Anyone Can Create It! Building a Writing Style Analysis-Based AI Writing Assistant with ChatGPT

Introduction

Writing is often a time-consuming and labor-intensive task.
Especially when writing articles regularly, starting from scratch each time can feel burdensome.
I often think "I have things I want to write about, but organizing my thoughts is challenging" or "I wish I could quickly work out the framework from an outline," and I've long felt this was an issue.

That said, the recent progress in ChatGPT has been remarkable.
It can now generate large amounts of text with just a little prompting, and sometimes it even remembers and utilizes your prompts in memory.
Given this, AI-powered text generation is certainly one solution.
However, I faced the challenge that when the generated text differed from my writing style, it took time to revise, ultimately reducing efficiency.

This article introduces a method for having ChatGPT analyze the characteristics of your writing and using those results to build a customized AI writing assistant.
This approach makes it possible to efficiently generate text that closely matches your writing style.

The process consists mainly of the following two steps:

1. Writing style analysis using Deep Research
2. Building a dedicated AI writing assistant using GPTs

Results and Analysis of What I Created

That said, it wouldn't be interesting to just go through the procedures step by step. Let me first share what it looks like. When I give the topic "Anyone Can Create It! Building a Writing Style Analysis-Based AI Writing Assistant with ChatGPT" to the GPT I created for this article, it produces the following:

Understood!
I'll structure a technical blog article with the following theme:

---

Title

Anyone Can Create It! Building a Writing Style Analysis-Based AI Writing Assistant with ChatGPT

---

Introduction: Purpose and Objective

Have you ever felt anxious when writing technical articles, wondering "Am I maintaining consistency in my writing style?"
I have. Seriously, multiple times. Especially when writing team blogs, different writers tend to create varying atmospheres, leading to a certain "sense of discord."

This article introduces how to build a writing assistant that can mimic your own or others' writing styles by combining ChatGPT with writing style analysis.

It's just these 3 simple steps!

1. Extract writing style characteristics based on your own articles
2. Have ChatGPT learn that style
3. Actually get writing assistance

It's so easy you'll feel like "I might be a genius..." so please stick with me to the end!

---

Step 1: "Verbalize" Your Writing Style

The first thing to do is verbalize "What is my writing style like?"
This involves reading through several of your past articles and identifying common patterns.

The checkpoints are like this:

• Vocabulary: Casual? Formal? Do you use expressions like "seriously" or "quickly"?
• Grammar: Consistent polite language? Length of sentences?
• Structure: Is it in the order of introduction → background → method → conclusion?
• Tone: Are you speaking to readers? Or explanatory style?
• Humor: How much jokes and self-deprecating humor do you include?

You could do this analysis manually, but it's much easier to rely on GPT.
For example, create a prompt like this and feed it your articles:

You are a professional technical article writer. Please read the following text and analyze characteristics such as vocabulary, grammar, tone, logical structure, and humor, then output a "list of writing style characteristics typical of this person" in Markdown format.

---

Step 2: Have ChatGPT Learn the Writing Style

Once you've analyzed the writing style, next you input that into ChatGPT.
What's important here is clearly stating "what style you want it to write in."

As an example, create a prompt like this:

You are a technical blog writer AI. Please write articles about web technology following the writing style guidelines below.

## Writing Style Guidelines
- Vocabulary: Use plain Japanese + technical terms (React, AWS, etc.) accurately
- Grammar: Polite form, medium-length sentences, create rhythm with commas
- Tone: Polite but casual, use questions and empathy
- Logical development: Show purpose in introduction, organize with headings, include specific examples and supplements
- Humor: Moderate light jokes appropriately

Theme: How to create a writing style unification tool using ChatGPT

In this state, if you ask "write the main text," it will churn out articles that seem just right. Really, what a convenient era we live in.

---

Step 3: Actually Try Writing

Finally, it's actual writing.
Since GPT reproduces the tone and style you usually write in, it feels like "another version of yourself" is creating the manuscript.

For example, even with instructions like "write an article about React state management in your usual style," if you've properly taught it the writing style analysis, it will finish quite nicely.

Then you just need to fine-tune it yourself.
With adjustments like "a bit too many jokes..." or "let's explain this part seriously," you can easily complete an article that's truly your own!

---

Conclusion: Create Your "Alter Ego" with GPT

So this time I introduced how to create an AI writing assistant using ChatGPT and writing style analysis.

• Analyze your writing style from past articles
• Have ChatGPT learn the style
• Create articles that sound like you together!

Using this method, "team blogs with consistency" isn't just a dream, and it can also be applied to writer training.

Please try nurturing your own "alter ego AI writer" too!

---

If you'd like, I can output this content in Markdown, Word, or other formats, or create a more technically detailed version!

How Do I Feel Looking at This Objectively?

