A hook has exactly one job: to give the viewer a reason not to swipe on within the first second or two. It does that job when it names a concrete problem this viewer actually has right now, not when it follows a formula. Why that is not a matter of taste, Instagram says itself: for reels, the most important predictions the system makes are how likely you are to reshare a reel, watch it all the way through, like it and go to the audio page. The hook buys you the seconds those first two can happen in at all. Three rules carry almost everything: concrete over clever, problem over topic, readable with the sound off. Concrete means a number, a timeframe, a place or a sentence somebody actually said. This is exactly where collected hook lists fall apart: they hand you the container, but the content has to come out of your own week, and without it the same template sounds identical on a thousand channels.
I film my own reels, but I no longer write the hooks by hand. What I collect about AI at work sits on the AI page; the full reel workflow at my place is in Can AI Create Instagram Reels?, and how I decide what to hand off at all is in Which Tasks You Can Hand Off to AI. This piece is only about the first second and a half, because everything else hangs off it.
What a hook has to do
A hook does three things at once.
It stops the thumb. For that, the first line has to be readable without sound. I plan every on-screen line as if the sound were off, because I do not know in advance who has it on. Five to eight words, one line, no term that needs explaining first: that is my working rule from production, not a spec from Instagram.
It filters. A good hook drives the wrong viewers away on purpose. If you write for accountants, "For everyone who sends invoices" is weaker than "If you handle clients on electronic invoicing", even though it speaks to fewer people. Fewer but the right ones is the better trade on a channel where people are supposed to hire you.
It opens a tension the rest of the reel resolves. This is where clickbait gets expensive: a hook that promises more than the reel delivers wins you two seconds and then costs you the signals the system looks at, watching all the way through and resharing.
On length: Instagram lets you record up to 20 minutes, but does not recommend videos over three minutes to accounts that do not follow you yet. So your hook pays into 30 to 60 seconds of content, not into a lecture.
Why generic hook lists fail
Take the most common template there is: "The three biggest mistakes in X". The template is not the problem. What somebody puts in for X is. Every example sentence in this article is a constructed example, not a case from my own business: they show the form, not my week.
Left and right use the same structure. On the right, somebody named a problem that a specific person recognises at a specific point in their day. On the left there is a category, and a category cannot stop a scroll, because nobody has a category-shaped problem.
So a hook list does not replace the work, it speeds it up. Anyone with raw material from their own week gets a long way with three patterns; anyone without it will not find a hook that carries in a list of two hundred templates either.
Five patterns that work, and why
1. The counted incident
A number and a timeframe from your own work. It works because nobody else can have that number: it is the part of your content no template supplies. Which is exactly why it has to hold up. Write down the source and the period before you publish it.
- Weak: "A lot of clients keep asking me this."
- Strong: "Third website this week where the contact form went nowhere."
2. The viewer's symptom
Describe the state they recognise in themselves first, and the cause second. It works because recognition is faster than understanding.
- Weak: "Today we are talking about newsletters."
- Strong: "If your newsletter gets opened but never clicked, it is rarely the subject line."
3. The correction
Contradict an assumption your audience really holds. It works because a contradiction opens a question the viewer wants closed. Important: it has to be a genuine, widespread assumption, not a straw man.
- Weak: "The biggest myths about taxes."
- Strong: "The home office deduction is not where most self-employed people leave money on the table."
4. The look behind the door
Show what you see as a professional and the client does not. It works because you are offering access rather than knowledge anybody can look up.
- Weak: "A look behind the scenes of my work."
- Strong: "What a roofer sees when he is the second one up on the same roof."
5. The real question
Quote, word for word, something somebody asked you, and answer it. It works because a real question sounds different from a constructed one, and the viewers asking it too feel addressed immediately.
- Weak: "Frequently asked questions from my clients."
- Strong: "'Can I still accept the proposal next week?' If someone writes you that, your proposal was unclear."
Five is enough. Collect twenty and you end up picking patterns while writing instead of naming problems, which puts you right back at the list.
How to apply the patterns to your own topic
Step 1: spend two weeks collecting incidents, not topics. One line per incident, with a date and what somebody said verbatim. "Sept 14, client asks whether she can still correct an invoice the customer has already paid." That is the entire raw material, and after two weeks it is surprisingly large, large enough for a content plan covering four weeks. Keeping it up is the hard part. If you would rather not do that on your own, put your first ten lines up for discussion in the community: the people there have the same two weeks behind them.
Step 2: write the problem behind the incident in one sentence. Three questions are enough: who does it hit, how do they notice it, and what does it cost them? From the example above: "Self-employed people without a bookkeeping routine only notice at year-end that a paid invoice cannot simply be changed, and they pay for it in stress and accountant hours."
