Hold Rate: A Metric No Standards Body Has Ever Defined
It appears in no industry glossary, no measurement guideline and no platform documentation. The two definitions in circulation divide incompatible metrics.
Search the industry video glossary for “hold rate” and you get zero results. Same in the video impression measurement guidelines. Same in the viewability guidelines. No platform defines it either.
It is a practitioner term, invented in agencies, that has acquired the tone of a standard without ever becoming one. That would be harmless if the arithmetic behind it worked. It does not: the two definitions in circulation divide metrics that the platform itself says are computed on incompatible methodologies.
This page shows what is actually defined, what is not, and what to measure instead.
Where the term does not appear
I looked in the places that would settle it.
The industry digital video glossary. Zero occurrences. It defines video ad completion rate as the percentage of video ads that play through their entire duration to completion, and stops there.
The joint digital video impression measurement guidelines. Zero occurrences.
The viewable advertising impression measurement guidelines. Zero occurrences.
The largest social platform’s help centre. No article. Searching returns unrelated pages about factors affecting view rate and about brand recall lift.
The short-video platform’s metric documentation. Not among the listed metrics.
What that adds up to. No standards body, no measurement council and no platform has ever published a definition. Every definition you have read was written by somebody selling analytics or writing a blog post.
Vendor glossaries offer two, and they are not variants of each other.
Definition one: 15-second plays divided by 3-second plays. The share of people who started the video and stayed to fifteen seconds.
Definition two: viewers past a milestone. The share reaching a defined point, typically 25%, 50% or 75% of the video.
They do not measure the same thing. One is an absolute duration ratio, the other is a proportional one. On a 60-second video, fifteen seconds is the 25% milestone. On a 20-second video, it is the 75% milestone. The two definitions cross over at different lengths, so the same creative gets different hold rates depending on which definition your agency uses.
Neither is authoritative. Both circulate in vendor glossaries, neither appears in a standards document, and nobody arbitrates between them.
And both have an arithmetic problem underneath, which is the part almost nobody knows about.
Duration metrics and milestone metrics are not the same family
This is documented by the platform itself, in plain language, and it invalidates the ratio.
What duration metrics do. The platform states that for these it sums the unique seconds that a video plays, and that time spent replaying the same segments for a single impression is not included. This family covers 2-second continuous plays, 3-second plays and ThruPlays.
What milestone metrics do. The platform states that these measure the number of people who reach certain points in a video, whether they played the video continuously or skipped to those points. The example it gives is video plays at 25%, and it says explicitly that the metric may include people who skip to the 25% point and play from there.
Why that breaks the ratio. Dividing a milestone metric by a duration metric produces a number whose numerator counts skippers and whose denominator does not. The result is not a retention rate. It is a mixture.
And the other definition has a subtler version of the same problem. 15-second plays divided by 3-second plays stays inside the duration family, so the replay handling is consistent. But it compares two thresholds on a distribution whose shape you cannot see, and it tells you nothing about the 4 to 14 second range where most departures actually happen.
What the platform recommends instead, for anyone who wants duration including replays: a separate average play time metric, which is a different number again.
One family excludes replays and counts unique seconds. The other counts anyone who reaches a point, including people who skipped there. Source : Platform metric documentation, read directly (2026)
The shorter metric is the stricter one
This is the finding that surprises everyone, including people who buy video every day.
The 3-second play. The platform counts the number of times a video played for three seconds, or for 97% of its length if shorter. Replays are excluded. It gives the example that someone who plays a 3-second video for two seconds, rewinds and watches the first two seconds again does not register a view, because replays do not count.
What that definition does not contain. Any requirement about pixels on screen. Any requirement about sound. Any accreditation.
The 2-second continuous play. The platform states that most 2-second continuous plays will have at least 50% of the video pixels in view, notes the exception where a system overlay obscures part of the screen, and states that the metric is accredited by the measurement council.
So the ranking is inverted. The two-second metric is pixel-gated and accredited. The three-second metric is neither. A longer threshold does not mean a more demanding measurement, and anyone building a funnel on 3-second plays is building on the looser number.
ThruPlay, for completeness. It optimises and bills for completion on videos shorter than 15 seconds, and for at least 15 seconds on longer ones. The platform defines full length as at least 97% of the video, because people drop off as credits roll or content fades.
The practical implication. If you want the most defensible retention denominator on that platform, it is the 2-second continuous play, not the 3-second play that every dashboard defaults to.
A view is five different events
Across platforms the word is doing entirely different work, and nobody reconciles it.
Two seconds, pixel-gated. The accredited metric on the largest social platform.
Three seconds, not pixel-gated. The default on the same platform.
Six seconds, or an engagement. On the short-video platform, a 6-second view counts a video that played at least six seconds in an impression session, played in full if shorter than six seconds, or that received at least one engagement within the first six seconds. That last clause means it is not purely a duration metric: two ads with identical retention can report different 6-second view counts if one attracted more taps.
Fifteen seconds, or 97%. ThruPlay.
Thirty seconds, or an interaction. On the video platform, you pay when a viewer watches 30 seconds, or the entire duration if shorter, or interacts with your video, whichever comes first.
And a replay asymmetry worth knowing. On the short-video platform, video views exclude replays, but average play time per video view includes time spent replaying. Multiplying one by the other does not give you total watch time.
