How to make ChatGPT sound less robotic and more like you

You spend ten minutes writing a prompt to get ChatGPT to draft an email. The output comes back. It uses words like delve, testament, and multifaceted. It sounds like a machine trying to wear a suit.

ChatGPT has a default voice. It is verbose, overly formal, and robotic. When you paste it into your email, everyone knows an AI wrote it.

You want to save time. But editing the AI output to sound like you often takes longer than writing it yourself.

There is a way to fix this. You can train AI to adopt your natural writing style without spending hours on prompt engineering.

Large language models are trained on vast amounts of internet text. When they generate text, they predict the most likely next word based on that training data. You can read more about this mechanism in the GPT-4 Technical Report.

Because the model averages out millions of different writing styles, its default output is a bland and safe tone. It relies on common structural patterns and overused vocabulary. This is why its writing lacks the rhythm and personality of a real human.

You can try to fix this by adding instructions to your prompt. You might tell the AI to follow these rules:

• Make it sound casual.
• Use short sentences.
• Do not use jargon.

This helps a little. But the AI often falls back on its default habits. Or worse, it overcompensates. It might become too casual and sound unprofessional. Describing your style with adjectives is rarely enough to override the default behavior of the model.

The most effective way to change AI output is to provide examples of your own writing. This technique is known as few-shot prompting. Providing examples significantly improves the ability of the model to match a specific tone and format. You can see the research on this in the paper Language Models are Few-Shot Learners.

Instead of telling the AI how to write, you show it. You give the AI a sample of an email you wrote in the past. Then you ask it to write the new email using that exact tone.

Here is an example of how to structure your prompt:

Here is an example of my writing style.

[Insert your past email here]

Now, write a new email about the upcoming project deadline using this exact tone and style.

The model analyzes the sentence length, vocabulary, and rhythm of your example. It then applies those patterns to the new text. The result sounds much closer to your actual voice.

The limitation of few-shot prompting is friction. Doing this for every single email or document is tedious. You have to find a good example, copy it, open ChatGPT, paste it, and write the prompt. Then you have to copy the result back to your original app. This breaks your workflow.

This friction is why we built WordPolish. WordPolish is a system-wide AI writing tool for macOS. It removes the need to copy and paste between apps.

You highlight text in any app, like Gmail or Slack. You trigger WordPolish from the menu bar or with a keyboard shortcut. The app uses your saved writing traits and the surrounding context of your document to rewrite the text in place.

You review the changes in a diff overlay. You can apply the rewrite immediately or ask for another version. It keeps you in your flow and ensures your writing always sounds like you.

AI should adapt to your voice, not the other way around. By using your own writing as a baseline and relying on tools that integrate into your workflow, you can stop sounding like a robot and get back to sounding like yourself.