Guide
Hooks on X: seven kinds of opening line, and how to test yours
The hook is the part of a post people read before deciding to read the rest. There is no hook that works for everyone, but there is a way to find the kinds that work for you. This guide gives the kinds Crux recognises, the exact rules, and the test.
Updated
Crux sorts the opening line of every post into one of 7 kinds: list, announcement, contrarian claim, question, number or statistic, first-person and plain statement. To learn which kinds work on your X account, compare each post with the 20 posts you published nearest to it in time, take the median result for each kind, and trust a kind only when it holds at least 5 posts and beats what random groups of your own posts produce. Averages from other accounts will not tell you: a kind that works on one account can do worse than usual on another.
What counts as the hook
The hook is the first line with words in it, and inside that line, mostly its opening sentence. A figure in the fourth sentence never had the chance to stop anyone scrolling, so it does not make a number hook. A few things are skipped before the reading starts:
- lines that are only a link or only a handle;
- the handles at the start of a reply;
- thread numbering such as โ1/โ or โ(1/7)โ;
- a short lead-in such as โWow.โ or โBig news.โ, which is read together with the sentence after it.
The seven kinds
The kinds overlap, so they are checked in a fixed order and the first match wins. That order puts the more specific reading first. The example lines are made up for this guide, and each one is classified the way shown by the same code that reads real accounts.
List
The first line starts with a bullet or an enumerator, such as โ-โ, โโขโ, โ1.โ or โ2)โ. Later on, a post whose body has three or more such lines counts as a list too, if nothing above matched first.
- Ship the smallest version first
- Watch what people do
- Fix the first thing they trip onAnnouncement
The opening sentence launches something: launching, just launched, shipped, introducing, announcing, just released, now available, out now, weโre live, now live. Intent beats form, so an announcement with a number in it is still an announcement.
Introducing our public changelog: 3 things we fixed this week
๐ Launching our pricing calculator todayContrarian
The opening sentence makes a claim about the crowd (everyone, nobody, most people, unpopular opinion, hot take, conventional wisdom, overrated, underrated, wrong), or leads with stop, donโt, never, quit, forget or actually. A contrarian claim phrased as a question still counts here.
Unpopular opinion: onboarding emails are overrated
Stop adding features before anyone asks for them
Why does everyone still send a welcome email?Question
The opening ends with a question mark, or starts with a question word (why, how, what, which, do, does, is, can, shouldโฆ). A question word followed by I, we, you, my, our or to is a title, not a question: โHow Iโฆโ and โWhat we learnedโ are not asks.
Why do onboarding emails get ignored?
1/ Which metric do you check first on a MondayNumber or statistic
The opening sentence holds a figure: a digit, a percentage, a currency sign, or a written number (one, ten, hundred, million, dozen, half). Thread numbering such as โ1/โ or โ(1/7)โ does not count, but a real fraction such as โ1/2โ does, and a figure buried in the fourth sentence does not.
40% of our churn came from the first week
1/2 of our signups never open the app
Wow. Our signup form now takes 20 secondsFirst-person
The opening starts with I, my, me, we, our or us, or is a first-person title such as โHow I rewroteโฆโ.
I stopped sending a welcome email
How I rewrote our pricing page in a dayPlain statement
Everything else: a claim or observation with none of the cues above.
Pricing pages should answer the buyerโs question firstWhich kinds work for you
Counting likes by hook kind will mislead you on a growing account: whichever kind you used most recently will look best, because more people followed you by then. Crux measures each post against the median of the 20 posts nearest it in time instead, so every post is judged against what your account usually got at that moment.
Then it takes each kindโs median multiplier, not its average, so one post that went far does not carry the rest. A kind needs at least 5 posts to be judged at all. Finally it draws the same number of your posts at random 500 times: a kind whose median lands above the 95th percentile of those draws works for you; one below the 5th does worse than usual; anything in between is not reported. A worked example shows the test on a sample account.
Crux publishes no averages across accounts, and there is no percentage here for how well questions or numbers do in general. That is on purpose: such a figure describes the average account, and the point is to learn about yours.
Habits that make the test possible
- Use more than one kind. A kind you used only 4 times cannot be judged; one you used forty times can.
- Put the hook in the first line. Crux reads only the first line, and mostly its opening sentence, so a question at the end of a post does not make a question hook.
- Judge a kind by its median, not its best post. One viral question says little about your questions.
- Keep your weaker posts. Deleting them removes the evidence of what does worse than usual.
Related: how to download your X archive and the full method.