// the hook survival taxonomy
Every ad hook has a shelf life.
We tagged the opening hook of 4,000+ ads running right now behind ~100 growing companies, then measured how long each ad has stayed live. The kind of hook predicts the shelf life. A bold claim runs a median of 70 days and the tail reaches years. A pattern interrupt runs a median of 28 days and rarely sees a birthday. An ad that keeps running is an ad that keeps paying its way, so the hooks that survive longest are the ones the market keeps rewarding.
How long does each kind of hook survive?
Each line is the share of currently-running ads of that hook type that have stayed live for at least that many days. The gap is real: bold claims and social proof sit highest, pattern interrupts drop away first. Switch the slice to see the same ranking hold inside SaaS and DTC. Hover or tap to read every type at any age.
The five hooks, ranked by how long they last
Ordered most durable to least. Each card is a real, currently-running example, quoted from the public ad libraries and drawn from the longest-running ads of its type. The copy is the hook: the first line the scroll meets.
States a big, specific promise up front.
Leads with who already trusts it.
Opens a loop the viewer has to close.
Names the pain before the fix.
Breaks the scroll with the unexpected.
The rest of the growth data
Method: every ad counted is live right now (still marked active) in the public Meta or Google ad libraries, bound to a company by verified page or domain, never by name. A model reads each ad and tags its opening hook into one archetype; an ad's age is its run time to date, so “surviving N days” is the share of currently-live ads of that type that have already run at least that long. Only splits with at least 30 ads are shown. A residual “other” bucket (296 ads of un-hooked evergreen catalog inventory) is held out of the ranking. Figures round and update automatically. Built from the same ad libraries and traffic data behind our daily teardowns.
Run a free growth audit on your own domain →