How I Started Writing with AI
English translation of the Chinese original (AI-assisted). Read the original in Chinese.
I want to record how I use AI to help me write.
First, I have to admit something: I have always believed that I am very poor at written language. I could even say that writing itself is painful for me.
Back in my school days, especially the six years of junior and senior high school, whenever a midterm or final Chinese exam included an essay, I almost never wrote one.
It wasn't that I didn't understand the question. It was that I knew very clearly that I did not have the ability to turn my thoughts into words.
During those six years, there were only two exceptions: the high school entrance exam and the college entrance exam. Those were exams too important to skip.
I mention this not to complain about the past, but simply to explain that I have always had a very clear understanding of my ability to express myself in writing.
But that doesn't mean I had no ideas. On the contrary, I have always been someone who likes to think and observe. It's just that, in the past, those thoughts had no outlet.
Later, I began using AI.
It was less like writing and more like having a conversation. I recorded the thoughts in my head by "speaking," and then handed them over to AI.
Gradually, I discovered that my thinking already existed, complete, within my conversations with AI.
When this material had accumulated to a certain point, I had a clear thought for the first time: perhaps I could organize it all.
This was the starting point of my writing with AI.
⸻
As for tools, my main one is ChatGPT.
The reason isn't complicated: in my view, its abilities are relatively balanced across the board, and it is more stable.
Besides that, I also use other models, including DeepSeek, Kimi, and Qwen from China, as well as Claude, Google's Gemini, and Grok from overseas.
I mostly use the free versions of these tools.
But they are not my main tools. Within my system, they mainly play supporting roles, for cross-checking and proofreading.
I deliberately avoid relying on a single model.
⸻
When I have a new writing idea, I usually start a new conversation in ChatGPT. That conversation becomes the starting point of a new project.
The first stage of the project is not to write the main text directly. I do only one thing: make clear "why I want to write this project."
This includes why I have these ideas, what my purpose is, and what I do not intend to write.
Through repeated conversations, AI gradually understands my purpose and boundaries. At the same time, through this process, I also clarify my own thinking for the whole project.
Once all of this is clear, I move on to the next step: I ask AI to organize the entire conversation and generate a structural outline.
Once the outline appears, the project has officially entered the writing stage.
⸻
Next, following the outline, I create a separate conversation for every chapter and every important supplementary section.
When I have a new idea, I go into the corresponding conversation and put the thought back where it belongs.
When a chapter is finished, AI helps me organize it into a first draft.
I first read and revise it myself, and then give it to other AIs, usually no fewer than three, for independent evaluation.
I collect their feedback and give it back to my main AI to hear its judgment.
In most cases, it accepts the suggestions; but sometimes it insists on the original structure. The reason is simple: it understands my true purpose and boundaries best.
In the end, all decisions about what to change and what to keep are still made by me.
The draft goes through multiple rounds of feedback and revision until the final version is complete.
For me, the changes brought by AI are not limited to individual articles.
After finishing a manuscript, I can use AI to translate it into Traditional Chinese, Japanese, and English versions. Then, through repeated cross-checking with multiple models, I correct shifts in meaning and problems of expression.
The result: one book becomes versions in four languages.
Without AI, this would be almost unimaginable in terms of time, money, and even practical feasibility. Now it has become something that can be planned, verified, and continuously improved.
⸻
Looking back, what AI has truly changed is not merely writing efficiency.
It has taken "writing," which used to depend heavily on individual ability, and broken it down into a process that can be shared, verified, and repeatedly corrected.
I no longer need to sit alone in front of a blank page and burn time; I can put my energy into judgment, choices, and structure.
But throughout the whole process, one thing has never changed—and it is also the part I have deliberately not handed over to any AI.
That is judgment.
What to write and what not to write; which suggestions to accept and which opinions to reject; and, in the end, the person who must take responsibility for these choices can only be oneself.
AI can improve efficiency, amplify thinking, and reveal blind spots.
But it cannot decide for you who you are, what you want to say, and what you are willing to take responsibility for.
Images in the original are not included in the web version.