Thumbstop rate is meant to tell you whether your creative stopped somebody mid-scroll, and at least three different calculations circulate under that single name. Nobody quoting a figure says which one they used, because most people do not know there is more than one.
That makes any comparison between two thumbstop rates a comparison of unknown quantities, which is worse than having no metric at all.
This is for you if somebody has quoted you a stop rate target and you want to know what was actually measured.
What thumbstop rate is trying to capture
The idea behind it is sound and worth separating from the execution.
In a feed, the first job of any advert is to interrupt scrolling. Everything else depends on that, so a metric isolating the interruption from what follows would be genuinely useful.
The intent is to measure that moment specifically, separately from whether the rest of the creative held anybody or converted anybody.
Where it falls apart is that Meta does not publish a metric by this name. It is an industry coinage, assembled from platform numbers by whoever needed it, and different people assembled it differently.
The three calculations sharing one name
Each of these appears in published material, presented as the definition.
Three second video views divided by impressions. The most common version, and it counts every time the advert was shown as an opportunity to stop somebody.
Three second video views divided by reach. Counts people rather than showings, so somebody who saw the advert four times and stopped once is treated differently.
A different view threshold entirely. Some versions use two seconds, some use a platform-specific view definition, and some use the video play count rather than a timed threshold.
Those produce materially different numbers from identical creative. Nobody publishing a thumbstop rate figure states which they used, and the fact that three exist rarely comes up.
Why the denominator makes them diverge
This is not a rounding difference. The versions move apart over the life of a campaign.
Reach counts people. Impressions count times shown. Early in a campaign they are close, because most people have seen the advert once.
As frequency rises, impressions grow while reach flattens. So the impressions version falls automatically, while the reach version stays roughly where it was.
Read the impressions version alone and you conclude the creative is decaying. Read the reach version and you conclude nothing changed. Both are describing the same adverts.
That is why frequency belongs beside any rate metric, and why comparing your thumbstop rate against somebody else’s is comparing two campaigns at unknown frequency using unknown formulas.
Why no thumbstop rate benchmark can exist
Even setting aside the sourcing problem, a benchmark here is structurally impossible.
A benchmark requires a shared definition. This metric has three, so a single published figure would be wrong for at least two thirds of the people reading it, regardless of how carefully it was measured.
Then add the sourcing problem on top. Every figure circulating comes from the same layer of vendor content citing other vendor content, with no sample size and no method.
The research behind this cluster catalogued that pattern across sixteen sub-topic searches. On thumbstop rate specifically, the entire visible first page was AI-generated vendor content, with precise-looking numbers and no traceable origin.
So you have an undefined quantity with an unsourced benchmark. Comparing your number against it means nothing at all, however carefully you calculate your side.
What the metric can legitimately do
Relative comparison inside your own account, with your definition written down.
Pick one calculation. Write it down somewhere your successor will find it. Then compare creatives against each other on the same placement, the same period and similar frequency.
Used that way it is a perfectly reasonable diagnostic. You are asking which of your adverts interrupts scrolling more effectively, and that question has an answer.
What it cannot do is tell you whether you are doing well. Absolute performance requires a reference point, and there is not one.
That distinction is the whole practical content of this article. A metric can rank your options without being able to grade them.
Why I use retention instead
Same reasoning as everywhere else in this cluster: fewer numbers, clearer decisions.
I track whether somebody watched more than half the video and stayed past three seconds. If they did, the creative stopped them, so the stopping is already measured.
Adding a separate stop metric gives me another row in the report and no decision I would make differently. It also gives me a number with three definitions, which is a liability rather than an asset.
There is one more consideration. A high stop rate with poor retention describes an opening that interrupted people and then lost them. Optimising towards stopping specifically rewards that, and I would rather not.
How to handle a thumbstop rate figure you are shown
Two questions, asked plainly, and they resolve most of it.
Which calculation is that? If the person cannot say, the number cannot be compared to anything, including your own.
Where did the benchmark come from? Follow it once. It will terminate in an unsourced article within two or three steps, and doing that exercise yourself is more convincing than being told.
Neither question is confrontational. Both are the sort of thing a careful person asks, and both save you from adopting a target that was invented.
Apply the same two questions to every creative metric you are given. Several will not survive them, and knowing which is a real advantage.
What the pattern behind thumbstop rate tells you
This metric is a clean example of something worth recognising generally.
Meta has data on scroll behaviour and publishes no metric by this name. The industry needed one, assembled it from available numbers, and ended up with several incompatible versions carrying one label.
The same shape appears wherever a platform holds data it will not publish and a query has commercial demand behind it. Frequency thresholds are the clearest parallel: Facebook analysed 2,439 Brand Lift campaigns, confirmed a plateau after a handful of impressions, and deliberately published no number. The internet publishes several.
So the transferable lesson is not about this metric. When the party with the data stays silent and everybody else is specific, the specificity was manufactured.
Once you recognise that condition you stop needing to check every individual figure, because you can predict which ones will not survive checking.
What to change this week
Three steps.
If you track a stop metric, write down which calculation you use and put it next to the number in your reporting, so nobody compares it against a different one.
Then put frequency beside it, so you can distinguish real decline from a growing denominator.
Then drop any external target for it, and compare your creatives against each other on matched placement and period instead.
Hook rate has no published benchmark covers the closely related metric and the same sourcing problem, hold rate tells you more is the measure I use in place of both, ad creative metrics covers what every denominator actually does, and fund the ad set and read what the algorithm chose is the position underneath. The case studies show results reported on a stated basis. To have your creative reporting checked for numbers nobody can source, book a teardown.