What is Hook rate?

A video ad is served 480,000 times and records 62,400 three-second plays, the column Meta reports. Hook rate is 62,400 ÷ 480,000 = 13%. Thirteen percent of the impressions produced somebody still there at three seconds, and the other 87% were scrolled past before the ad had said anything. Whatever the rest of the video does, it does it to that thirteen percent.

Hook rate divides the number of short-threshold video views by the number of impressions. It answers one narrow question: of the people the ad was served to, how many were still there after the opening.

Hook rate = three-second video views ÷ impressions

Its value is the isolation. Every other video metric mixes the opening with what follows. Completion rate depends on the whole edit; cost per view depends on the auction; conversion rate depends on the page. Hook rate depends on the first seconds and very little else, which makes it the cleanest available read on whether an opening works.

That also bounds what it can tell you. A strong hook rate means people stayed past the opening. It says nothing about whether they stayed after that, understood anything, or did anything.

No platform reports a metric called hook rate. It is assembled from two figures platforms do report, and the assembly is a local decision.

The threshold varies with what the platform exposes: Meta reports three-second video plays, TikTok reports a two-second equivalent, and a team computing the metric across both has to decide which to use or to accept that the two figures are not comparable. The denominator varies: impressions on some teams' definition, reach on others'. Neither choice is wrong and they produce different numbers from the same campaign — the reach-denominated version is substantially higher, because each person is counted once however many times they were served.

The consequence is that a hook rate quoted without its definition is uninterpretable, and that cross-account or cross-agency comparisons of the figure are usually comparing conventions.

The fix is trivial and rarely done: write the definition next to the number, once, in the report template.

The single largest influence on this metric is not the creative. It is where the ad was shown.

Autoplaying in-feed video crosses a short threshold constantly, including from people scrolling past at speed who never decided to watch anything. Surfaces where somebody pressed play, or chose not to skip, produce a much lower rate from a much more deliberate audience.

So a blended hook rate across placements is a weighted average of several unrelated behaviours, with the weighting chosen by the delivery algorithm. It moves when the mix moves, which happens continuously and without anybody deciding.

The practical rule: split by placement before drawing any creative conclusion. A creative test that looks decisive is frequently a delivery-mix difference, and this metric is more susceptible to it than most.

On its own the figure can be high for the wrong reason. Cheap autoplay inventory produces impressive hook rates and very little attention.

Read against hold rate — the share of those early viewers still present at a longer threshold — the pair becomes diagnostic.

High hook, high hold. The opening earned attention and the video repaid it. Scale it and watch for fatigue.

High hook, low hold. The opening stopped people and the rest did not hold them. Either the hook promised something the body did not deliver, or the body is longer than the idea in it.

Low hook, high hold. The opening is failing and what follows works. This is the most actionable pattern in video creative: a better opening on the same body is a cheap edit with a large effect.

Low hook, low hold. Usually an inventory or audience problem before it is a creative one, particularly if the pattern holds across several different creatives.

The opening, and almost nothing else.

Open on the situation rather than the solution, because people recognise their own problem faster than a product. Put the subject in the first frame, which is also the thumbnail. Start mid-motion, because a shot already in progress reads as content rather than as an advertisement beginning. Write for silence, on screen as well as spoken. Move the branding after the point rather than before it.

Each of these is a change to a second or two of footage, which is what makes hook work the cheapest high-leverage editing in advertising: the same body, several openings, tested against each other.

Hook rate is a creative diagnostic and never a target. Optimising toward it directly buys cheap autoplay impressions and calls them attention.

Its place is in the creative review, beside hold rate and the retention curve, answering a question the outcome metrics cannot: whether the ad is failing because the offer is wrong or because nobody got far enough to hear it.

Those are different problems with different fixes, and an account that cannot tell them apart will keep rewriting offers that nobody ever saw.

Teams that compute the figure over reach rather than impressions get a much larger number from the same campaign, and both camps defend their choice.

The case for impressions is that it is the unit the money bought, it keeps the metric consistent with every other rate in a media report, and it does not flatter.

The case for reach is that it describes people rather than deliveries, and that a person served the same ad five times had one opportunity to be hooked rather than five.

The second argument is reasonable and it has a practical problem: frequency varies by campaign and over time, so a reach-denominated hook rate moves when frequency moves even if nothing about the opening changed. That makes it a worse instrument for comparing creatives, which is what the metric is for.

The defensible resolution is to use impressions for creative comparison, to state the choice, and to stop renegotiating it.

Hook rate, hold rate and thumbstop rate all describe the same early moments and overlap enough to cause confusion.

Hook rate and thumbstop rate are frequently the same arithmetic under two names — short-threshold views over impressions — with thumbstop carrying an implication about scrolled feeds specifically.

Hold rate is the one that is genuinely different, because its denominator is the early viewers rather than the impressions. It measures what happened after the hook rather than at it.

An account using all three should write down which is which, because the failure mode is not disagreement about the numbers but three people reporting the same number under different names and concluding they disagree.

• Rebuild the first second before anything else. Open on the situation, put the subject in frame, and start mid-motion rather than easing in.
• Move the branding after the point. An opening that establishes who is speaking before saying anything worth hearing loses the moment it needed.
• Write the opening for silence, on screen as well as spoken, because most feed views start with sound off.
• Test openings against a fixed body so the comparison isolates the hook rather than the whole ad.
• Split the figure by placement before concluding anything, because surfaces differ more than hooks do and the delivery mix moves on its own.