What makes a hook work?
It sounds like a person talking, not a brand announcing.
In September we sorted the posts on my Spanish and French endometriosis accounts by hook shape and compared like rates.
The listicle, the shape most of us reach for first, came last.
| Hook shape | What it sounds like | Like rate |
|---|---|---|
| Quoted dismissal | «no te puede doler tanto, es solo la regla» (“it can’t hurt that much, it’s just your period”) | 1.69% |
| Question | The question your audience is already asking themselves | 1.27% |
| Plain complaint | “odio que el dolor me despierte a las cuatro de la mañana” (“I hate that the pain wakes me at four in the morning”) | 1.21% |
| Listicle | “things nobody tells you before surgery” | 0.95% |
Like rates are averages per shape from the September comparison.
The example lines come from a later test on one of my Spanish accounts. Want hooks to test this week? Get 30 free hook prompts.
The winners have something in common. They’re what a friend would say out loud, or what someone said to you that still stings. Nobody talks in listicles.
Averages hide outliers, though.
One French list hook reached a 7.93% like rate:
“ce dont personne ne te prévient avant une opération de l’endo” (what nobody warns you about before endo surgery).
A shape is a starting bet, not a law.
How long should a hook be?
Short enough to read before the thumb moves. The rule I teach: under 14 words, lowercase, in the words your audience actually types.
Most guides land on five to fifteen words, and the first few carry the weight.
Completeness matters more than length. The hook should be the whole thought in the first line. On a slideshow, slide two is the punchline, not the lesson.
One test hook reads «seguro es psicológico, los análisis salen bien» (“it must be psychological, your tests are fine”).
Slide two: “nueve años después: endometriosis.” Nine years later. Nothing else needed.
Can AI write your hooks?
Yes, if you give it real shapes to copy. This is the method from my social media workshop. Find ten posts that outperform in your niche.
The filter is likes divided by followers: a 300-follower account with 4,000 likes is a signal, a two-million account with 50,000 likes is noise.
Name the hook shapes those posts use, ask AI for twenty hooks in those shapes, then cut to ten by hand. The cutting is the skill.
“Here are 10 posts that outperformed in my niche, with their first lines.
Name the hook shape each one uses.
Then write 20 new hooks in those shapes for [audience], in their words, lowercase, under 14 words.
No hashtags, no emojis, no promises we can’t keep.”
Then let results write the next batch. On my app’s accounts, a weekly job ranks which hooks earned views and feeds the winning shapes back into the hook library.
The machine rewrites its own inputs. Your version can be a Sunday spreadsheet.
When hooks matter less
A hook gets attention. It can’t keep it. If the post doesn’t pay off the first line, people learn to stop trusting your first lines.
In health, money or anything with stakes, a hook that promises a cure is a lie with good reach.
My system refuses them. In one test it turned down a hook that said “te cura”: it cures you.
And for a coach with a dozen clients, the hook matters less than whether the post sounds like you. A booked call comes from trust, not a scroll-stop.
Next: content pillars decide what the hook is about, and posting cadence decides how often you get to test one.