If you want to train ChatGPT to write like you, you’re in luck.
It is 100% possible to feed ChatGPT your writing samples and have it output something similar. This can result in ChatGPT writing content like a human and creating text that is undetectable by AI detectors.
However, due to certain limitations with ChatGPT, you need to be methodical in how you approach this.
In this post, I’ll explain exactly how to make ChatGPT write in your style.
I’ll also share the pitfalls you need to be aware of, and a few clever hacks that let you get around the annoying token limit.
Let’s get started.
How To Train ChatGPT To Write Like You
To train ChatGPT to write like you, follow this 6-step process:
- Choose your writing samples
- Create writing guidelines based on your samples
- Add your writing guidelines to Custom Behavior
- Get ChatGPT to generate your writing
- Refine ChatGPT’s output
- Consider the token limit (Optional)
1. Choose Your Writing Samples
The first step is to gather your writing samples.
The samples should be in the same format and serve the same purpose as the piece of writing you want ChatGPT to create.
For example, If you use some emails as a sample of your writing but want ChatGPT to write a blog post, you probably won’t be happy with the results.
If possible, collect multiple samples. The more ChatGPT has to work with, the better.
2. Create Writing Guidelines Based On Your Samples
Now, it’s time to generate writing guidelines based on your samples.
To do this, we’re going to feed your samples to ChatGPT using this prompt:
You are an expert at analyzing writing samples from a specific writer and generating a concise and actionable style guide based on those samples. The guidelines should address all elements of writing, including overall structure, sentence structure, sentence length, paragraph length, vocabulary, transition words/phrases, punctuation, purpose for writing, and all other elements that define an individual’s writing style.
Take the following samples of [type of writing] and generate a style guide based on them:
- Sample 1: [Paste sample]
- Sample 2: [Paste sample]
- Sample 3: [Paste sample]
Output the style guidelines in bullet point format as concisely and directly as possible.
3. Add Your Writing Guidelines to ChatGPT’s Custom Behavior
To make it easy to generate as much content in your writing style as necessary, we’re going to use ChatGPT’s custom behavior setting.
This enables you to start new chats without having to input your guidelines each and every time.
To input new custom behavior into ChatGPT, follow these steps:
1. Click on your profile icon in the bottom left corner of the ChatGPT interface.
2. Click on the “Custom instructions” button that appears below the toggle.
3. In the box that asks what ChatGPT should know about you to provide better responses, share some relevant background info on why you’re writing this content.
4. In the box that asks how you want ChatGPT to respond, paste your writing guidelines.
5. Click on the “Save” button to confirm your custom instructions.
As long as the “Use Custom Behavior” box is toggled on, you’ll use these guidelines automatically whenever you start a new chat.
4. Get ChatGPT to Generate Your Writing
Now that your style guidelines are in place, you can get ChatGPT to write like you.
We’re not going to solely rely on your writing guidelines. They’re just an extra quality assurance feature.
The first step is to start a new chat. This allows ChatGPT to use the guidelines you added to Custom Behavior.
Next, use the following prompt to re-feed ChatGPT your writing samples. This is necessary because this instance of ChatGPT doesn’t have access to the instance in which you previously provided the samples:
“Analyze the unique style of the provided writing sample below. Using the information extracted from the sample, write a ___-word (length) ________ (type of work: e.g., essay, blog post, short story) about ________ (topic).
Your task is to:
- Extract and Emulate Style Elements: Analyze the tone, word choice, mannerism, sentence structure, pacing, and explanation style of the sample. Replicate these elements in the new piece.
- Aim the Piece At: ________ (target audience: e.g., teenagers, tech enthusiasts, historians).
- Maintain Rhythm: If the original piece used a mix of short, medium, and long sentences, ensure to retain a similar rhythm.
- Incorporate Specific Techniques: If the original sample used humor, metaphors, analogies, or any other specific technique, incorporate them where appropriate in the new piece.
- Represent Multiple Perspectives: Consider the views of both ________ (perspective 1: e.g., an advocate, a tech lover, an expert) and ________ (perspective 2: e.g., a critic, a traditionalist, a novice).
- Iterative Drafting: Use the iterative approach to enhance your drafts. After each version, critique your work, assign it a score out of 10, and if the score is below 9, refine the previous draft. Do this until you achieve a score of 9 or 10. In each iteration, edit to make the content more similar to the writer’s style.
- Maintain Authenticity: While emulating the style, retain clarity in your explanations, employ open-ended prompts for brainstorming, and infuse some originality to the text.
Writing Sample for Analysis: [Insert the writing sample you want analyzed here.]”
ChatGPT will generate a writing based on your samples, and you can revise it further as needed.
5. Refine ChatGPT’s Output
Even with all of your prep work, it’s likely ChatGPT won’t produce something that 100% sounds like you.
You can start by getting ChatGPT to try again. If it’s a shorter response, you can simply ask ChatGPT to “Try that again.”
If it’s a longer response, the smart move is to re-submit the initial prompt so you don’t use extra tokens.
Here’s the process for that:
1. Click the “edit” button in the top-right corner of the message that generated the sample.
2. Scroll to the bottom of the message and click “Save and Submit.”
Regenerate the content a few times and pick the result you like best.
However, even the best result will likely need additional editing. You can do this by asking ChatGPT to revise certain aspects of the prompt, or you can edit the writing yourself.
6. Dealing With the Token Limit
The most frustrating part of training ChatGPT to write like you is the token limit, which caps how much you text ChatGPT can process and generate.
The token limit in GPT-3.5 — the free version — is 4,096, which is about 3,072 words.
GPT-4 is a little better, with a token limit of 8,192, or about 6,144 words.
That might sound like plenty — until you realize your initial prompt, your writing samples, and ChatGPT’s output all have to fit in that word count.
If you want to avoid the token limit issue altogether, consider using Claude.ai — a free generative AI similar to ChatGPT that has a limit of 100,000 tokens. That massive limit should remove any issues with sample or output size.
However, its outputs often aren’t as good as GPT-4, and you need to be located in the United States or Canada to access it. (Though you can also spoof your location using a VPN.)
If you’re set on dealing with ChatGPT’s token limits, you’ll need to write your content in phases.
For example, if you want ChatGPT to generate a long blog post — and your samples are similar long blog posts — it might be smart to first train ChatGPT on your blog post introductions and have it generate the intro, then do the same for your body content and conclusion.
To figure out how close to the token limit you’ll be, paste your writing samples, custom behavior, and prompts into the OpenAI tokenizer.
Getting ChatGPT to write like you is a great way to use this amazing technology.
However, there are many more ways to leverage ChatGPT — and the many other amazing AI tools coming out — to increase your productivity.
To stay up-to-date on the latest advances in AI and ensure you don’t miss out on game-changing new tools, join our email newsletter.