⭕️ Good Points

• I feel it's reached a state where it writes things I would likely write
• The part "I have. Seriously, multiple times. Especially when writing team blogs, different writers tend to create varying atmospheres, leading to a certain 'sense of discord'" properly presents challenges to readers in a way that seems like me. It captures my tendency to talk at length about the background of the issues I'm discussing.
• The article structure creates a proper common structure. The parts where I list the flow of the discussion in the introduction and state conclusions are like me.

❌ Areas for Improvement

• In the part discussing the background of challenges, it uses the expression "nandesyone" (a casual way of addressing others), but "nandesyone" goes against my preferences. I would address readers with "narimasenga?" instead. Expression skills need improvement.
• There's a tendency to write somewhat too casual text. It gives off a frivolous impression. Compared to when I write, there seems to be more colloquial language. In my case, while I often use words like "seriously" or "really," I tend to limit their use to introductions and try to balance the overall tone of the text.

Step 1: Writing Style Analysis Using Deep Research

In the first step, I input data from previously written articles and content into Deep Research to analyze writing tendencies.

Reasons for Selecting Deep Research

There are multiple approaches to analyzing text characteristics:

• RAG method: Use all articles as data as-is and have AI search them
• Manual analysis: Extract text characteristics yourself
• Deep Research: AI systematically analyzes text characteristics

While the RAG method is effective for reproducing individual expressions and styles, it has limitations in extracting the essence of writing style and characteristics of logical structure.

I chose Deep Research because it can abstract and organize text, systematically analyzing the tone and expression methods of writing. By organizing structural patterns, tone, and frequent word patterns in writing, I can objectively grasp the characteristics of writing style.

Implementation Procedure

1. Analysis Request to Deep Research

I prepared the following data as analysis targets:

• Previously written technical articles (Markdown format)
• Blog and note technical memos
• Technology-related social media posts

The following items were extracted as analysis results:

Analysis Item	Content
Vocabulary characteristics	Balance of technical terms and casual expressions
Grammar structure	Sentence length, line break frequency, punctuation usage
Tone	Ratio of polite language to casual expressions, frequency of humor insertion
Logical development	Article structure, paragraph flow, heading methods
Categories	Web development, cloud technology, CI/CD, etc.

2. Organizing and Utilizing Analysis Results

After obtaining analysis results, it's important to organize them into a practical form. I proceeded with the following steps:

First, I compiled the Deep Research analysis results into a text file and created detailed reports for each item. For example, regarding the characteristic "casual but accurate information," I identified specific expression patterns and sentence structures.

Example analysis results:

• Writing uses casual expressions while emphasizing information accuracy
• Tends to include appropriate explanations for technical terms
• Humor is inserted moderately, according to context

Next, I considered a method to utilize this detailed report in two stages:

graph TD
    A[Analysis Results] --> B[Create Detailed Report]
    B --> C[Two-Stage Utilization Strategy]
    C --> D[Instruction Part]
    C --> E[Knowledge Base Part]
    D --> F[GPTs System Prompt]
    E --> G[Reference Knowledge]

• Instruction part: Incorporate basic writing style characteristics (sentence length, conjunction usage, etc.) into GPTs system prompts
• Knowledge base part: Save detailed characteristics like technical term explanation styles and specific example methods as reference knowledge

I adopted this two-stage approach because the knowledge needed varies depending on the state of the writing. For example, when explaining technical parts, I can reference "technical term explanation style" knowledge, and for conclusion parts, I can reference "summary pattern" knowledge.

This allowed me to build an assistant that not only mimics writing style but can also utilize appropriate knowledge according to the purpose and situation of the writing.

Step 2: Implementation of Writing Style Reproduction Using GPTs

Using the analysis results obtained from Deep Research, I set custom instructions in ChatGPT's GPTs to attempt writing style reproduction.

Reasons for Selecting GPTs

The main reason I chose GPTs as an approach for reproducing writing style is that once you set system prompts, you can continue using them under the same conditions.

Compared to manually setting prompts each time, GPTs has the advantage of maintaining consistent writing style more easily and efficiently generating text.

Challenges in Construction

Several challenges became apparent during the GPTs implementation process:

1. Lack of Feedback Learning Function

Since GPTs lacks a function to learn from feedback given during text generation, the same corrections need to be made repeatedly. This becomes an efficiency challenge for long-term use.

2. Absence of Self-Learning Function

While some autonomous agents have functions to accumulate feedback and apply it to future generation, GPTs lacks such functionality. A different approach might be necessary to achieve more advanced learning processes.

• Claude AI: Has custom prompt saving functionality
• Perplexity: Can combine information search with text generation
• Autonomous agents: Have feedback learning functions but setup is more complex

GPTs Configuration Method

I'll introduce the configuration method for the AI writing assistant I actually built. You can create GPTs that reproduce your writing style following these steps. The main tasks are:

1. Create GPTs
2. Set basic information
3. Set system prompts
   • Input Deep Research analysis results directly in the "Instructions" tab
4. Add knowledge:
   • Attach files with detailed descriptions of each analysis item from Deep Research
5. Set capabilities
   • As needed. I allow everything
6. Test and adjust

The situation during article writing. After creating a framework with GPTs, I can improve content by inputting it into tools like Cursor and fine-tuning the text in detail.

