Your Scriptwriting Partner

If you've tried using AI to write YouTube scripts and found the outputs are rubbish - and that it can even take longer - listen up.

Because the truth is, AI is really good at completely ruining scripts.

And it's not because it's "not good enough yet."

It's because the way you're asking it to generate scripts is broken.

But I've spent the last four years systemising my scriptwriting process for hundreds of students and clients, and I now use a hybrid AI / human method to write scripts in under two hours.

So... here's why AI is ruining your scripts, and how to fix it.

The AI-mistake almost every YouTuber makes:

You might think that after giving ChatGPT/Claude a few examples of “gold standard” scripts + a personal writing style guide...

...you can simply ask it to generate full scripts from scratch.

And while having a style guide + past script examples are absolutely necessary for writing scripts with AI, they’re not sufficient.

Because writing a full script involves a tonne of different, complicated, interlocking steps.

The AI has to interpret your notes, build a structure, expand every sentence in your voice, check that expanding those sentences doesn't break the structure, audit your hook, check for stakes… all at the same time.

It’s too much. And even AIs with higher token counts like Claude and Gemini will struggle to do these tasks to a high standard.

The solution, then, is to create your scripts in stages with AI.

Here’s the process I’ve been following with Claude recently...

Stage 1: Brainstorm collaboratively

Asking AI simply to "create the structure" for a video leads to disaster.

I’ve found AI consistently creates the "most logical" sequence possible… which might seem ok at first, but rarely holds up to closer inspection.

So instead of asking AI to hand you a complete structure, have a conversation:

Keep it flexible and free-flowing... there is no "right" way to do this part.

• Pass the AI your titles ideas.
• Discuss your core idea, and the different angles of approach you could take.
• Be honest about any blindspots or uncertainties.

Gradually, you'll settle on your angle of approach, and find your way towards a mutually agreed-upon structure.

Let the structure emerge from that back-and-forth, rather than a single prompt.

Stage 2: Write one “phase” at a time

This is the same principle I teach for writing scripts manually.

Writing a whole script from start to finish is overwhelming, and the quality suffers without a step-by-step system.

That’s why (as you may know) I always start by writing just the payoffs, word for word.

Then the setups. Then the tension. Then the hook and CTAs.

With AI, your approach should be no different.

Ask it to generate just the payoffs at first, then provide feedback.

Are they in a logical order? Are they individually compelling? Are they too similar to each other?

Once those are locked, move to the setups. Then the tension. Then the hook and CTAs.

This gives you the continuous ability to course correct… before the AI goes off and writes something insane.

Now you have your V1 script… but it’s not going to sound right just yet.

Stage 3: Rewrite segment 1, then ask AI to learn from your edits

This is where it gets really powerful.

Rewrite segment 1 to your liking - after all, the AI won’t have written it perfectly.

Then paste your re-write back into the chat, and ask AI to rewrite segment 2 while focused on:

1. What it learned from your segment 1 re-write.
2. Your style guide.

The second point is extremely important.

I’ve found, even with my style-guide pre-loaded into my scriptwriting project on Claude, it forgets to adhere rigorously to it as the chat gets longer.

Repeat this process across the entire script - re-write segment 2, and tell the AI to learn from it when re-generating segment 3.

Etc, etc.

Each subsequent segment will get closer and closer to your style.

Stage 4: The final pass

Asking AI to do a "final check" of the whole script is, once again, too broad.

You're basically asking it to "unbake the cake" and check whether all the frameworks, structures, tonal references (etc) are up to standard.

In my experience, this leads to the AI making bad suggestions.

So accept, if you can, the need to use your human brain for this part.

Read the script back to yourself, out loud if possible, and fix anything that sounds off. After everything you've just done, this part shouldn't take long.

Not only does this give you final oversight on the script...

...but it means you'll always provide the AI with a little more data about your writing preferences.

Data you can then ask it to learn from next time you write together.

In an upcoming newsletter, I’m going to give you my exact system in Claude for capturing that data and feeding it into your next script.

That's all for this week.

Any questions? Reply to this email and I'll get back to you.

Speak soon,
George 👋

‍

When it comes to AI, I go through phases of burying my head in the sand.

Everything is moving so quickly, and it sometimes feels like my AI knowledge is out of date the moment I learn something new.

But recently, I've started to find a nice balance of integrating AI into my scriptwriting workflow - without it feeling overwhelming.

And it's helping me write scripts faster than ever.

As I write this email, I'm on a train to Amsterdam for a long weekend away with my partner.

Which meant this week, I had just 3 days at home to write and record two videos for my editor.

Videos I had not even ideated at the start of the week.

Here's how I managed it, using AI to speed me up.

1) 1of10

It’s much easier to dive into writing a script when you can already see the packaging.

I’ve spent the last two weeks generating a batch of new thumbnail ideas using 1of10’s thumbnail generator.

Here are some of my favourites (all generated in under 60s):

We all know the advice is to “create packaging first”.

But in a busy week, it’s not practical to wait until you’ve completely finished a thumbnail before starting the script.

With 1of10’s generator, you get the benefit of seeing an almost finished thumbnail in literally 60 seconds - and that’s where the value truly is.

With 1of10, you can:

• Describe any thumbnail you can imagine and it will re-create it in your channel’s style.
• Physically sketch an idea and it will do the same.
• Give it a reference thumbnail from any channel, and it’ll create your version of that thumbnail.

I had to move fast this week... and having my packaging ready to go by 9:30am on Monday morning got me off to a great start.

Want to give it a try? Get 20% off a 1of10 subscription here:

Get 1of10 (20% Off)

2) “Live” Style Guide

Every week, I give feedback to my community students.

This feedback constitutes my most up-to-date advice.

