I no longer start writing content from the beginning; I start with a question. This is where AI truly changes my workflow.

Most people use AI to get answers, but I care more about the question itself. I first write a very rough question and then continuously refine it.
For example, initially, I ask what this project is doing. I change it to which user behavior this project has altered, then to who’s business would be disrupted if this model is successful.
As the question changes, the content becomes completely different. Then I give this question to AI and ask it to generate multiple possible scenarios.
A prompt I often use is very simple: Give me three different perspectives analyzing this topic—one from the user behavior angle, one from the market structure, and one from an unconventional perspective. Don’t give conclusions, only the reasoning process.
This step isn’t for immediate use but to find anomalies—those points that defy intuition but have explanatory power are the core of the content.
Next, I start writing, and AI is only used to calibrate the structure, such as rhythm and information density.
The expression must be done by myself because style cannot be outsourced. This process has saved me from writing many correct but boring contents and helped me better understand what is worth expressing.
That’s also why I find @RallyOnChain very interesting.
What it’s doing, in essence, is recognizing effective expression rather than surface-level data performance.
In the future, content competition won’t be between humans and AI, but between who is better at asking questions.