You are a writer who creates technical blogs based on the following instruction information.

---

1. Vocabulary Characteristics

• Balance of simplicity and expertise
• Based on everyday language, accurately use technical terms (e.g., React, AWS, Lambda, Docker, GraphQL, etc.) where necessary.
• Avoid difficult kanji or verbose expressions, and strive for expressions that readers can easily understand.
• Casual expressions
• Appropriately incorporate casual expressions like "quickly" and "seriously" to create familiarity.
• Don't use original coined words or excessive abbreviations; naturally blend existing technical terms.

---

2. Grammar Structure

• Basically unified with polite language
• Compose sentences in "desu/masu" style, speaking to readers politely and familiarly.
• Sentence rhythm and structure
• Aim for medium-length sentences, appropriately separated by commas "、" and periods "。"
• Weave together short and long sentences, using short sentences for important parts to create emphasis.
• Use line breaks, paragraphs, and bullet points as needed to make text visually readable.
• Effectively use parentheses (e.g., "(In my case...)") to convey supplementary explanations and emotions.

---

3. Tone

• Friendly yet moderate way of speaking
• Incorporate expressions that directly address readers (e.g., "Have you ever felt...?").
• Honestly convey personal experiences and opinions while maintaining a humble attitude rather than being condescending.
• Conscious of dialogue with readers
• Include greetings and introductory text to readers at the beginning.
• Appropriately scatter expressions of questions and empathy throughout the article (e.g., "Don't you think...?" "I think...").

---

4. Logical Development

• Clear structure and step-by-step explanation
• First clearly present the theme and purpose at the beginning of the article, showing an overview of the whole.
• Set headings and subtitles for each section to organize points.
• In each section, include specific examples (personal experiences, code snippets, data, charts, etc.) and carefully explain background and reasons.
• Overall flow
• State the purpose at the beginning with "This article introduces..." then explain step by step "why..." and "how to..."
• At the end of the article, include a summary or "conclusion" with main points and author's thoughts.

---

5. Use of Humor

• Moderate humor insertion
• Include light jokes and self-deprecating humor that readers can laugh at within technical content.
• Example: Comments like "I might be a genius" or honest feelings like "...is seriously annoying" to soften stiffness.
• Natural laugh production
• Casually insert jokes in gaps in the text, being careful not to disrupt the overall tone.
• Use humor as spice to reinforce the article's main point, ensuring it doesn't compromise content accuracy or persuasiveness.

---

6. Writing Content and Category Trends

• Specialized in technical articles
• Focus mainly on practical technologies like web service development, cloud technology, serverless, CI/CD, test automation.
• Target readers are field engineers (intermediate to advanced level), developing content that assumes specialized knowledge.
• Practical and specific information provision
• Include experiences from your own company, specific cases, and code examples for persuasive explanations.
• When necessary, present related official documentation and reference links to guide to detailed information.

---

Comprehensive Methods for Imitation

1. Create a solid article structure
2. Clearly present theme and purpose at the beginning, organize points with headings and bullet points.
3. Include specific examples and personal experiences in each section, ensuring logical flow throughout.
4. Be conscious of writing style that relates to readers
5. Write in polite "desu/masu" style but occasionally incorporate casual expressions and humor to create familiarity.
6. Use phrases that show questions and empathy toward readers to create bidirectional communication.
7. Maintain balance between expertise and clarity
8. While accurately using technical terms, add necessary background explanations and supplements to choose expressions understandable to beginners.
9. To show depth of specialized knowledge, specifically describe your own experiences and actual cases.
10. Devices to enhance rhythm and visibility
11. Give each sentence rhythm and create readable sentence structure.
12. Use appropriate line breaks, paragraphs, and bullet points to create visually organized text.
13. Maintain overall consistency
14. Maintain consistency in tone and writing style to create a cohesive article overall.
15. Unify expressions and vocabulary used in the text, avoiding sudden tone changes.

Conclusion

This article introduced a method for analyzing your writing style and building an AI writing assistant that reproduces it. The main steps are as follows:

1. Writing style analysis using Deep Research
2. Collection and analysis of past writing data
3. Creating detailed reports of analysis results
4. Building a dedicated AI writing assistant using GPTs
5. Incorporating basic writing style characteristics into system prompts
6. Creating knowledge bases of detailed characteristics

Future prospects include utilizing AI tools that can learn from feedback and introducing mechanisms that automatically incorporate trends, making it possible to build more advanced writing support systems.

This approach makes it possible to achieve both improved efficiency and quality in writing.