Building a hooks library with AI is a Skill job you finish in one sitting, about forty-five minutes for the first batch. You give ChatGPT or Claude your services and the words your customers use, it writes fifty first lines for social posts, and you keep the ones that sound like you, then rank them as your posts come back. This page covers how to build and rank your own list. The data on which hook types work for small businesses lives on our hook benchmark of 648 real posts, and this page uses it only as a starting order. The AI writes the fifty. You decide which ones stay, and your own posts decide which ones rise.
- Shape
- Skill
- Sitting
- one sitting
The shape: Skill
A Skill job is one prompt, one answer, and your judgment on top. There is no system to build, no tool to connect, and nothing that runs on its own.
A hooks library fits this shape because the hard part of a first line is range, not effort. Most owners open every post the same way, often with the name of the service or a greeting, because that is the first thing that comes to mind on a busy day. Fifty lines side by side show you openings you would not have reached alone: the customer's problem in their words, a comparison, a plain how-to, a question you actually want answered. Seeing them together also makes the weak ones obvious, so cutting is fast.
The other half of the shape is that the AI cannot tell you which line your followers will respond to. It has never seen your audience, and no prompt changes that. So the job splits in two: the AI writes the batch in one sitting, and your published posts do the ranking over the next few weeks. The library you keep is the part that lasts.
What this means for you: put your effort into the inputs and the cut. A vague prompt returns fifty vague lines, and ranking fifty vague lines teaches you nothing.
What you give the AI
The AI writes a first line as a reason to keep reading, so it needs to know what is worth reading about. Give it these, by name, in one message:
- Your three to five main services or products, one line each, in the words on your own website.
- The problems customers bring you, in their words. Pull these from reviews, emails, intake forms or what people say on the phone: "the door sticks every summer" beats "door repair".
- The questions you answer over and over, the ones you could recite in your sleep.
- Who your customer is, in one sentence, such as "homeowners in older neighborhoods who want one person they trust".
- Five of your own past captions, pasted as text, with a note on which ones got comments, so the AI hears your voice.
- Words you never use, such as "game-changer", "secret" or anything your industry rules forbid.
- The platforms you post on, since a first line on LinkedIn and one on Instagram can carry a different tone.
If you have an honest opinion that sets you apart from others in your trade, add it in one line. The AI will otherwise invent one, and an invented opinion is the fastest way to sound like nobody.
What comes back
You get fifty first lines, grouped by hook type, each one short enough to read before a phone cuts the caption off. Each line carries a label for its type and a note on which service or problem it belongs to. You also get a starting order for the types, based on the benchmark, so your first posts test the likelier ones first.
Paste this brief into ChatGPT or Claude, with your inputs below it:
Write fifty first lines for my social media posts. A first line is the opening sentence of a caption, the part people see before they decide to keep reading.
Group them by hook type, spread evenly so the total is fifty:
- Problem: names a problem my customer has, in their words.
- Comparison: sets two choices side by side (this or that, before and after, old way and new way).
- How-to: promises a short, useful step.
- Question: a question I genuinely want my followers to answer.
- Plain: no formula, just a specific, true statement about my work.
- Contrarian or myth: only if it matches an opinion I gave below. If I gave none, skip this type.
Rules:
- Every line is about my business, using the services, problems and questions below. Nothing generic that any business could post.
- No made-up numbers, results, prices, reviews or customer stories.
- No exclamation marks, no emoji, no capital letters for emphasis.
- Never use the words on my "never use" list.
- Each line is one sentence, short enough to show in full before a phone cuts the caption off, and reads well on its own.
- Match the voice of my past captions. Do not repeat any of them.
- Next to each line, note which service or problem it belongs to.
After the fifty, list the hook types in this starting order for me to test: problem, comparison, how-to, plain, question, contrarian or myth.
Then name the five lines you think are strongest and say in one sentence why each one is specific to my business.
My services:
[list]
Problems customers bring me, in their words:
[list]
Questions I answer over and over:
[list]
Who my customer is:
[one sentence]
An opinion I hold that others in my trade may not:
[one line, or "none"]
Platforms I post on:
[list]
Words I never use:
[list]
Five past captions, and which got comments:
[paste]The starting order in that brief comes from our benchmark. Among standard text-and-image posts in that 648-post cohort, problem hooks averaged 4.3% engagement across 44 posts, comparison hooks 3.5% across 40, how-to hooks 3.1% across only 15, and posts with no deliberate hook 2.9% across 238. Question hooks averaged 2.2% across 97 posts, but drew 0.44 comments per post, close behind problem hooks at 0.48. So a question earns its place when you want replies, not reach. The benchmark page states its own limits plainly: several groups are small, the businesses are mostly local and service-based, and the figures are observations, not proof of cause. That is why the order is a place to start, and your own posts get the final say.
