🚨 Quick announcement: I’ve been working on a new course and I’m excited to share it with you next week. Watch out for an email from me!

Hey Warrior,

Most people don’t think of using multiple AI models when writing content.

But with so many different AI language models out there today, it pays to tap into the unique strengths of each.

Every model has a distinct tone and style that suits different needs when writing with AI.

In this post, I’ll share my process for writing with AI by leveraging the unique strengths of each model.

Here’s what we’ll cover:

• Step 1: Create Your Base Prompt

• Step 2: Test with Multiple Models

• Step 3: Regenerate and Refine

• Step 4: Compile and Curate

Read time: 5 minutes

🔧 Tool of the week: Guidde

Guidde is an AI-powered screen recording tool (think Loom on steroids).

You simply record your screen and then AI will automatically turn that recording into a FULL step-by-step tutorial!

I can then share this how to guide with anyone, which makes it really easy and fast to demo a workflow or show a process.

Try it out for free here.

*sponsored content

Step 1: Create Your Base Prompt

First, you’ll need to have a strong base prompt that clearly explains the content you want the AI to generate.

As I've covered in my previous post, I often use one-shot prompting when I have a very clear idea of what I want to write.

So in that case my base prompt would look something like this:

You are a world-class …

Your task is to rewrite the provided [asset] below, using the specific context I give you.

[Enter your CONTEXT. This can be as detailed as you want and have multiple sections.]

[Paste the ASSET, i.e. landing page, email, video script etc]

If I don’t have an example (i.e. asset) at hand to be able to do one-shot prompting, the best way is usually:

1. Write a short draft of whatever it is you want to write about (can be bullet points).

2. Create a prompt that converts the draft into a polished version.

3. Use the 3 writing adjustments I talked about in this post.

The prompt I use in this case goes like this:

#INSTRUCTIONS: 

You're a professional blog post editor. Please assist me in converting the CONTENT DRAFT I've provided below into a final state. My CONTENT DRAFT is an unfinished draft. It contains some sentences that are close to final, which may only need some proofreading. And it contains some lines where content is still missing and needs to be completed by you (often denoted by square brackets).

Use these guidelines for your writing style: 
- Formality: 3/10 (1 = casual, colloquial language suitable for informal settings; 10 = highly formal, structured language appropriate for official or academic contexts) 
- Sentence Structure: 3/10 (1 = predominantly simple, short sentences, easy to follow; 10 = extensive use of complex, compound sentences, potentially challenging for average readers) 
- Word Choice: 3/10 (1 = basic, straightforward vocabulary, easily understood by a general audience; 10 = specialized, technical language or sophisticated lexicon, more suited for expert readers)

#CONTENT DRAFT

[Enter your draft]

Step 2: Test with Multiple Models

Once you have a solid base prompt, plug it into at least two different models to start. I usually go with ChatGPT and Claude first.

The goal here is to take advantage of each model's unique strengths.

For example, I've found ChatGPT usually makes stronger attention-grabbing sentences, while Claude has a more natural human tone.

Here, I used both Claude and ChatGPT with the same one-shot prompt. But you can see that the outputs are vastly different.

Claude wrote in a very natural, conversational tone:

Whereas ChatGPT chose a pretty formal tone in this case:

In this case, I definitely liked the Claude output more.

For further exploration, you could also experiment with Bing (in creative mode) or Bard. However, I tend to find that with writing, those models never beat ChatGPT or Claude (so far).

For those seeking even more options, Perplexity (labs.perplexity.ai) also offers a handy tool for quick, free access to open-source models like Llama and Mixtral.

Step 3: Regenerate and Refine

The next step is to iterate a bit further on the outputs from each model and identify the versions closest to your goals.

In the example above, I really liked Claude’s first try, but I wanted certain sections expanded and other reduced.

So I followed up with this prompt:

The ChatGPT version was pretty far away from what I wanted. So I used my lever prompt to try and transform the tone into something sounding more conversational.

The key is to know when to stop iterating.

In this case for example, I would have had to iterate on the ChatGPT output for quite a while, because it was still so far off from what I wanted.

But the Claude output was already so good on the first and second try that I didn’t need to do that.

So I saved myself a lot of time, by using both models and not just one.

Step 4: Compile and Curate

As a final step, I’ll copy my 2-3 favorite outputs into a single document.

Now comes the editor's job - piecing together the best parts from each output to build your finished text.

Think of it like a puzzle - you want to pick out the paragraphs, sentences, and phrases from each model that resonate with you the most.

In this case I took around 70% from Claude, 10% from ChatGPT, and 20% were my final manual edits.

Other times it might be the other way around, 70% ChatGPT and 10% Claude.

It always depends on the type of text.

Wrapping up

Leveraging multiple models allows you to tap into the unique strengths of each for tone, style and optimized content.

With the right prompting and curation, it can help you write high-quality content faster than relying on a single model.

Test it out and let me know how it works for you!

Thanks for reading!

If you enjoyed this email, consider forwarding this newsletter to a friend or colleague.

See you in the next one!

Moritz

P.S. Whenever you’re ready, here are two ways I can help you:

1. Get your product in front of 20,000+ solopreneurs, business owners and professionals. Sponsor this newsletter here.

2. Do you need help with an AI-related problem you’re facing? Grab some time with me here for an AI consulting session.