Attention without sacrificing the payoff.

The problem

A generic hook library treats attention as one problem with one solution: a stock formula ("Most people think X. They're wrong.") applied regardless of platform, audience or what the content actually delivers. The same structural hook that stops a scroll on a founder-audience LinkedIn post can fall completely flat on a short-form video aimed at a technical audience — the mechanism that earns attention is not the same everywhere, and a one-size template quietly optimises for interruption at the expense of trust.

Why not just ask for it in the prompt

Asking a model to "write a better hook" in the moment tends to reach for whichever pattern is statistically most common — which is exactly how the same handful of tired constructions end up everywhere. Fixing that requires a process that deliberately generates from several different attention mechanisms and screens for platform and voice fit before ranking, not a single-shot request repeated slightly differently each time.

How it works

• —Extracts the truthful payload of the content first, then generates five to seven structurally different candidate hooks from different attention mechanisms.
• —Scores each candidate on relevance, reward, tension, specificity and credibility, but uses platform fit and voice fit as hard gates — a candidate that fails either is not recommended, whatever it scores.
• —Checks that the hook’s promise matches what the content actually delivers before handing it over.

Evidence and provenance

• —Reads performance data as early attention plus downstream outcomes together, so it does not mistake an isolated viral post for a repeatable pattern.
• —Asks at most three compact questions, and only when the missing answer would materially change the hook.

What it produces

• —A recommended hook with the reasoning behind it, plus labelled alternatives — usable copy first, explanation after.
• —Format-specific structure for short video (first frame, spoken opening, on-screen text) and long-form video (title, thumbnail concept, opening, the promise being fulfilled).

What it must not do

• —Never invents a number, result, client, event, quote, credential or personal experience to make a hook land harder.
• —Never copies signature wording or a distinctive personal anecdote from a source hook it is learning from.
• —Never promises virality or presents one platform’s benchmark as a universal guarantee.

In practice

The same underlying insight (that a decision framework produced a "provisional" recommendation instead of a false-confident number) earns a different opening depending on where it runs: a LinkedIn post for founders might open on the specific conflict ("Two sources, same metric, two different numbers — which do you trust?"), while a technical audience post might open on the mechanism itself. Same truthful content, deliberately different entry points, because the two audiences arrive caring about different things.

Why a reusable skill, not a one-off prompt

A hook that works is easy to imitate badly — the surface pattern gets copied while the underlying judgment (platform fit, voice fit, promise-to-payoff match) does not travel with it. Packaging the six-step engine and its hard gates as a skill means that judgment is applied consistently, not just remembered by whoever wrote the last good one.

Mohammed's contribution

Defined the six-step engine and, specifically, the rule that platform fit and voice fit act as hard gates rather than scored inputs — a candidate that fails either is excluded outright, regardless of how well it scores on relevance or tension.