What hook-rate actually measures
Hook-rate is the percentage of viewers who watch past the first 3 seconds of a short-form video. It is not the same as retention or completion rate — those measure middle and end. Hook-rate isolates whether your opening earned the attention required for everything else.
Strong hook-rate on TikTok is 70%+ for established creators, 80%+ for top-tier. Reels and Shorts skew slightly higher because the audience self-selects more. If your hook-rate is below 60%, the rest of the video does not matter — most viewers never see it.
The asymmetry: improving hook-rate from 50% to 75% typically doubles total views on the same content. Improving the middle of the video from "good" to "great" produces single-digit lift. Hook-rate is the highest-leverage knob you can turn.
The gate: write 10 hooks, ship 1
Before you film, write 10 candidate hooks for the same video. They should differ on dimension and angle — not just word swaps. Some are pattern interrupts, some are bold claims, some are direct questions, some are stat reveals, some are "what most people get wrong" framings.
Then score each hook on three quick gates: would you stop scrolling if you saw it (gut check), does it imply a specific payoff (curiosity gap), is it specific enough to be unfakeable (no "5 tips to grow your business" generic). Hooks that fail any gate get cut.
You typically end up with 2-4 hooks that pass. Pick the one you have least seen done elsewhere — novelty beats polish at the hook stage.
10-hook brainstorm structure
- Hook 1 — Pattern interrupt: "I deleted [thing everyone uses] and grew 3× faster."
- Hook 2 — Bold claim: "Most [people in your niche] are doing [thing] wrong."
- Hook 3 — Direct question: "Why do [your audience] keep failing at [thing]?"
- Hook 4 — Stat reveal: "73% of [niche] never make it past [milestone]. Here is why."
- Hook 5 — "What if": "What if I told you [counterintuitive thing]?"
- Hook 6 — Personal stake: "Last week I [specific embarrassing thing]. Here is what I learned."
- Hook 7 — Anti-pattern: "Everyone says [common advice]. They are wrong."
- Hook 8 — Curiosity gap: "There is one thing [successful person] does that nobody copies."
- Hook 9 — Direct address: "If you are [specific persona], this is for you."
- Hook 10 — Timestamp specific: "I just discovered [thing] yesterday. Telling you immediately."
Where AI helps and where it does not
AI is good at generating hook candidates from the same topic. Give Claude or GPT the topic, your niche, and the 10-hook structure above, and you get 10-20 candidates in 30 seconds. Use them as raw material, not final copy — AI hooks tend to be too smooth, too marketing-speak.
AI is bad at predicting which hook will perform. Hook-rate is downstream of taste, audience knowledge, and timing — variables AI does not have access to. The validation has to come from posting, not from asking GPT "which one is best."
The hybrid that works: AI generates 10-20 raw candidates, you rewrite the top 3-5 in your voice, post one as the lead video and one as a remix variant a few days later. Compare hook-rate in the analytics, learn which framings your specific audience responds to, feed that back into future hook prompts.
The remix variant — test without doubling work
The cheap way to A/B hooks: post the same video twice with different hooks, 2-7 days apart. Same script body, same b-roll, swap the first 3 seconds. Most platforms (TikTok especially) will not penalize the second post if the rest of the content is genuinely valuable.
This produces real performance data on hook-rate from your actual audience, not predictions. Over 5-10 such tests you build a "hook style guide" for your specific niche and your specific channel — what framings consistently outperform.
The whole loop fits inside the short-form workflow in workflow 02. The AI agent generates candidates, you pick winners, you post the lead and the remix variant, the agent watches the analytics and reports back which framing won. You stop guessing and start compounding.
The 3-second test
Before you ship a hook, ask: would I keep watching if a stranger sent me this video? If you hesitate, the hook is not done. The hesitation is the same one your viewers will have — and they will scroll instead of giving you the benefit of the doubt.