None of the five mentions sound. Whether anyone heard the ad is not part of any platform’s definition of a view.
There is one measurement in this area that a standards body has genuinely defined, and it is worth holding onto.
The viewable video ad impression. To count, two continuous seconds of the video advertisement must play, meeting the same pixel requirement as a viewable display ad.
The clause that matters for creative. The guidelines state that this required time is not necessarily the first two seconds of the video ad: any unduplicated content comprising two continuous seconds qualifies.
The other defined term. Video ad completion rate: the percentage of all video ads that play through their entire duration to completion. Simple, unambiguous, and defined in the glossary that has no entry for hold rate.
Why this matters to your reporting. These two are the only terms in the area you can put in a client deck and defend if challenged. Everything else is platform-specific or invented.
What follows about hold rate specifically. It is not that the concept is worthless. Retention past a point is a real thing worth knowing. It is that the word has no fixed referent, so two people using it are usually not discussing the same number.
Read the curve, not the ratio
A single number cannot describe retention, and the four milestone points can.
What a ratio hides. Two ads with the same 50% milestone can have completely different shapes: one loses viewers steadily, the other loses almost everyone in the first two seconds and then holds the survivors to the end. Those are different problems and they need different fixes.
The steep early drop. Most departures between the start and 25%. The opening is not earning the next five seconds. Change the first frames, not the ending.
The mid-video collapse. Retention fine to 25%, falling apart before 50%. The ad made a promise it did not keep quickly enough. Usually the message arrives too late.
The steady decline. An even slope across all four points. This is normal and mostly a length problem: the ad is longer than its content justifies.
The flat tail. Heavy loss early, then almost none. You have found your audience within the audience. The targeting is wide and the creative is doing the qualifying, which is expensive but informative.
What to do with the shape. Fix the segment where the slope is steepest, then re-measure. That is a real diagnostic loop, and it is available on every platform without inventing a metric.
Five rules that survive the fact that the vocabulary is broken.
Pick one platform’s metrics and stay inside them. Never build a ratio from two platforms, and never build one across the duration and milestone families within a platform.
Use the accredited denominator where one exists. On the largest social platform, that is the 2-second continuous play, not the 3-second play your dashboard shows by default.
Define your own hold point in seconds, not in percentages. Identify the timestamp where your message actually lands. A 25% milestone on a 12-second ad and on a 90-second ad are not comparable events.
Read the whole curve, not one ratio. The milestone metrics give you 25, 50, 75 and 100%. Four points describe a shape. One ratio describes nothing, and hides where people actually left.
Judge on cost per meeting held. Retention is a diagnostic, not an objective. If better retention does not move the downstream number, it improved a metric and nothing else.
And say what you mean in the deck. Write “share of viewers reaching 50%, milestone metric, includes viewers who skipped to that point” rather than “hold rate”. It is longer, it is defensible, and it prevents the argument three months later about which definition was used.
“Hold rate” appears zero times in the industry video glossary, the video impression guidelines and the viewability guidelines. No platform defines it.
Two definitions circulate and they disagree: 15-second plays over 3-second plays, or share past a percentage milestone. They cross over at different video lengths.
The milestone version is arithmetically unsound. Milestone metrics count viewers who skipped to a point; duration metrics exclude replays and count unique seconds. The ratio mixes families.
The shorter metric is the stricter one. The 2-second continuous play is pixel-gated and accredited. The 3-second play, which every dashboard defaults to, is neither.
A view is five different events: 2 seconds pixel-gated, 3 seconds not, 6 seconds or one engagement, 15 seconds or 97%, 30 seconds or any interaction.
None of them involves sound. Whether anyone heard your ad is not part of any platform’s definition of a view.
Only two terms here are actually standardised: the viewable video impression at two continuous seconds, and video ad completion rate.
And the viewability standard refuses to privilege the opening: the required two seconds need not be the first two.
No. The term appears zero times in the industry video glossary, zero times in the video impression measurement guidelines and zero times in the viewability guidelines, and no platform publishes a definition of it.
What is the usual definition?
There are two, and they disagree. One divides 15-second plays by 3-second plays. The other measures the share of viewers passing a milestone such as 25%, 50% or 75%. Neither is authoritative and both have arithmetic problems.
Why can't you divide those metrics?
Because they count different things. Duration metrics sum unique seconds of play and exclude replays. Milestone metrics count anyone who reaches a point in the video, including people who skipped straight to it. A ratio across the two mixes methodologies.
Which video metric is the most trustworthy?
On the largest social platform, the 2-second continuous play. It is accredited by the measurement council and most such plays have at least 50% of the video pixels in view. The 3-second metric has neither property.
So a longer view is not a stricter view?
Not necessarily. On that platform the 2-second metric is pixel-gated and accredited while the 3-second one is not, so the shorter threshold is the more demanding measurement.
What counts as a view across platforms?
Two seconds with a pixel requirement, three seconds without one, six seconds or a single engagement within six seconds, fifteen seconds or 97% of the video, or thirty seconds or any interaction. They are five different events sharing one word.
Do any of these definitions involve sound?
None of them. Every platform view definition is about duration, and in one case pixels. Whether anyone heard the ad is not part of any of them.
What should I use instead?
The share of viewers still present at the timestamp where your message lands, measured with one platform's own milestone metrics and never as a ratio across two different metric families. Then judge on cost per meeting held.