Hook And Human

Hook and Human

The persuasion sibling of clear-and-human. That skill strips engagement patterns to protect credibility; this one deploys them on purpose to win attention and conversions — without crossing into slop. Marketing copy that reads like a person, hooks in the first line, and asks for exactly one thing.

Use this for social, ads, sales, and landing copy. Use clear-and-human for docs, runbooks, and anything where authority beats reach.

Modes

• Write — generate copy from a brief ("write a LinkedIn post about X"). Pick a framework, lead with a hook, close with one CTA.
• Punch-up — the user has flat copy that's accurate but doesn't land. Add a hook, tighten rhythm, sharpen the CTA. Don't change the claims.
• Review — score the copy for stopping power and conversion (see scoring). Offer a punch-up.

Default to Write for a brief, Punch-up for pasted flat copy, Review if they ask "is this any good?".

The floor (non-negotiable — this is what keeps it human)

These carry over from clear-and-human unchanged. Persuasion is not a license to break them:

1. Never fabricate specifics — including the soft ones. No invented stats, testimonials, customer names, results, or credentials. Use only what's in the brief or WRITING_CONTEXT.md. The trap in marketing copy is the soft fabrication, because it doesn't look like a lie: a derived number (rounding an approved "40→6 min" into "saves 30 minutes"), an aggregate-time claim ("get your afternoon back" — you don't know how often they deploy), or a soft outcome ("fixes ship the same hour they're reported") are all fabrications too. The reader can't tell invented color from a measured result, so each one spends trust you haven't earned. Stick to the approved fact and let it carry the weight; if you want another proof point, leave [ADD PROOF] and say so. A made-up implication damages credibility as fast as a made-up number.
2. Cut generic AI filler. Hooks are fine; filler is not. Still banned: delve, leverage, synergy, robust, seamless, transformative, tapestry, "in today's landscape", "unlock your potential". These don't persuade — they signal a machine.
3. No hollow vulnerability or fake humility. "Can I be real for a second?", "I'm no expert, but…" — these backfire; readers smell the performance. If a personal story is real, tell it plainly. If it isn't, don't.
4. One CTA, specific. Not three asks, not "thoughts?". One clear next action.
5. Vary rhythm and match voice. Short line, then a longer one. Pull voice from WRITING_CONTEXT.md. Uniform cadence reads as generated even when the words are punchy.

Context

Read WRITING_CONTEXT.md (or FOUNDER_CONTEXT.md) from the project root first — same file clear-and-human uses. Pull brand voice, audience/ICP, the offer, approved proof points, and banned phrases. If it's missing and the brief is thin, ask up to 3 questions (offer, audience, the one action you want) before writing.

What to deploy (the part clear-and-human suppresses)

Full detail in references/frameworks.md. The essentials:

• Lead with a hook. First line earns the second. Specific > clever. Open a curiosity gap you actually close.
• Pick a framework to structure the body: AIDA, PAS, or BAB (see reference). Don't show the seams.
• One CTA, concrete and low-friction ("Reply 'guide' and I'll send it", not "Let me know if interested").
• Format for the channel — line breaks, length, hashtag/emoji norms in references/channels-marketing.md.

The difference from slop: a hook backed by a real specific is persuasion; a hook backed by nothing is bait. Earn the emphasis with the actual number or story, never with ALL-CAPS or "Read that again."

Review scoring (1–10)

| Dimension | Measures | Target | |---|---|---| | Scroll-stop | Does the first line stop the reader? | 8–10 | | Offer clarity | Is the value/offer unmistakable? | 8–10 | | Conversion | One clear, specific, low-friction CTA? | 8–10 | | Authenticity | Sounds like this person/brand, not generic AI? | 7–10 |

If Scroll-stop is high but Authenticity is low, flag it: attention-grabbing but hollow, won't build trust or repeat.

Self-audit (run before presenting Write or Punch-up output)

1. Would the first line actually stop me mid-scroll, or is it a template hook with no payoff?
2. Is there exactly one CTA, and is it specific?
3. Did I invent any number, name, or result — including a derived number, an aggregate-time claim ("hours back", "your afternoon back"), or a soft outcome the approved facts don't actually support? If yes, cut it or replace with a placeholder.
4. Does it sound like the brand in WRITING_CONTEXT.md, or like generic marketing AI?

Fix what fails, then present.

Relationship to clear-and-human

Siblings, not a pipeline. Pick by intent: reach/conversion → here; credibility/clarity → clear-and-human. They share WRITING_CONTEXT.md and the no-fabrication floor, so a piece can move between them without contradiction. Don't run both on the same draft in sequence — they optimize for different things and will fight.

Provenance

Adapted (MIT / public domain): persuasion frameworks and hook/CTA patterns synthesized from common copywriting practice and ognjengt/founder-skills (MIT); the no-fabrication floor and voice/rhythm rules inherited from its sibling clear-and-human.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: kevin-burns
• Source: kevin-burns/claude-skills
• License: MIT
• Homepage: https://techblog.kevinburns.de

Install and usage instructions live in the source repository linked above.