§ 01 · The Hook
The bait, then the rug-pull.
There is one step almost every AI-assisted scriptwriter skips — and it is the step that would have made every script after the first one faster to write. George Blackman's answer is three prompts and two document re-uploads, run once after every video.
Where the time goes.
00:00 – 00:24
01 · Hook + promise
Pattern interrupt opener, 3-month usage claim, sets up the one missing habit that compounds across every future script.
00:24 – 01:05
02 · The problem: static documents
Audience avatar and style guide alone are not enough. Making edits in Google Docs or Notion keeps Claude blind to its own mistakes.
01:05 – 01:33
03 · The insight: force the diff
If Claude never sees the difference between what it generated and what you filmed, it cannot improve. The solution is forcing it to compare.
01:33 – 02:14
04 · Collaboration report prompt
Paste the prompt alongside your final edited script. Claude surfaces Tier 1 (style) and Tier 2 (structure) learnings in a structured report.
02:14 – 02:53
05 · Absorbing the learnings
Two follow-up prompts: one updates the style guide with Tier 1 learnings, one updates the training doc with Tier 2. Both get re-uploaded to the Claude project.
02:53 – 03:58
06 · Compounding loop + CTA
Each script adds context. The caveat: the system amplifies the human's own writing quality, good or bad. CTA to next video on curiosity mistakes.
Visual structure at a glance.
Named ideas worth stealing.
01:33
list
Three-Prompt Feedback Loop
- Collaboration Report Prompt — compare AI draft vs. final script, extract Tier 1 + Tier 2 learnings
- Style Guide Update Prompt — absorb Tier 1 (voice/style) learnings into the style guide document
- Training Doc Update Prompt — absorb Tier 2 (structural) learnings into the training document
A post-script ritual that feeds the gap between AI output and edited final back into Claude project documents, compounding style accuracy across every future script.
Steal for Any AI-assisted writing workflow — replace script with newsletter, email sequence, or blog post
02:14
model
Tier 1 / Tier 2 Learning Classification
- Tier 1: style, voice, language patterns, sentence rhythm → updates Style Guide
- Tier 2: structural changes, segment order, hook architecture → updates Training Document
Separates subjective voice corrections from objective structural corrections so each type flows to the right document.
Steal for Any prompt engineering or AI workflow where style and structure need separate treatment
Lines you could clip.
01:20
"If every time an AI generates a script, you go and make your edits somewhere else, how is it gonna know what it should be doing better the next time?"
01:31
"Force Claude to learn from its mistakes."
03:40
"Because otherwise, you're simply gonna be training the AI on bad script writing principles."
Things they pointed at.
How they asked for the click.
03:40
next-video
"So watch this video next to understand the huge mistake most YouTubers make when adding curiosity to their scripts"
Soft bridge CTA — no subscribe pitch, no newsletter, no sponsor. Just a contextually relevant next-step video. Clean and friction-free.
Word for word.
HOOK opening / re-engagementCTA the pitch analogy story
00:00HOOKOkay. Don't write another YouTube script using AI until you've seen this. I promise you I'm about to save you a whole bunch of time.
00:07HOOKNow I've been using Claude to help me write YouTube scripts for the last three months and today I wanna show you the most important thing that you're probably not doing that's gonna make each script faster to write than the last one. This takes two minutes and you will be kicking yourself you didn't implement this sooner.
00:22HOOKNow you might think that giving Claude an audience avatar and a style guide document might be enough to make it sound like you and to make it generate scripts in your tone. But the truth is those two documents are not enough on their own. Now I don't know exactly what your process looks like for writing a script with AI, but I imagine it will be something like this.
00:39There might be some back and forth going over the brainstorm. You might be deciding on the topic. You might be fleshing out the segments of your video piece by piece before asking the AI to just completely generate an entire script for you.
00:50Now I'm not sure exactly what your process looks like, but the point I'm getting at here is you reach a point where you have a fully written script that in some way you have used AI to help you generate. Now what most people will do here is take this AI generation, put it into Google Docs or Notion and make the edits that they wanna make to turn it into a script that's ready to film.
01:09And this is fine. Right? There are always gonna be changes that you have to make to something that's AI generated.
01:14But the problem is by doing that, the AI isn't getting a chance to learn from its mistakes. Think about it. If every time an AI generates a script, you go and make your edits somewhere else, how is it gonna know what it should be doing better the next time?
01:26So the simple solution here is to force Claude to learn from its mistakes. And I'm gonna show you how to do that now with one of the simplest yet most powerful prompts you will ever need for writing a YouTube script. And by the way, to make this easier, you can grab my whole Claude script writing process including the prompt I'm about to show you in the description below.
01:44It's completely free and honestly, this has been a game changer for my channel. So this is the prompt that we're gonna be using. I want you to copy and paste this exact prompt into your Claude chat.
01:53Now what it's gonna do is look for differences between the portions of the script or the whole script, if that's what you used it for, that it generated compared to the final version of the script that you have edited and thrown back into the chat. That's the key thing here. Make sure you've pasted back in the final version of your script so that Claude has something to compare its own generations to.
02:14Put it in alongside this prompt, and as you can see, what it's gonna do is look for structural changes, segment level changes, style language patterns, all of this, and convert these into what I call tier one and tier two learnings. All of this with the goal, right, that the next time you sit down to write another script using Claude, it's gonna have this continuously updated context on the way that you write.
02:36So run that prompt and that will give you your collaboration report. Next up, and I still get excited that this is even possible, we're gonna ask Claude to absorb those learnings into its own training. So let me show you how to do that now.
02:48So this is the first of the two prompts that we're gonna use to update our two documents. Uh, this one is to update the style guide. So you'll simply copy and paste this in.
02:57It's going to generate another doc with the updated changes to your style guide, you can then simply re upload back into the project on Claude and it's gonna have an updated style guide for the next script. Then finally, we have the training doc update. Now once again, exact same principle, copy and paste this into Claude it's going to update those key structural changes to the training document.
03:18So once again, on your next script, it's gonna know better how to structure out script in a way that you would. Everything the AI learned about how you write in comparison to how it thought you wrote has been absorbed back into its own training document. So on the next script, it's got more context.
03:33CTAAnd as you keep repeating this process script after script, it's gonna get closer and closer to your style each time. Now the key to this process was forcing Claude to learn from its mistakes. But it's important that you're doing the same too.
03:46CTARight? Because otherwise, you're simply gonna be training the AI on bad script writing principles. So watch this video next to understand the huge mistake most YouTubers make when adding curiosity to their scripts and how you
— full transcript
§ 05 · For Joe
Feed the gap back in after every script.
WHAT TO LEARN
AI scriptwriting compounds only if the model sees the difference between what it generated and what you actually filmed — every edit made outside the chat is a lesson that never arrives.
- A style guide tells Claude what to aim for; it never tells Claude where it missed. The diff between AI draft and edited final is the missing signal.
- Tier 1 learnings (voice, sentence rhythm, vocabulary) belong in the style guide. Tier 2 learnings (segment order, structural architecture) belong in the training document. Mixing them into one document dilutes both.
- The three-prompt ritual — collaboration report, then style guide update, then training doc update — costs about five minutes per script and compounds indefinitely.
- Re-uploading the updated documents to the Claude project is the step that makes the learnings stick. Without re-upload, the next session starts from scratch.
- The system amplifies whatever writing principles it is trained on. Improve your own scripts first, or the AI learns and scales your weaknesses.
- A next-video CTA with no subscribe pitch or sponsor ask is a valid low-friction engagement signal — the absence of friction is the strategy.
