Verified against ChatGPT · 2026-07-24
Rewrite just the first 15 seconds when a finished script isn't landing
Isolates the cold-open hook as its own deliverable — for testing multiple openings against one already-written video body — instead of rewriting the whole script when only the opening is the actual problem.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are rewriting only the first 10 to 15 seconds of a YouTube video — the cold open — for a script whose body is already finished and is not the problem. Treat this as an isolated component with its own job: earn the next 30 seconds of attention, nothing more. Do not summarize the whole video here and do not try to make this section carry the video's full argument. VIDEO TOPIC AND ANGLE Why "just drink more water" is bad advice for most people who are actually tired EXISTING SCRIPT BODY A 7-minute script arguing that chronic tiredness in most sedentary adults correlates more strongly with poor sleep consistency and blood sugar swings than with hydration, citing three specific studies WHY THE CURRENT OPEN ISN'T WORKING Current hook opens with "Hey everyone, welcome back, today we're talking about hydration" — 9 seconds pass before any actual claim, and Studio shows a 22-point retention drop by second 15 VIEWER'S STARTING KNOWLEDGE Already knows the basic '8 glasses a day' advice and has probably heard it debunked once before — a hook that re-explains that baseline will feel slow NUMBER OF VARIANTS 4 HOOK RULES The first sentence must contain zero setup — no channel name, no "in this video," no scene description before the substance starts. Start either mid-action, mid-claim, or mid-question: a viewer who has watched zero seconds before this one should already be oriented toward a specific stake within that first sentence. Never promise something the script body does not actually deliver — read the existing script body before writing a single hook line, and reject any hook angle that the body cannot honestly pay off, even if it would test well as an isolated line. Match the register of the script body; a hook pitched more sensational than the video that follows it creates a mismatch that shows up as a fast drop the moment the tone shifts, not a save. Where the audience's starting knowledge is specified, do not open with information they already have as if it were new — that reads as slow to the exact viewer this video is for, even if it would be a fine hook for a beginner audience. If a stated reason explains why the current hook fails — too slow, too vague, promises the wrong thing, sounds like every other video on this topic — every variant you write must specifically avoid that named failure, not just be "punchier" in a way that could reintroduce the same problem in a new form. OUTPUT FORMAT Produce the requested number of hook variants, each as spoken lines only, each on its own numbered block. Under each variant, one line stating the specific psychological or curiosity mechanism it uses — a stated stake, a contradiction, a specific number, a direct question — and one line confirming exactly where in the existing script body this hook's implicit promise gets paid off.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Isolating the hook as its own artifact, separate from the script body, matches how the failure actually surfaces in the data: a retention graph with a sharp cliff in the first 15 seconds but a normal-looking curve for the rest of the video is telling you the opening is the specific broken part, and regenerating the entire script in response would touch nine minutes of content that was never the problem, discarding a body that already works while giving the model far more surface area to introduce a new mistake. Requiring the model to read the existing script body before writing any hook line, and rejecting any angle the body can't honestly pay off, targets the most common failure mode of hook-only rewrites done in isolation — an LLM asked only for "a punchier opening" has no way to know if a punchier claim is even true of the video that follows, and will happily generate a hook that promises a twist, a number, or a stake the body never actually delivers, creating exactly the bait-and-switch mismatch that produces a fast second drop the moment viewers hit the calmer, honest content underneath. Naming the specific reason the current hook fails, and requiring every variant to specifically avoid that named failure rather than just sound more energetic, prevents a subtle regression: a hook can get more exciting in tone while making the identical structural mistake — still opening with nine seconds of preamble, just louder preamble — and a vague "make it punchier" instruction will not catch that, while an instruction anchored to the specific diagnosed cause will. Tying each variant to the audience's actual starting knowledge stops a second common mismatch: a hook that re-explains a baseline fact the target viewer has already heard debunked reads as slow to exactly the audience the video is trying to hold, even though the identical hook line would work perfectly for a true beginner — the fix is not a better hook in the abstract, it's the right hook for who is actually watching. Requiring a stated payoff location for each variant turns an unfalsifiable creative claim ("this hook is more engaging") into something a reviewer can actually check against the finished script in thirty seconds.
What you get back
1. "Your tiredness has nothing to do with water — it has to do with a spike your blood sugar took two hours ago." Mechanism: direct contradiction of the assumed cause. Pays off at 1:40, where the blood-sugar-swing study is introduced. 2. "Three studies just found the exact opposite of what you've been told about why you're tired." Mechanism: stated number + contradiction. Pays off across 1:40-4:10, where all three studies are covered in sequence.
Verified against
ChatGPT GPT-5.1 · 2026-07-24
Gemini Gemini 3 Pro · 2026-07-29
Changelog
- 2026-07-24 — Initial publish, verified against ChatGPT (GPT-5.1) and Gemini (Gemini 3 Pro).
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
