Hook rate benchmarks: the only one that applies is your own

Write down your definition before you open anything. Hook rate is three second views divided by impressions, as a percentage.

Pull your last twenty creatives that got meaningful impressions. Fewer than twenty and the median moves every time you add one.

Compute the figure for each one under that single definition. No exceptions for placements you dislike.

Sort them and take the middle value. That is your benchmark. It is the only one that describes your account.

Write the date next to it. It will drift as your placement mix and audience temperature change.

Recompute quarterly. A benchmark from last spring is describing an account you no longer run.

Hook rate moves with factors that have nothing to do with your creative.

Placement mix moves it, because feeds that autoplay behave differently from feeds that do not.

Audience temperature moves it, because people who know your brand stop scrolling more readily. Category moves it. Even the time of year moves it.

Definitions move it most of all. Some people divide by reach rather than impressions. Some use video plays, which inflates the figure substantially.

So an article quoting thirty percent is describing somebody else's placement mix under somebody else's denominator.

A benchmark you cannot reproduce is a rumour with a decimal point.

The same applies inside your own company. Two teams quoting hook rates are frequently measuring different quantities and comparing them anyway.

Agree the denominator once, in writing. Half the disagreements about creative performance disappear before anybody opens a spreadsheet.

Comparing two hooks needs enough impressions per arm to distinguish them. Most people declare a winner well before that point.

The approximation for comparing two proportions is 16 times p times one minus p, divided by the difference squared.

Here p is your own median as a decimal. The difference is the gap you want to detect, also as a decimal.

The figures below use an illustrative 30 percent baseline. They show the shape of the arithmetic rather than a target.

Substitute your own median before any of these numbers touch a real decision.

The pattern to notice is that the bill rises with the square of how fine a distinction you want.

If a 1 point difference costs 33,000 impressions per arm to detect, subtle rewrites are not testable at most budgets.

So test openings that are genuinely different. A number against a question. A result against a problem. A face against a product in hand.

Those produce gaps large enough to read on a normal week's spend.

Making several genuinely different openings is cheap here. One three minute take can hold four of them, and each batches separately at 100 credits.

Highlight a phrase and a clip lands over exactly those words, so the visual opening changes without refilming.

That is the link between the arithmetic and the production. The test design demands variety. This is where variety stops being expensive.

A high hook rate with a low conversion rate is often a mismatch rather than a success. Something in the first two seconds attracted the wrong people.

Hook rate does not measure whether anybody bought. It measures whether anybody stopped, which is the first of several hurdles.

Optimising it alone reliably produces ads that get watched and sell nothing. That costs more than a low hook rate does.

It also cannot be compared across accounts, across campaigns with different placement mixes, or before and after a targeting change.

If you take one thing from this page, take the definition and the date. Everything else follows from having those two written down.

If your median is below where you want it, that is a first-four-seconds problem rather than a benchmark problem.

Fixing it is a writing job. Put a number, a name or a result in the opening. Emphasise one word so it pops in the captions.

Then run the new opening against the old one with enough impressions per arm to read the gap.

Recompute the median each quarter and keep the old ones. A benchmark that has moved twice tells you more than any single figure ever did.

Keep them in the same place as the definition and the dates. Three numbers and three dates is the whole record.