The short version: a single "write me a YouTube script about X" prompt will always sound generic, because ChatGPT has no idea how you talk, what you have already said on camera, or where your audience drops off. The prompts below work because they feed ChatGPT your own material first and build the script in layers, hook, outline, draft, tightening pass, rather than asking for a finished script in one go.
Why one prompt never gives you a usable script
I have watched dozens of small business owners try this the fast way. They type "write a 5 minute YouTube script about email marketing for small business" into ChatGPT, get 800 words back, read them out loud once, and quietly delete the video before uploading it. The script is not wrong exactly. It is just nobody. No filler words, no personality, no sense that a real human with opinions wrote it.
That is not a ChatGPT failure. It is a prompting failure. The model is doing exactly what you asked, writing generic content for a generic audience because that is all the information it was given. If you want a script that sounds like you, you have to feed it you first.
The five step prompt sequence that gets a usable draft
Step 1: Give it your voice before you give it a topic
Before I ask for anything, I paste in the transcript of two or three of my own past videos, or if a client has never filmed before, three minutes of them talking on a voice note about their business. The prompt is simple:
"Here are three transcripts of me speaking on camera. Read them and describe my speaking style in bullet points, sentence length, pace, use of humour, how I open, how I use examples, what phrases I repeat."
This single step is the one most people skip, and it is the reason their scripts read like blog posts with line breaks. It takes two minutes and changes everything that follows.
Step 2: Ask for an outline, not a script
"Based on my speaking style above, give me a 6 point outline for a YouTube video on [topic]. Each point should be one sentence, in the order I should say them, aimed at someone who is [specific description of viewer, not "beginners"]."
Being specific about the viewer matters more than people think. "Someone who has tried Mailchimp once and gave up because open rates were under 15%" produces a better outline than "small business owners interested in email marketing."
Step 3: Nail the first 15 seconds separately
YouTube's own retention data shows most drop off happens in the first 15 to 30 seconds, so I never let the hook get buried inside a full script prompt. I ask for it alone:
"Write 5 different opening lines for this video, under 12 words each, that state the specific result or surprising claim, no 'in this video I'm going to' phrasing, no questions."
That last instruction matters. ChatGPT defaults to rhetorical questions as openers ("Have you ever wondered why..."). They are the single most overused hook in YouTube scripts right now and viewers scroll past them.
Step 4: Now ask for the full script, in chunks
"Using outline point 1 and this opening line, write 150 to 200 words of spoken script in my voice from step one. Write it the way I would say it out loud, not the way I would write it, short sentences, contractions, occasional incomplete sentence."
I do this outline point by point rather than asking for the whole script at once. Chunked requests keep the pacing tighter because ChatGPT is not trying to manage the whole runtime in one pass, and 150 to 200 words matches roughly one minute of natural spoken pace for most people.
Step 5: The tightening pass
"Cut this by 20% without losing any specific numbers or examples. Remove any sentence that just repeats the point before it."
This is the pass that turns an AI draft into something you would say. First drafts from ChatGPT are almost always over-explained, it will make a point, then restate it, then give an example, then restate the example.
A real example from my own channel
I tested this exact sequence on a video I filmed about LinkedIn lead generation. My one-shot prompt draft came in at 640 words, opened with "Have you ever struggled to generate leads on LinkedIn?", and took me 40 minutes to rewrite before I could film it. The five step version, using two of my own old transcripts as the voice sample, came in at 610 words, opened with "I got 14 client calls from one LinkedIn post, here's the exact structure", and I filmed it almost word for word with only small changes as I spoke. Same topic, same length, roughly the same time spent prompting, a completely different amount of rewriting needed afterward. The difference was not the topic or the length. It was that the second version had my actual habit of leading with a number.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Prompts for specific video types
Different formats need different framing. These are the ones I reuse most:
- Talking head explainer: "Write this as if I am speaking directly to camera, no cutaways or b-roll notes, just spoken words, in my voice from the transcripts above."
- Listicle style video ("7 tools for..."): "For each of the 7 items, give me one sentence stating what it is, one sentence with a specific number or price, and one sentence on who it is wrong for. Keep each item under 40 words total."
- Client story or case study video: "Structure this as: the situation before, the specific thing we changed, the specific result with a number, what I would tell someone in the same position. Do not use the word 'journey'."
- Short-form (under 60 seconds): "Write this in under 100 words total, one idea only, cut every sentence that is not the hook, the point, or the call to action."
- Tutorial or how-to: "Number each step, one sentence of what to do, one sentence of why it matters or what goes wrong if you skip it, no more than 5 steps."
If you are choosing between ChatGPT and a dedicated writing tool for this, I have gone through what holds up in the best AI tools for script writing in 2026, and for most solo creators ChatGPT with a good prompt sequence still beats a paid script-specific tool.
The part nobody wants to say out loud
Here is the uncomfortable bit. A better prompt will not fix a boring video. I have seen creators spend three hours refining a script through five rounds of ChatGPT prompts, get a sharp piece of writing, film it flatly with no energy and no eye contact, and get worse retention than a rough unscripted take filmed with actual enthusiasm. The script is maybe 30% of why a YouTube video works. Delivery, pacing on camera, and whether you believe what you are saying carry the rest. If your last five videos have flopped, the fix is not a better prompt, it is filming test footage and watching it back before you touch ChatGPT again.
I also do not use ChatGPT to write full scripts word for word anymore for anything I plan to say on camera myself. I use it for structure, for the hook options, for tightening a draft I already wrote by hand. If you read your own script and every sentence is perfectly formed and grammatically neat, that is usually the sign a script will sound written rather than said. Real speech has fragments in it.
Where ChatGPT still gets it wrong
It will overuse "let's dive in" and "game changer" unless you specifically ban them in the prompt. It cannot judge pacing against your actual video length without you telling it words per minute, I use 130 to 150 words per minute of usable footage as my rough guide for talking-head style. And it will happily write a script that is factually confident and completely wrong on any statistic, so if you ask it for a number, check it before you say it on camera.
For anyone building a wider content workflow rather than one script at a time, it is worth reading how these tools fit into a broader system, my piece on how productivity tools work on a computer covers the setup side, and if you are choosing between different AI writing tools for the business overall rather than just video, I compared several in the best AI assisted writing tools for small business owners.
If you are weighing up whether to keep doing this yourself in ChatGPT or bring someone in to build the whole system for you, I wrote a straight comparison of costs and outcomes in do you need an AI consultant, or can ChatGPT do the job, worth reading before you spend on either.
Related tool: my free prompt generator.
For the rest of the series, see the ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps.
Frequently asked questions
Can ChatGPT write a full YouTube script in one prompt?
It can produce something, but a single prompt without your speaking style, your specific viewer, and a word count guide will read as generic. The five step sequence above, voice sample, outline, hook, chunked draft, tightening pass, takes about 20 minutes longer but needs far less rewriting afterward.
How long should a ChatGPT generated YouTube script be for a 10 minute video?
Aim for 1,300 to 1,500 words as a starting draft, based on 130 to 150 spoken words per minute for talking-head style delivery. Always cut it down after reading it aloud once, first drafts run long almost every time.
Should I feed ChatGPT my old video transcripts before asking for a script?
Yes, this is the step that makes the biggest difference in how natural the output sounds. Paste in two or three transcripts of you speaking and ask ChatGPT to describe your style before you ask for any script content.
Why do ChatGPT YouTube scripts all sound the same?
Because most people prompt with a topic and nothing else, so the model defaults to safe, generic phrasing, rhetorical question hooks, "let's dive in," repeated restating of points. Specific viewer descriptions, banned phrases, and your own speaking sample fix most of this.
