Hook Analysis: the complete framework for scoring, testing, and rewriting the opening that decides retention

Short-form retention curves consistently show the steepest drop between second one and second three. That window is where the viewer asks three silent questions: do I understand what this is, do I care, and is the next moment worth more than the next swipe. If the opening cannot resolve those three questions immediately, the algorithm sees a swipe-away and learns not to recommend the video again.

Long-form is no different. The retention chart of a 20-minute YouTube video lives or dies on the first 30 seconds. A weak cold open does not just lose those early viewers — it tells the algorithm that this video does not deserve impressions, which compounds across the lifetime of the upload. Hook analysis is the only review layer that catches this before it happens.

What makes the first three seconds difficult is that the creator already knows the payoff. The viewer does not. Hooks fail when the creator writes them from inside their own head instead of from the position of a cold stranger swiping fast. Hook analysis is the practice of forcing yourself back into that stranger's seat.

Every hook can be scored across five independent dimensions. Treating them independently is what turns hook analysis from vibes into a repeatable system. A hook can have a brilliant curiosity gap and still fail because its promise is unclear. A hook can have perfect clarity and still fail because there is no gap to pull the viewer forward.

Promise clarity asks whether a stranger understands within one sentence what they are about to watch. Curiosity gap asks whether the stranger has a specific question that only this video can answer. Pattern interrupt asks whether the opening visual, audio, or framing breaks the feed's rhythm. Proof density asks whether the first ten seconds deliver evidence that the promise is real. Payoff alignment asks whether what the video promises matches what it actually delivers.

When you grade a hook on all five dimensions, the weakest one is usually the cause of the drop-off. A 90/90/90/90/40 hook performs like a 40, not like an 80 average. The diagnostic value of the framework is in finding the one dimension that needs the rewrite.

Most strong hooks fit one of seven repeatable patterns. The contrarian hook reverses a widely accepted belief. The numbered list hook front-loads a quantified promise. The story hook drops the viewer into a moment of consequence. The before-after hook shows transformation in one frame. The mistake hook names a failure the viewer has felt. The forbidden-knowledge hook offers information the viewer believes others do not have. The stakes hook makes an irreversible decision visible.

These are not gimmicks. Each pattern works because it satisfies one of the underlying psychological triggers — curiosity, identity, stakes, novelty, social proof, or self-interest — within the first few seconds. The choice between patterns is not about which is best. It is about which fits the video's actual content. A story hook on a tutorial video creates friction. A list hook on a personal story video kills the emotional stakes.

Strong creators usually build a hook library — a personal collection of formulas that work for their niche, with three rewrites of each — so they never start from a blank page. Hook analysis becomes a library lookup, not a creative invention, which is what makes weekly publishing sustainable.

Pre-publish hook testing has two layers. The first is internal — write three variants and score each across the five dimensions, then ship the highest weighted score. The second is external — when stakes are high, post a 15-second silent test of the hook on Stories or a private Discord, watch how many people stop scrolling, and decide whether the hook deserves the full production.

The pre-publish workflow takes 20 minutes per video and prevents the most expensive mistake in creator life — editing a 6-hour video around a hook that nobody clicks. The math is simple: 20 minutes of hook analysis can save the next 30 hours of production from being wasted.

The mistake most creators make is testing after publishing. By then the data is contaminated by the algorithm's distribution decision, which was itself based on the hook. Test before. Test cheap. Test often.

YouTube long-form hooks need a clear promise within five seconds plus an explicit payoff timeline — telling the viewer what they will get and roughly when. Long-form viewers are willing to invest more time but punish vague openings harder than short-form viewers do because the time commitment is larger.

YouTube Shorts hooks live or die on the first frame and the first spoken word. Sound on is rare in the early seconds, so the visual has to carry comprehension on its own. The most reliable Shorts hooks use on-screen text to state the promise within the first 800ms.

TikTok hooks are spoken-word dominant because TikTok viewers default to sound on. The opening line is everything. The visual matters but it is the audio cadence that signals whether to swipe. TikTok hooks also benefit more from contrarian framing than YouTube hooks because TikTok's algorithm rewards comment volume, and contrarian openings generate comments.

Reels hooks fall between Shorts and TikTok — sound is on more often than Shorts but less often than TikTok. The strongest Reels openings combine an on-screen text overlay with a spoken hook that says the same thing differently. That redundancy catches both sound-on and sound-off viewers.

The most common failure is the throat-clear — opening with 'Hey guys, today we are going to talk about...' which the algorithm sees as filler. The rewrite is to cut everything before the first specific promise. If the video can start at second four with no loss of meaning, those four seconds were dead weight.

The second most common failure is the buried lede — the actual hook is at 0:15 instead of 0:01. The rewrite is to take the most interesting moment of the video and play it as a teaser inside the first three seconds. This is sometimes called a 'cold open' and is the single highest-leverage edit a creator can make.

The third failure is the over-promise — a hook that sets expectations the video cannot deliver. This wins clicks but destroys retention, which is worse than no clicks at all. The rewrite is to soften the promise until it matches what the video actually delivers. A hook that promises 60% and delivers 60% beats a hook that promises 100% and delivers 40%.

The fourth failure is the abstract opening — claims with no specifics. 'This will change your life' is a non-hook. 'I quit my $180,000 job after watching this' is a hook. The rewrite is to find one concrete detail and lead with it.

The creators who win at hook analysis treat it as a database problem, not a creative problem. They keep a spreadsheet — or a Notion table, or a private hook-library tool — of every hook they have ever written, with its 0-100 score, the pattern it used, the video CTR, and the 30-second retention.

After 30 to 50 entries, the pattern becomes obvious. Some niches reward contrarian hooks. Others reward story hooks. Personal hook history becomes the most accurate predictor of what will work next time, far better than generic best-practice articles.

The compounding effect is that hook analysis becomes faster over time. The first 20 hooks take 20 minutes each. The next 100 take five minutes each, because the creator now has internal patterns and a personal library to draw from.

Hook generators invent hooks from a prompt. Hook analysis takes a hook the creator has already written and tells them why it works or fails. The two are complementary. Generators are useful when the creator is stuck. Analysis is useful when the creator already has an opening and needs to know whether to ship it.

The risk of pure generation is that AI-generated hooks tend to sound generic — they hit the surface patterns but lack the specific detail that makes a hook real. The risk of pure analysis is that it provides feedback but not new ideas. The strongest workflow is to generate three drafts, analyze each, then write a fourth version that combines the strongest dimensions from each.