So, for the last 2 weeks, I’ve been passing all my feedback into a continuous training document.

At the end of each month, I ask Claude to create a summary of my recent advice to my students, followed by:

1. Updating my current style guide.
2. Creating 5 YouTube script ideas based on that advice.

So... think about where you’re already writing/thinking about your area of expertise each week:

• On Zoom calls.
• In a newsletter.
• In Discord Groups / Group Chats.

Transform that knowledge into a continuous stream of new ideas and pre-written chunks of your next script.

3) Use "Projects".

This is basic, but most YouTubers still don’t do it.

If you’re already using ChatGPT or Claude (I recently switched to Claude btw), stop wasting time explaining what you need it to do each time you're using it to help with scripting.

Instead, create a project and pre-train it with:

1. Your personal style guide.
2. A detailed breakdown of its purpose (to write YouTube scripts).
3. Any additional materials you think it needs to write well (e.g. I upload the full 36,000 word PDF of my course to any scriptwriting-related project).

4) Ask AI to create a report on itself.

If you've used AI in any way to help structure or draft your scripts, always provide it with the final version of the script once you've made your edits.

Then ask it to create a report on itself, identifying the differences between what it generated and what you ultimately wrote yourself.

Then use that report as training data for your next conversation with it.

I've found Claude gets 10% closer to my exact style and structure with every script.

I'll speak more about this process in the coming weeks.

That's all for this week.

And, if you're interested in trying 1of10's new thumbnail generator (which continues to blow my mind), don't miss out on our exclusive 20% discount:

Get 1of10 Now (20% Off)

Any questions? You can to reply to this email and I'll get back to you!

Speak soon,
George 👋

In December, a student of mine was about to launch a new channel.

Tim had three videos scripted, filmed, and ready to launch (following my “Whirlpool” strategy).

But, days before hitting publish… Tim’s computer died.

The scripts were safe in the cloud - but the footage was lost.

So today, I want to share how he turned this apparent disaster into an amazing opportunity…

…and the important lesson he taught me about scriptwriting.

Success from Disaster

So the videos were gone... but the ideas were still top of mind for Tim.

After all, he’d just spent weeks writing and recording them.

So rather than wallowing in misery (like I would have done), he got curious.

He decided to record some off-the-cuff podcast episodes covering similar topics.

This stripped away some of the more heavily scripted elements and allowed him to experiment with riffing.

And let me tell you something - I am secretly (although not actually) a teeny bit glad this all happened.

Because the resulting podcast is fantastic.

It’s called: “The Fiction Writer’s Toolshed: Practical Tools for Authors”

Check it out here!

Technically, it's aimed at fiction authors.

But every episode is full of wisdom for any type of writer - and that includes YouTubers.

I'll finish today's newsletter with my favourite lesson from Tim's podcast so far.

But first, let's take a moment to consider Tim's response to this tech disaster, and how we can apply it to our own YouTube videos.

Ask yourself this - how many times have you recorded an entire YouTube video only to find the mic wasn’t switched on, or the focus was playing up, or even that you never hit record?

I’ve experienced all these and more - and the funny thing is, the next take is nearly always much better.

Practice makes perfect, and - albeit unintentionally - mistakes like these force us to practice.

And this was a heck of a lot more than "forgetting to hit record".

Tim had put hours into filming, editing, and preparing for publish.

But who would bet that when he re-records these same YouTube videos...

...they'll turn out way better the second time around? ☝️

Technical disasters are often blessings in disguise.

Here’s how you could apply this thinking to your next video:

• Experiment with an alternate format - a podcast, a Short, or even a newsletter - and see whether this changes how you express your thoughts on the topic. Then rewrite the long-form piece.
• Allow yourself a “trial run” during each recording - hit record just in case, but work on the assumption that you’ll record the whole thing at least twice.
• Discuss a topic with a friend, and record the conversation - and notice the objections or questions your friend has. This could totally change how you approach the script.

Scriptwriting Lesson - Farming vs Cooking

So what have I learned from The Fiction Writer's Toolshed?

Well, a lot! But here's my favourite takeaway so far:

This is from episode 5: “Fiction Farming vs Fiction Cooking” (listen here).

Tim says:

“Writing is not the same thing as coming up with ideas. They are not the same process.”

What Tim’s describing are the two different “modes” our brains enter when writing.

In associative mode, you’re “farming”.

You’re researching ideas and throwing them into a document, without knowing exactly how they’ll fit together yet.

In analytical mode, you’re “cooking”.

You’re combining those ideas, narrowing your options, and working towards a final script.

The problem Tim identifies is this:

“As writers, we try to cook while we’re still farming […] judging seeds that could grow into incredible ideas by the standards of a finished meal. If you do that long enough, writing starts to feel bad.”

This resonated strongly with me.

How many times have you opened a new script full of excitement, but quickly found yourself stressed out and unable to make decent progress?

And the longer this goes on, the more you start to believe you’re just a bad writer.

This is part of the reason I switched my scriptwriting approach - using pre-existing material from my newsletters, and plugging in my setups and payoffs in a methodical, repeatable way.

That’s farming.

Then I switch to my analytical brain and connect each setup to each payoff using bullet points - and often write the hook later once I’m sure all ideas have been gathered and arranged.

That’s cooking.

Now, if the way I explained this feels a bit overwhelming, go check out The Fiction Writer’s Toolshed.

Not only does Tim have an incredible radio voice, but each episode explains a writing problem + solution simply yet elegantly - and you won't find it difficult to see how his ideas connect to YouTube.

Thanks for letting me share this, Tim!

That's all for this week.

Any questions? You can to reply to this email and I'll get back to you.

Speak soon,
George 👋