GOF Quick Wins - How to Create a GPT That Thinks and Writes Like You

I'm Damon Nelson from GeekOutFridays, and if you're anything like me, you love what AI can do — but you hate how generic it often sounds. In this post I’ll walk you step-by-step through a simple, no-code method to train a GPT to sound and think like you.

This approach uses content you already own (emails, blog posts, transcripts), a couple of minutes of setup, and OpenAI’s custom GPT interface. I’ll also give you my free set of 10 knowledge prompts to jumpstart the process (you can find the Notion guide here: https://rssmasher.notion.site/10-Knowledge-Prompts-to-Create-a-Personal-AI-Assistant-2547123fca208029a43ef6fbdb16e54e).

Why your AI doesn’t sound like you (and the simple fix)

Here’s the core problem: you give AI short commands and expect it to know how you think. When you type “write an email about X,” the model defaults to generic patterns it has learned across millions of users. It doesn’t know your voice, your point of view, your favorite analogies, or the way you structure a sentence. That’s why replies feel bland, safe, and—let’s be honest—forgettable.

The fix? Train the AI on your own writing so it learns your patterns. Instead of telling it what to say from scratch, you give it examples of how you already say things. The result: responses that sound like they came from you, not from a factory reset assistant.

The three-step methodology (what to do in five minutes)

There are three simple steps. I promise they’re practical and repeatable:

1. Collect pre-AI writing: blog posts, emails, transcripts, long-form bios.
2. Upload those text files to your GPT knowledge bank or custom instructions.
3. Use a short, consistent instruction (example: “Act like my brand’s content manager and write a social post about {topic}”).

That’s it. The upfront work takes a few minutes to gather files and a little longer to upload them. I once pulled content all the way back to 2017 and had a working GPT in about 15 minutes. But for most people, the basic setup is faster.

Step 1 — Collect your writing (the gold mine)

Not all content is created equal for training voice. The best sources are pieces you wrote before you started using AI—that’s pure, unfiltered you. Here’s what to gather:

• Blog posts you wrote manually (longer pieces are better).
• Email sequences or solo emails where you explain things in your tone.
• Video transcripts from unscripted or lightly scripted videos (I use my GeekOutFridays livestream transcripts because I don’t script those).
• Author bios, back-of-book blurbs, and website About pages.

Put all these as text files into one folder. Keep them simple (plain .txt or .md). The goal is to create a compressed knowledge base that represents your voice, favorite phrases, and typical structure.

Step 2 — Build your GPT and upload the knowledge bank

I recommend using OpenAI’s custom GPT builder for this. The interface allows you to:

• Give the GPT a name and short description.
• Upload files into a knowledge bank (or add custom instructions).
• Add conversation starters and suggested inputs for users.

How I do it: I create a new GPT, name it (e.g., “Damon AI”), then upload the entire folder of text files into the knowledge bank. The upload process is as simple as selecting files and confirming. Once uploaded, the GPT can reference those files as if they're part of its training context. This is where the magic happens—now the model has the patterns that make your writing distinct.

What to include in the GPT settings

Keep the starter instructions short and practical at first. You can later expand with guardrails and longer guidelines, but start with something like:

"Act like my brand's content manager and write a social media post about {topic}."

Then add conversation starters so people (or you, later) know how to prompt the GPT. Example suggestions:

• “Input your main topic to get the best response.”
• “Try: ‘Write a short Facebook post about launching a new course’.”
• “Ask: ‘How would I write this?’ to get it to match my voice.”

Step 3 — Use short, smart prompts (and a fun trick)

Once your GPT is trained on your content, you don't need elaborate prompts. Give it a simple role and task, and the knowledge bank does the rest. My go-to is:

"Act like my brand's content manager and write a social media post about {topic}."

That’s concise, repeatable, and scalable. The knowledge bank supplies voice and structure, while the instruction supplies purpose.

Here’s a neat creative trick I use: tell a story that mirrors the emotional tone of the topic instead of explaining the topic literally. For example, to write about launching Facebook ads for the first time, I might prompt:

"Make a Facebook post about diving with sharks related to running Facebook ads for the first time."

Why this works: the shark-cage metaphor lets the model pull from your storytelling style and emotional cadence. The result is an engaging social post that’s uniquely yours—narrative-driven instead of robotic bullet points.

Practical tips, extra features, and sharing

Small touches elevate the experience:

• Images: I create quick cover images for posts using DALL·E 3 through the GPT builder to make demo cards or thumbnails. It pairs nicely with your social post output.
• Conversation starters: Offer sample prompts so teammates know how to get useful output immediately.
• Sharing options: You can keep the GPT private, share a link with specific people, or submit it to the GPT store to distribute or monetize it.

Real-world example: turning fear into a story

Let me walk you through a quick example I used during testing. I asked the GPT to make a Facebook post that compares jumping into Facebook ads for the first time to diving with sharks. The output borrowed this structure from my transcripts and blog writing:

• Open with a visceral image (descending into a cage).
• Connect the fear to the practical risk of ad spend.
• Offer a small, tactical next step (start small, test, iterate).
• Close with a personal sign-off that feels like me.

The result reads like a short narrative with a moral—exactly the kind of copy I’d write myself. And it’s faster than drafting from scratch.

How to iterate and improve

After you have your GPT up and running, here’s how to refine it:

1. Run a handful of prompts and compare the output to how you would write the same piece. Note the differences.
2. Add targeted samples to the knowledge bank (e.g., a few high-quality email examples if you want better emails).
3. Tweak the short instructions to add constraints or tone directives (e.g., “short, witty LinkedIn post” or “long-form newsletter at 800 words”).
4. Create multiple GPTs for different use cases—one focused on social, another on long-form blogs, and another on customer support replies.

Over time, your GPT becomes a reliable content assistant that saves you time and keeps your voice consistent across channels.

Free resource: 10 Knowledge Prompts

I promised a free resource. Grab my list of 10 knowledge prompts that act like cheat codes for teaching an AI how to behave as your assistant. You don’t need to sign up for anything; the guide lives on Notion for quick copying and use. Here’s the URL again if you want to try it now: https://rssmasher.notion.site/10-Knowledge-Prompts-to-Create-a-Personal-AI-Assistant-2547123fca208029a43ef6fbdb16e54e

Wrap-up and final thoughts

Training a GPT to sound like you is straightforward and powerful. The steps are simple: collect your authentic content, upload it into a GPT knowledge bank, and use short role-based prompts to get outputs that reflect your voice. This approach scales: you can automate social posts, generate first drafts of blogs, create email sequences, and even personalize customer replies—all while preserving the human flavor that makes your brand unique.

If you want help or want me to send the Notion prompts, leave a comment on the original post and I’ll share the link. But more importantly—start with your own content. The more real, pre-AI writing you give the model, the truer it will sound to you.

Thanks for reading. I’m Damon Nelson with GeekOutFridays—let’s make AI sound like you.