Step 3: pour that same sentence into three different patterns. Not three hooks on three topics, but three hooks on one problem. From the sentence above:
- Counted incident: "Twice this week, the same question: can I still change an invoice that has been paid?"
- Symptom: "If your accountant calls in March asking about an invoice from September, something went wrong."
- Correction: "You do not correct a paid invoice. You do something else, and it takes ten minutes."
Step 4: test every candidate against two hard criteria. First: does it stand on the first frame without sound, and can it be read in one second? Second: does the last sentence of the reel resolve it? Whatever passes both goes into the shortlist, the rest is out.
Step 5: let viewers pick, not you. Instagram has trial reels for this. You mark a reel as a trial before sharing, it is shown first to people who do not follow you, and your followers do not see it. Roughly 24 hours later you get views, likes, comments and shares plus a comparison with your earlier trial reels. If it performs, you share it publicly afterwards. That lets you run two hooks on the same content without filling your grid with experiments. Read the result as a direction check, though, not as a controlled A/B test: two separately published reels do not reach the same audience at the same time, and it only carries weight once several runs point the same way. Instagram's originality guidelines target other people's material rather than two versions of your own, so they are not in the way; still, shoot the second variant fresh instead of uploading the same file twice.
The usual mistakes
One addition to the second-to-last point: under its own originality guidelines, Instagram describes as original what you shot yourself, what reflects your own perspective or what you have meaningfully edited, and it states that accounts posting mostly unoriginal content are no longer eligible for recommendations to people who do not follow them. Adopting a hook structure is ordinary craft, reposting other people's reels with the audio swapped is something else.
Who writes the hooks at my place
An AI employee drafts them. Conny takes a content brief, a reel, a hook, a script, a carousel or a caption, and delivers a ready-to-approve draft in my voice, checked against the audience, the content pillars and the house rulebooks. A hook is therefore never just a nice line with her: it fits a specific person, one of my pillars and the things I do not say. Her suggestions do not come from gut feel either: behind the rulebooks she works from sit 15,536 analysed Instagram hooks, and she checks patterns against real numbers rather than opinions. In her own review, reels on my account reached a median of 485 accounts and carousels 57; carousels then repeatedly fell below 200 reach with zero saves. What followed was not more polishing of the old format but a cut: production shifted to reels, and the carousel came back only in a reworked form.
Conny invents no personal facts about me. No anecdote, no number, no client detail. When in doubt the spot in the draft stays blank and I fill it. That is precisely why the hooks sound real: a model allowed to invent the incident delivers the generic hook list this article opened with in new packaging, while one that requires the incident from me can sharpen it but not replace it. Nothing gets posted without my approval anyway.
How to set up a role like that with a rulebook rather than a prompt is in How to Hire an AI Employee, the craft behind it in CLAUDE.md: Claude Code's Memory, and who does what at my place in My AI Workforce.
Frequently asked questions
What is a hook in a reel?
The hook is the opening: the first one to two seconds of image, on-screen text and first spoken sentence. Its job is to supply a reason to stay. It does not decide how good the reel is, but it does decide whether anyone finds out.
How long should a hook be?
Short enough to read in one second. Five to eight words in a single on-screen line is the frame I work with, and one spoken sentence: a working rule from production, not a spec from Instagram. Anything that needs a second line belongs at second three, not at second zero.
What are the best hooks for reels?
The ones that name a problem your audience has that day. The pattern is interchangeable, the content is not: the five above are containers, and what they carry comes out of your week. Which is why the question cannot be answered with a list.
How do I test whether a hook works?
With trial reels, and with patience. A trial reel goes first to people who do not follow you, and after roughly 24 hours you get views, likes, comments and shares plus a comparison with earlier trials. Read that as a direction check, not as a controlled A/B test: two variants never run under identical conditions, so only trust a result once several runs say the same thing.
Can an AI write hooks?
Yes, and this is one of its strongest spots. It holds the audience, the pillars and the exclusion rules in view on every draft, proposes variants rather than one version, and checks them against the numbers from your last posts. What it cannot do is live through the incident. Give it three lines from your week and you get five variants.
What to do next
Open a file today, call it incidents, and for two weeks write one line into it every evening: the date, and who asked what or got what wrong. Then take the first entry, put the problem behind it into one sentence, and build three hooks from it in three patterns. Film exactly one, as a trial reel, so the decision sits with people who do not know you yet. The step after that is a rulebook of your own instead of a pile of prompts, and that is where most people get stuck alone. So bring your three hooks into the community: we go through them on the calls, and in between, people keeping the same file will tell you in a post which one carries. How I set the rulebook up, and what Conny's template looks like, is there too.
