have learned something. When the two disagree, trust the cost per result and use retention to work out why.
Why this metric survives platform changes
One practical advantage over most creative reporting, and it is the reason I keep using it.
Retention is measured entirely on the platform, from behaviour the platform observes directly. Nothing about it depends on a pixel firing, a cookie surviving, a consent choice or a click identifier reaching your CRM.
That makes it one of the few numbers in this whole programme that has not been degraded by the last few years of privacy and browser changes. Conversion measurement got harder. Watching whether somebody stayed with a video did not.
So when your conversion data is thin, patchy or under investigation, this remains readable. That is genuinely useful during the periods when everything downstream is uncertain.
It does not make it a substitute for outcome data, and it does make it a reliable second opinion when the outcome data is behaving strangely. If retention is stable and conversions collapsed, the creative is probably fine and the problem is further down the chain.
What to change this week
Three steps.
Set up hold rate on both conditions, more than half viewed and more than three seconds, so you are measuring the thing rather than a proxy for it.
Then look at where retention falls away on your best and worst performers, rather than only at the summary number. The shape is where the diagnosis lives.
Then drop any external video benchmark from your reporting, because none of them are sourced, and replace it with comparison against your own creatives.
Hook rate has no published benchmark covers why the metric this replaces is unsourceable, fund the ad set and read what the algorithm chose is the position underneath, creative volume covers how many creatives you should be comparing, and the weekly rhythm I run puts it into a testing cadence. Organic video has the same measurement gaps, which my SEO work accounts for. To have your creative reporting reviewed, book a teardown.