If the batch misses, ask for one more round with a single direction, such as "ten more problem lines, shorter, using only the words in my reviews". A third and fourth round tends to sand the lines down until they sound like everyone else's.
How to judge it
Read each line as a stranger scrolling past would, with your business name above it and nothing else. You can reject a bad line in seconds once you know the signs.
A bad result looks like this:
- Any business could post it. "Here's something nobody tells you" is a hook for everyone and therefore for no one. If you could swap in a dentist or a florist and the line still works, cut it.
- It invents a fact. A percent, a number of customers, a result or a review that you did not give the AI has to go.
- It promises more than a post can deliver. "The one trick that changes everything" sets up a letdown, and the reader remembers the letdown.
- It forces a formula. A question nobody would ask, or a myth nobody believes, reads as a template. In the benchmark, posts with no deliberate hook beat question, contrarian and myth-busting openings on engagement rate, which is a good reason to let plain lines stay in the library.
- It sounds unlike you. If the line is far louder or far more formal than your captions, your followers feel the switch before they read the second sentence.
- It shouts. Capital words, stacked punctuation and urgency words read as an ad.
A good line passes one test: the right customer reads it and thinks "that is me", and the wrong customer scrolls on without feeling tricked.
Then cut hard. Keeping fifteen good lines is a better result than keeping forty middling ones, because the list only helps if you trust everything on it.
What stays with you
The AI can write fifty lines. It cannot know your customers, check your facts or watch your posts.
These parts stay with you:
- The cut. Only you know which lines sound like the business your customers already know.
- The facts. Every claim in a hook has to be true of your work today, including services you still offer and seasons that still apply.
- The tracking. Write the hook type next to each post when you publish it, then read comments, saves and shares in the platform's own insights after a week or two. Compare like with like, since the benchmark found results differ by format.
- The ranking. Once a month, move the types that earned responses to the top of your list, and move the ones that did not to the bottom. Your own audience outranks any benchmark, ours included.
- The rest of the post. A strong first line on a weak post wastes the reader's attention. The hook opens the door, and the caption has to be worth walking into.
- The publish button. Read the line once more on your phone, in the app, before it goes out.
An example (invented business)
This is an editorial example with a made-up business, to show the shape of the work.
Say you run a two-person handyman business that fixes doors, decks and small leaks in older homes. Your captions usually open with "Another job done", and you know they get scrolled past.
You paste three services, a handful of phrases from your reviews ("the door that stuck for ten years", "he actually showed up"), your common questions and five past captions. You give one honest opinion: you would rather repair an old wooden door than replace it.
Five of the fifty might come back like this:
1. Problem: If your back door sticks every July, the frame is usually the culprit, not the door.
2. Comparison: Repair the old door or replace it: here is how we decide in ten minutes.
3. Plain: We spent Tuesday rehanging a 1940s door that had not closed properly in years.
4. How-to: One thing to check before you call anyone about a sagging deck step.
5. Question: What is the one repair in your house you keep putting off?You check them. Line one makes a claim about frames, so you confirm it matches what you see on most jobs before keeping it. Line two promises a ten-minute decision, so you keep it only because that is true of how you work. You keep all five, cut most of the generic lines from the batch, and post line one first, since problem openings led the benchmark. After a month, if the plain lines about real jobs draw the most comments from your neighbors, they move to the top of your list, whatever the benchmark said.
Keep, rank or rewrite
Each line in the batch gets one of three calls, and the call is quick once you know which applies.
| What you see | What to do | Why |
|---|---|---|
| Specific, true and in your voice | Keep it in the library | It is ready to open a post this week |
| Specific and true, but louder or stiffer than you | Rewrite it in your words, then keep it | The idea is right, only the voice is off |
| Generic, invented or a forced formula | Cut it | Ranking a weak line wastes a post |
Once lines start running, add a fourth column to your own copy of the list: what each post earned. That column is the ranking, and it is the part that makes the library more useful every month.
Where the same habit goes next
Fifty options, then a hard cut, works for other short lines too. Email subject lines are the same job for the inbox, and a brand voice guide makes every batch start closer to how you sound. If the caption after the hook is the harder part, what to do if you hate writing captions picks up from there.
Of every job on this list, social media is the one that comes back every week, and Boomp is software that does that one for you. See ten posts made from your website, free
