About a year ago, I was absolutely convinced I could build an AI that could write in my voice.

I gathered hundreds of my newsletters, social posts, and articles. I spent days organising everything, sorting, filtering and tagging it all nicely.

The result? Complete rubbish.

Sounded like ChatGPT with a few more swears. And ellipses. I do love ellipses…

No matter what I tried, the AI output felt generic. It might mimic a turn of phrase here or there, but it never quite captured... me.

I gave up on the whole concept. AI just wasn't ready for this particular task.

I also saw lots of people touting AI clones and the like and everything I played with was … disappointing!

Fast forward to today, and everything has changed. Funny that! Change in the AI world? Who’d have thunk?

What was impossible just a year ago is now not just possible but surprisingly straightforward. The latest models from OpenAI and Anthropic (hugs for Claude) have made genuine tone of voice replication accessible to anyone.

So even if you've tried this before and failed - now is the perfect time to try again.

Let's get started

The Game-Changer: Projects

There's one feature that's completely transformed the voice replication process: Projects.

Both Claude and ChatGPT now offer Project spaces where you can

• Upload multiple files (all that content you collected in Part 2)

• Create custom instructions that tell the AI how to use those files

• Have the model reference specific examples when generating content

This is dull but revolutionary.

Instead of trying to stuff everything into a single prompt (which never worked well!), we can now create a permanent environment where the AI has constant access to all your voice examples.

The Voice Extraction Process

Here's the straightforward process we'll follow

1. Upload all your collected content to a Project (in ChatGPT or Claude) as a knowledge base

2. Create an analysis prompt that examines the entire knowledge base

3. Use the output to create custom instructions for the Project

4. Test and refine with real content creation tasks

Let's break down each step.

Step 1: Creating Your Project

Both Claude and ChatGPT handle Projects slightly differently, so I'll cover both.

For Claude:

• Go to claude.ai/projects

• Click "New project"

• Name it something clear like "[Your Name/Brand] Voice Project"

• Click "Upload files" and add all your collected content

• We’ll create custom instructions in the next step

For ChatGPT:

• Go to chatgpt.com

• Click "+" next to Projects

• Give the project a name

• In the Configure tab, upload your files with the Add File button

• You'll create the Instructions in the next step.

Step 2: The Analysis Prompt

Now comes the magic. We'll create a prompt that analyses your entire knowledge base and extracts the patterns that make your voice unique.

Here's the prompt I've refined over dozens of projects

I need you to analyse all files in the knowledge base to create a comprehensive brand voice guide. This guide will serve as custom instructions for this project, helping you write in this exact voice in future conversations.

Process:
1. Thoroughly analyse all content in the knowledge base
2. Identify distinct voice patterns, writing style, sentence structures, vocabulary, and tonal qualities
3. Note both consistent patterns and how the voice adapts across different contexts
4. Create a structured brand voice guide with specific examples from the knowledge base

Your guide should include:

VOICE OVERVIEW:
- 3-5 sentence summary of the overall voice
- 5-7 key personality traits with evidence
- Overall writing style (formal/casual, direct/indirect, etc.)

LANGUAGE PATTERNS:
- Sentence structure preferences (length, complexity, variation)
- Paragraph construction
- Transition techniques
- Distinctive punctuation or formatting choices

VOCABULARY ANALYSIS:
- Frequently used words/phrases
- Unique expressions or catchphrases
- Words/phrases that are deliberately avoided
- Technical vs. simplified language choices

TONAL VARIATIONS:
- How tone shifts across different topics
- How tone adapts to different audiences
- Emotional range and how it's expressed

RHETORICAL DEVICES:
- Preferred storytelling techniques
- Use of questions, analogies, metaphors
- Humour style (if present)
- How complex ideas are explained

SPECIFICS TO EMULATE:
- Opening techniques
- Closing techniques
- Transitional phrases
- List/example structures
- Call-to-action approaches

CONTENT EXAMPLES:
- Include at least 10 specific quoted examples from the knowledge base that perfectly capture the voice
- For each example, explain what makes it representative

FORMAT:
- Create this as a comprehensive reference guide
- Use sections and subsections for clarity
- Include direct quotes from the knowledge base as examples
- Add notes about when/how to reference the knowledge base for specific voice elements

This guide will become the custom instructions for generating content in this exact voice. Be thorough and specific, as this will be the foundation for all future writing.

This prompt asks the AI to perform a comprehensive analysis of your content, looking for patterns at every level - from word choice to sentence structure to rhetorical devices.

If you’ve fed it enough information it’ll be quite spooky!

Step 3: Creating Custom Instructions

Take the output from the analysis prompt and refine it into custom instructions for your Project.

For Claude

• Go to your Project top level and look for this

• Click Set Project Instructions

• Copy/paste in your tone of voice guide from the prompt.

• Add any additional notes about how you want Claude to use the knowledge base (see below)

For ChatGPT

• Inside your project look for the Add Instructions button

• Paste the voice guide into the "Instructions" section

• Add any specific guidance about referencing the knowledge base (see below)

Here's a crucial addition to make to the instructions

When asked to write content in this voice, always:
1. Reference the knowledge base for similar examples first
2. Follow the patterns identified in this guide
3. Explain your approach if asked, citing specific examples from the knowledge base
4. Maintain this voice unless explicitly instructed otherwise

Add this at the TOP of custom instructions and then copy/paste in your voice guide from above. This addition ensures the AI knows when and how to use the voice guide you've created.

The Magic Combination

What we've built is essentially a specialised writing system with two key components

1. A data source / Project Knowledge (all your uploaded content)

2. Custom instructions (your voice guide)

Together, these create a powerful system that can consistently replicate your voice across any content type.

The beauty of this approach is that the AI can now reference specific examples from your knowledge base when creating new content. If you ask it to write a social media post, it will look at similar posts in your uploaded content to ensure the tone and style match.

We don’t have to put ALL the references in a mega prompt (thus filling the chat memory instantly!). Instead we have our references in place and a prompt that can tell our AI to refer to them when needed.

What's Next?

Now that we've extracted your out tone and voice created a basic system, it's time to explore deployment options. In Part 4, we'll look at different ways to implement your tone of voice system, from custom GPTs to API integration and more.

Keep Prompting,

Kyle