Prompts That Produce Great Video Scripts (With Before & After Examples)
A video about Rome gets you Wikipedia read aloud. Here is the 4-part prompt formula, 3 narrative frameworks, and before-and-after openings showing what changes in the first 5 seconds.
Type "a video about Rome" into VidNext and you get Rome read aloud, all 10 minutes of it.
Founded on seven hills. Republic, then empire. Julius Caesar. Aqueducts were impressive. The fall was complicated.
Nothing in that is wrong. Nothing in it is yours either, and a viewer who has seen one Rome video has seen that one.
The 4-part formula below is what changes that, and Script Review is where you check it worked.
The fix takes about four minutes and it happens before you generate anything.
Why the generic prompt fails
A short prompt isn't underspecified. It's fully specified, just not by you.
Ask for a topic and the model fills every remaining decision with the most probable option. Most probable opening, most probable structure, most probable ending. That's the definition of interchangeable.
The four minutes you spend narrowing it are the four minutes where the channel becomes yours.
Here's what to narrow.
The 4-part formula
Subject. The specific thing, not the category. Not "Rome". The grain shipments that fed it.
Angle. What you're claiming. This is the part almost everyone skips, and it's the part that makes a video worth watching.
Structure. How it moves. Chronological, countdown, question and answer, argument and rebuttal.
Tone and constraints. Register, length, and what you're refusing to do.
Stack them and you get something like this:
A 10-minute documentary on how Rome's grain supply from Egypt decided its politics. Argue that the empire's foreign policy was a food-security policy. Chronological across three crises. Measured, no dramatics. Anchor each act to a dated source. Say plainly where historians disagree.
That's the same subject as "a video about Rome". It will not produce the same script.
Before and after: the first five seconds
Openings are where retention is won or thrown away, and the difference is visible in one line.
Before: "Rome was one of the greatest civilisations in history, and its story continues to fascinate us today."
After: "Rome imported enough Egyptian grain to feed a million people. When the ships were late, governments fell."
The first sentence tells the viewer they already know this. The second makes a claim they'll want checked.
Here's another pair.
Before: "Investing can seem complicated, but understanding the basics is important for everyone."
After: "Two people put in the same money for thirty years. One ends up with twice as much. The only difference is a fee of one percent."
Same subject. The second one has a number, a tension and a promise inside twelve words.
And one more.
Before: "In this video we'll explore some of the most interesting unsolved cases in history."
After: "The file was closed in 1974. Three of the four detectives who signed it later said they'd been wrong."
Three frameworks that hold ten minutes
Once the angle exists, the structure does the work. These three cover most of what a faceless channel needs.
Deep Dive. One question, four acts, an answer that arrives late. Best for history, true crime, science. Ask it in the first fifteen seconds and refuse to answer until the third act.
A 12-minute deep dive on why [event] happened, in four acts: the conditions, the trigger, the response, the part still argued about. Withhold the conclusion until act three.
Countdown. Ranked list, ascending, each item shorter than the last so pace increases. Best for niches with breadth rather than depth.
A 10-minute countdown of [seven things], ranked by [explicit criterion]. State the criterion in the first thirty seconds. Spend the most time on number one.
Myth-Buster. A claim most people believe, then the evidence against it. Highest retention of the three when the myth is genuinely common, and the worst when it isn't.
A 10-minute video examining the common belief that [claim]. Present it fairly first, in its strongest form. Then the evidence. End with what's actually true, including the part that's still uncertain.
Pick the framework before you write the prompt, not after you read the script.
If the framework is right and the writing still feels flat, the missing piece is usually rhythm rather than structure. You can point the pipeline at a video whose pacing you like and have your script written to that timing.
Then read the script
The prompt is half the job. Script Review is the other half.
The script comes back before anything is narrated. Seven minutes of reading, and it's the only quality gate that catches the thing no tool can measure.
Two questions.
Does it make a claim you'd defend in an argument? And does it say anything a hundred other channels wouldn't?
If either answer is no, change the prompt and run it again.
Verification runs alongside it, scored 81 out of 100 on the project we timed, at 18 credits. It catches factual drift. It cannot catch a boring angle, and it doesn't pretend to.
What re-running actually costs
This is the part that changes behaviour, so here are the numbers.
Script generation and verification came to 48 credits on the ten-minute project we measured. The full video was 442.
So re-running the script is roughly a tenth of the video. You can rewrite the prompt four times and still spend less than half of one render.
The Newbie pack is $15 for 500 credits. Three or four script attempts cost you well under a dollar each.
Credits do not expire, so iterating on the prompt is not burning an allowance you paid for by the month.
That's the whole argument for fixing the prompt rather than fixing the video. The prompt is cheap. Everything downstream of it is not.
Why a generic prompt is a monetization problem, not just a boring one
This is the part people find out too late, so here it is early.
YouTube's monetization policies treat mass-produced and repetitive content as ineligible for the Partner Programme. The rule exists specifically to catch channels that publish volume without a view.
The policy language is about "the creator's original, authentic insights or perspective." A prompt that says "a video about Rome" contains none, and neither does the script it produces.
That is the real cost of the generic prompt. Not that the video is dull. That the channel it belongs to may never be eligible to earn.
The angle in your prompt is the thing the policy is asking for. Write it down and you have complied. Skip it and no amount of production quality substitutes.
Why the prompt is the only cheap place to fix anything
Here's the structural reason this matters more than any editing tip you'll read this year.
Built by hand, a researched ten-minute video runs 16 to 30 hours across five or six tools, and by the time you discover the angle is weak you have already paid for footage, narration and a render you now have to throw away. That range is an estimate creators report, not something we timed.
Through the pipeline the same video is 65 minutes of wall clock and 22 minutes of your attention, and the script arrives before a single frame is sourced.
That ordering is the whole gift. You find out the angle is boring at 48 credits instead of 442.
There's also a date pushing on this. The Partner Programme threshold doubles on 1 February 2027, and clearing it beforehand takes roughly ninety-six videos, which is four and a half a week from September and eleven a week from December.
At eleven a week you do not have time to rescue weak scripts in the edit. You have time to write a better prompt, and that is the only lever that scales.
What we got wrong at first
Worth saying, because we made this mistake for a while.
We assumed longer prompts were better prompts and started writing paragraph-long instructions with tone notes, pacing notes and vocabulary lists.
The scripts got worse. Not dramatically, but noticeably flatter.
Our best guess is that piling on constraints crowds out the one thing that matters, which is the angle. A prompt with a sharp claim and three lines of structure beat a prompt with a vague claim and twelve lines of style direction, every time we compared them.
We have not run this as a controlled test with enough samples to call it a finding. Treat it as a working habit rather than a law: get the angle right first, and add constraints only when something specific goes wrong.
The 30-second version
Name the specific subject. State what you're claiming about it. Choose one of the three structures. Say what you're refusing to do.
Then read the script before you narrate it.
That's four minutes of prompt and seven minutes of reading, on a video that takes 65 minutes to exist and 22 minutes of your attention overall.
Those eleven minutes are the entire difference between your channel and a content farm.
Try one prompt three ways and read all three scripts before you generate anything.
Credit figures come from one VidNext project of 600 seconds, measured 18 June 2026. Monetization eligibility wording quoted from YouTube Channel Monetization Policies as of June 2026. The before-and-after openings are written examples, not retention measurements: we have no per-opening retention data and neither does anyone quoting a figure for it.
