Generate 20 hooks for a topic and rank them by a stated rubric
One prompt that produces 20 openers across the 8 archetypes, scores each on four named criteria, and returns the top 5 with the score shown, so the ranking can be argued with instead of trusted.
Generate 20 hooks for a short vertical video about {topic}, aimed at {audience}, where the video's goal is {reach / convert / trust}. Spread: at least 2 of each archetype: question, bold claim, pain point, curiosity gap, 'stop doing X', number, contrarian, result-first. Label each with its archetype. Rules for every hook: max 12 words; spoken and on-screen are the same words; present tense; no 'ever wondered', 'what if I told you', 'you won't believe', 'secret', 'hack', 'ultimate'; a number only if the reader could measure it. Score each hook 0-3 on four criteria and show the four digits: - SPECIFIC: names a place, object, number or moment (3) vs could be about any topic (0). - CHECKABLE: the video can prove or deliver it by 20 s (3) vs it is an opinion or a tease (0). - SOUND-OFF: works as text on a still frame (3) vs needs the voice (0). - MATCH TO GOAL: for reach, how many people have a guess or an opinion; for convert, how close it is to the moment of need; for trust, how much it admits (3 to 0). Return: a table of 20 rows (number, archetype, hook, four digits, total out of 12), then the top 5 by total with one line each on what frame 1 shows, then the 3 lowest with one line on why they failed. Do not round the scores up to make a nice top 5; ties are fine.
In Claude Code, paste the prompt as a message.
Steps
- Fill topic, audience and the goal. The goal changes the fourth score; 'reach' and 'convert' rank differently.
- Shoot the top 3, not the top 1. The rubric predicts the order wrong often enough that three is the cheapest insurance.
- Put each one on its own campaign link and judge after a week on installs (see the A/B entry).
Example
Topic: a weeknight cooking channel, one chicken into three dinners. Audience: parents cooking on weekdays. Goal: reach.
| # | Archetype | Hook | S C O M | Total |
|---|---|---|---|---|
| 1 | result-first | Three dinners. One chicken. Seven dollars forty. | 3 3 3 2 | 11 |
| 2 | number | One chicken, three meals, one day. Let’s see what it costs. | 3 3 2 3 | 11 |
| 3 | pain-point | It’s six pm, the chicken is frozen, and the kids are asking. | 3 2 2 3 | 10 |
| 4 | stop-doing | Stop buying chicken breasts. Buy the bird. | 2 3 3 2 | 10 |
| 5 | contrarian | Meal prep on Sunday is why you order takeaway on Wednesday. | 2 2 2 3 | 9 |
| 6 | question | Why is the whole chicken cheaper than two breasts? | 2 3 2 2 | 9 |
| 7 | curiosity-gap | The third dinner is the one nobody expects. It’s at the end. | 2 3 1 2 | 8 |
| 8 | bold-claim | Under two dollars a plate, four people, no coupons. | 3 3 3 2 | 11 |
| … | (12 more rows) |
Top 5: 1, 2, 8, 3, 4. Frame 1 for #1: three plates in a row, receipt propped behind them. Frame 1 for #2: the whole bird on the board, knife beside it.
Lowest 3: ‘Cooking doesn’t have to be hard’ (0 0 1 0, portable to any topic); ‘You won’t believe this chicken trick’ (0 0 1 1, banned phrase and nothing checkable); ‘Let me show you my process’ (0 1 0 0, no promise).
Tips
- Showing the four digits is the point. A ranked list with hidden scores cannot be corrected; a list with scores lets you say 'that one is not a 3 on checkable' and move on.
- In Claude Code, ask for the table as a CSV file so the three you shoot can be matched to their campaign links later.
| works with | Claude Code: Runs as a skill or a prompt in the terminal; can read your files and write the result. · Claude: Paste into claude.ai or the Claude app. · ChatGPT: Paste into ChatGPT; works with any GPT model. · Any LLM: Model-agnostic: Gemini, Mistral, a local model. |
| cost | Free. No API, no key. |
| status | Unverified: written from the technique, not yet run end to end by us. |
| license | CC BY 4.0 for the text here; skills are MIT. |
| updated | 2026-11-01 |
| for agents | generate-20-hooks-ranked-by-rubric.md |
Related
A/B test hooks with campaign links: judge after a week on installs, not views
The creator-growth method: same video, different hook, one store campaign link per variant, installs per campaign read after a week. Includes the iOS ct= link shape and the noise rule.
Hook rewrite: a boring opener in, 5 stronger versions out, each with the reason
A prompt that diagnoses why an opener is weak (generic, unprovable, needs sound, buried number) and rewrites it five ways, each labelled with the archetype and the fault it fixed.
Ask an agent to turn a script into a numbered generation shot list
A prompt for Claude or ChatGPT that converts a 30-second script into 6-8 shot prompts in the 7-part structure, ready to paste into a video model one by one.
Generate a 10-clip b-roll pack for a talking-head video
One agent prompt that reads your transcript and writes ten generation prompts for cutaways that show what is being said, each tagged with the sentence it covers.
The day as one text file: an agent reads the speech and proposes the story
The step before any timeline: the day's speech in shooting order, numbered by clip and sentence, read by an agent like a scriptwriter. It proposes story, scenes, hook and cutaways; you argue in text.
Forced reasoning before scripting: goal, hook type, demo moments, emotion arc
Four questions an agent must answer in writing before it writes a line of script. Skips the generic promo every model produces when asked for one directly.