It took me a long time using AI to gradually notice something.

“Writing fluently is not the same as writing like me.”

When I first used AI to write articles, I often gave it only a topic and added, “Make it sincere,” or “Don’t make it sound like AI.” There was not much obviously wrong with the result. But if I replaced my name with another creator’s, it seemed to work just as well.

This was not entirely the model’s fault. I had handed over the immediate task without telling it who I was, what I had been doing lately, which statements I agreed with, or which I would never say however polished they sounded.

I have freelanced in Shanghai for fourteen years. Watercolor, photography, reading, and writing have always mingled in my life. Now there are also fiction, AI videos, a personal website, and a knowledge base. If none of these experiences enters the context, all AI sees is one isolated question. Naturally, it can only produce an answer that is reasonably correct in an average sort of way.

“A prompt tells AI what to do this time. Context tells AI who you are.”

Many people place their hopes in prompts, looking for a sufficiently powerful magic phrase that will suddenly make AI understand.

Prompts are useful, of course. They tell AI what to do on this occasion. But why an article needs to be written, what setbacks you have experienced in the matter, and how your current circumstances differ from three years ago are hard to explain in a single prompt.

I now prefer to see AI collaboration as a process of gradually becoming familiar. Personal experience, ongoing projects, examples of expression, and the reasons left behind after each revision all give it a little less to guess.

As the materials become more settled, the changes I see are specific. AI starts to know that I work in watercolor, and that I do not want every article to become a story about painting. It knows I am exploring the possibility of a one-person business, and that plans must not be written up prematurely as results. It will even remind me that I recently used a particular example and that using it again might feel repetitive to readers.

Only then does the writing start to feel as if it has grown out of my own worktable.

You do not need to wait for a complete knowledge base to begin. Start with four simple documents.

The first is “About Me.”

Describe your work history, long-standing abilities, important turning points, and relatively stable values and judgments. There is no need to package a persona here. Being truthful matters more than being comprehensive. Begin with the experiences that influence your decisions today.

The second is “What I Am Working On.”

Projects change, and priorities change too. I keep ongoing work, sticking points, and next steps separately here. Otherwise AI can easily put watercolor, fiction, short videos, and course services into a single article, like a porridge containing every possible ingredient that no one can bear to eat.

The third is “How I Like to Express Myself.”

Choose three to five older articles you are truly satisfied with, and mark the sentences, rhythm, and spaces you like. “More natural” or “More like a real person” is still vague to AI. Actual text and specific choices are more useful than dozens of adjectives.

The fourth is “Why I Made This Edit.”

It is normal for AI’s first attempt to miss the mark. When deleting a passage, jot down the reason. Perhaps it sounds too much like course sales copy. Perhaps it compresses a complicated experience into a slogan. Perhaps it reveals something you are not ready to share.

As these edits accumulate, they preserve both your writing style and how you make judgments. For an independent creator, the latter is more important.

If you want to try today, begin by asking AI to write a self-introduction of about two hundred Chinese characters.

Then look at just three things.

Are the facts wrong?

Does the voice sound like you?

Has it crossed the boundary of what you are willing to make public?

For each problem you find, add one note to the knowledge base. Test it with the same task next time. After a few rounds, you will understand what AI lacks more clearly than you would by simply collecting prompts.

My own knowledge base grew this way. It is not perfect. There are still old views, duplicate materials, and plans that never happened. Whenever I actually write an article, work on a project, or receive feedback, I return and correct a little more.

AI is good at organizing things that have already been recorded, and it can help me find connections between different projects. It cannot go out sketching in the street for me, decide which experience I may share publicly, or take responsibility for what happens after publication.

Helping AI understand me better does not mean handing everything over. Identity documents, account information, client materials, and private health records all need clear boundaries. Personal circumstances also change. Old records need regular checking. “It remembers” must not be mistaken for “It understands correctly.”

“With these materials,

AI can write a little less of what anyone could say, and help me organize more of the expression that is truly mine.”

The views, experiences, and final judgments in this article are the author’s. AI assisted with organizing the text and generating illustrations.