Making AI Text Talk Like You: Training Steps

Making AI Text Talk Like You: Training Steps

Brian Lv13

Making AI Text Talk Like You: Training Steps

ChatGPT generates remarkably human-like write-ups but lacks something important: a unique personal voice. While ChatGPT can compose thoughtful prose on any topic, the phrasing often rings hollow. To make ChatGPT truly your writing companion, you need to train it to write with your style, pacing, word choices, and tone.

MUO VIDEO OF THE DAY

SCROLL TO CONTINUE WITH CONTENT

Fortunately, with a few prompting techniques, you can train ChatGPT to adopt a unique writing style that matches yours.

Disclaimer: This post includes affiliate links

If you click on a link and make a purchase, I may receive a commission at no extra cost to you.

2 Ways to Teach ChatGPT to Write Like You

To get started, we’ll assume the persona of a detail-oriented author who uses a lively, engaging tone, provides detailed, accessible explanations, make liberal use of analogies, and address the reader directly throughout their text. The author also has a knack for using analogies to explain key concepts. Here’s a sample of the author’s work .

The task is to get ChatGPT to adopt the author’s writing style. Here we have three options. We can ask ChatGPT to write in a “detail-oriented, conversational manner using a lively, engaging tone, providing detailed, accessible explanations, making liberal use of analogies, and addressing the reader directly throughout their text”—much like the author. ChatGPT will oblige, but this approach will produce unpredictable results for several reasons. For example, how frequently does the author use analogies in each write-up? How informal does the author get? Does the author use light-hearted humor?

Trying to describe your writing style to ChatGPT will make the instruction way too broad. There are ways too many details that this approach wouldn’t be able to capture, resulting in a style that might not be in tune with that of the author; creating a ChatGPT prompt to deliver this information would be a challenge.

To teach ChatGPT to write like you, you can:

  1. Train ChatGPT with several samples of text you’ve written
  2. Use a Personal Preference Map (PPM)

Both options require a different approach, which you can check out below.

1. Training ChatGPT With Samples of Your Work

To train ChatGPT with samples of your work, head over to ChatGPT and use the prompt below, immediately followed by a sample of your written work that best captures your writing style.

Extract the writing style in the text provided below. Please study the tone, word choice, mannerism, sentence structure, pacing, explanation style, and other stylistic elements in order to mimic this author’s unique voice: [Paste the sample here]

After using the prompt above, you should get a result similar to this:

Result of writing Style extraction by ChatGPT-2

Of course, one sample might not be enough to thoroughly capture every aspect of your writing style. So, you can repeat the prompt above with three to five more samples within the same ChatGPT conversation. After iterating through the number of samples you wish to use, you can then use the prompt below to unify the extracted styles.

Unify all the extracted writing styles and present them in a clear detailed form. Use the tone, word choice, sentence structure, pacing, explanations, and other stylistic elements you have extracted from the different samples provided to mimic this author’s unique voice. Your instruction is to write an article on the topic: “Topic to write goes here.” Maintain the author’s perspective and attitudes while covering new subject. Write smoothly and convincingly in the author’s distinctive voice.

The prompt above should immediately apply a combination of all the writing styles extracted from all the samples to write whichever topic you provided in the prompt. It doesn’t have to be an article, it can be an email, a speech, an essay, jokes, or even a song.

When working with normal write-ups like blogs or articles, if you want ChatGPT to stick to the structure and style of a writing sample as much as possible, using one sample (and, to a lesser extent, two samples) seems to be much more effective. To do this, use the prompt below:

Pay attention to the tone, word choice, mannerisms, sentence structure, pacing, explanation style, and other stylistic elements in order to mimic the unique style of the author of the text below. Use the same stylistic elements to write an article on the topic: “Some article topic goes here.” [Paste the sample here]

Also, for the best results, we recommend using GPT-4 and, specifically, the GPT-4 Code Interpreter plugin for the task. You’ll be able to work with more text or even, ideally, stack several articles in a text or word file and ask ChatGPT to analyze the content for its writing style.

2. Training ChatGPT Using a Personal Preference Map (PPM)

A Personal Preference Map (PPM) is a key-value list of preferences ChatGPT can use to produce responses that better align with your preference. In this case, ChatGPT can extract a PPM from written samples to learn about your writing preferences and use it to replicate your writing style on demand. If you are unfamiliar with PPM, we’ve discussed it extensively in MakeUseOf’s eBook on Unlocking the Potential of ChatGPT . It is lightweight and easy to read, so do check it out.

To use a PPM, you’ll need to first extract it before using it as a prompt whenever you have something to write. To extract a PPM from samples of your work, use the prompt below:

I want you to extract a Personal Preference Map (PPM) from the data I provide in the next prompt. Now, a PPM is a key => value pair of conditions mapped to preferences. Below are examples of key => value pairs:

Tone => sarcastic, sassy, and loving

Word choice => formal, complex

Sentence structure => mixed of short and long, mostly short

Explanation style => imagery, vivid, relatable

Only reply affirmatively if you understand the task and do nothing else. When I provide the next prompt, extract the PPM using the same logic and formatting used above. The key => value pairs should be separated using “=>” Apart from the tone, word choice, and explanation style, I want you to include 10 other stylistic elements that better capture a writing style.

Using the prompt above on this article about how Auto-GPT is different from ChatGPT , we captured several stylistic elements the author uses in the article.

Extracted PPM using ChatGPT

Although the PPM approach is slightly more complex, it offers enormous attention to detail. While our previous method takes a more generalist approach to describing and applying an author’s style, PPM can get as detailed as possible, far more than anyone can easily discern at first glance.

Another advantage of using a PPM is the flexibility and portability it offers. You can easily tweak the writing style with precision by hanging a few words. You can also use the PPM in a different AI chatbot like Claude AI or Google Bard. We used the PPM above on the Claude AI chatbot and asked it to give it a topic to write. It was able to replicate as many of the stylistic elements used by the target author as possible.

Using ChatGPT PPM on Claude AI chatbot

Make ChatGPT Work for You

The beauty of ChatGPT is its versatility—with the right guidance, the AI chatbot can be taught to write in practically any style you want. If you’re tired of ChatGPT’s soulless writing style, you don’t have to settle for it. ChatGPT can do better than bland, generic outputs. With the right mix of training data, prompts, and feedback, you can transform this AI chatbot into your own writing doppelganger. Go for it.

SCROLL TO CONTINUE WITH CONTENT

Fortunately, with a few prompting techniques, you can train ChatGPT to adopt a unique writing style that matches yours.

Also read:

  • Title: Making AI Text Talk Like You: Training Steps
  • Author: Brian
  • Created at : 2024-12-03 16:49:12
  • Updated at : 2024-12-07 01:05:40
  • Link: https://tech-savvy.techidaily.com/making-ai-text-talk-like-you-training-steps/
  • License: This work is licensed under CC BY-NC-SA 4.0.