How I Communicate Effectively with AI: An Introduction to 'AI Mastery'
Who hasn’t used AI these days?
Almost everyone around me is using AI, from the elderly to children. But different people use AI in different ways, and they get different results.
Some people use AI as a search engine to quickly find answers, which is fine. This is the starting phase of using AI.
Some have multiplied their work efficiency by ten or even a hundred times using AI. Some people use AI for programming, developing various tools; others do research, write, design, and run operations with it. Some even make decent profits from it.
If you don’t want to just stay in the starting phase of using AI and want to learn more, I’m happy to share my experience.
A brief intro: I’ve been doing internet product design for over 20 years and have a wealth of experience, especially in working with programmers. But, I could always say I hardly understand code.
I don’t know code, but now I’m using AI to program, and I’ve already developed several products. For me, the most important thing about AI isn’t suddenly becoming a programmer but having the chance to slowly turn my hard-to-execute ideas into real things.
So this article is not a tech tutorial.
I want to share more about how a non-tech person can communicate with AI effectively.
I later found that using AI well isn’t that complicated: make your goals clear, explain the background, state the constraints, set the standards for the results, and then continually adjust how AI works.
To put it simply, it’s still communication.
01 Before Communication, Understand Your Communication Partner’s Personality
The “personality” here, of course, doesn’t mean AI has emotions or motivations like a human.
I mean, AI will show some stable tendencies during usage.
It’s just like communicating with people. Different people have different ways of communicating. You need to know roughly what kind of communicator the other party is to know how to speak and get results.
I think AI has at least a few characteristics:
- Very capable.
- Likes to take shortcuts.
- Likes to please the user.
AI being very capable needs no further explanation. Many people are first blown away by AI exactly because of this.
But being capable doesn’t mean it will automatically give you the best results.
When I say AI likes to take shortcuts, it means if you don’t set standards or constraints, it often chooses the easiest, safest, most template-like answer.
You say “write an essay,” and it might give you a complete but very generic essay. You say “help me analyze,” and it might give you some common angles.
It’s not because it’s not smart, but because you didn’t tell it what “good” means.
AI is also eager to please. I believe many have experienced this. You show it a piece of text, and it might first say, “This is already well-written,” then give some gentle suggestions.
But many times, we don’t need comfort; we need more objective answers.
So, when communicating with AI, we can’t just treat it as a tool that answers questions. We need to know how it would typically answer and then actively give it rules, standards, constraints, and even require it to challenge us in return.
Precisely because of this, the ensuing communication methods are meaningful.
02 Before Asking Questions, Prepare Them
I think this point isn’t about sharing a method but a mindset.
Many use AI somewhat casually, opening the input box and typing in a bunch of mindless input. They end up complaining about AI, calling it dumb or stupid.
You ought to know, if you want a good answer, you first need to prepare a good question.
This doesn’t mean you need to learn how to craft a complex prompt, but at least consider a few things first:
What exactly do you want AI to help you with? Who is this result for? How do you want it delivered? What can’t it do? To what extent should it be done to be considered complete?
If you haven’t thought about these questions, AI will likely only guess. If it guesses right, you’ll think it’s smart; if it guesses wrong, you’ll think it’s talking nonsense. But at the end of the day, it’s often not about AI being incapable; it’s about us giving it vague tasks.
A good question is often half the work.
03 How to Start: Use Meta Prompting
If it’s your first time using AI, or the first time using AI for a professional task, it’s like getting a brand-new, never-used machine.
In such a scenario, how should you start?
Simple, the question to ask is: How do I begin using it?
For instance, if you want to learn AI programming, you don’t have to immediately ask:
“Please help me write a complete app.”
You can first ask:
“I’m a product designer who doesn’t know code, but I want to try making a small tool with AI. Can you tell me where I should start? What information do I need to prepare? What common mistakes should I avoid?”
This method is so simple it’s almost laughable. If you’ve used it once, you won’t even think of it as a method because it’s so simple.
But it can break your unfamiliarity and fear of AI.
More importantly, it frees you from the pressure of “I have to ask the right question all at once.”
You can also have AI help you design the question:
“I want to accomplish this, but I don’t know how to ask you. Could you help me refine my needs into a better prompt?”
This is so-called meta prompting. It’s not directly having AI do the job, but first having AI help figure out how to hand the job to AI.
I often use it myself. Because many times, it’s not about not knowing what I want, but not knowing how to articulate it properly in an unfamiliar field.
04 Before Asking AI, Let AI Ask You
When using AI for complex tasks, the task often exceeds our personal knowledge range.
For example, in product design, I can clearly explain who the users are, the scenarios, and where the experience is uncomfortable. But as soon as it involves engineering implementation, databases, deployment, permissions, or security, I find myself at a loss.
The best way, then, isn’t to pretend to understand but to turn the table and let AI ask you questions.
More accurately, let AI translate those technical questions I don’t understand into product and life questions I can answer.
You can say:
“I want to create a habit-forming app for kids. I don’t know code and don’t know how to translate a product idea into technical requirements. Please don’t write code yet or rush to give solutions. Please, like someone who understands both product and technology, ask me 10 questions.
These questions shouldn’t use technical jargon but should be in language I can understand, helping you determine: what pages the product needs, what information to save, whether parents and kids are different roles, if login is needed, if data needs syncing, if reminders are needed, what content involves privacy, and what can be omitted from the first version.
Once I answer, you can help me compile it into a requirements document suitable for AI programming tools."
This method is extremely helpful.
Because AI will break down questions you didn’t know how to express into a set of questions you can answer.
It might ask: Are the child and parent using the same device or different ones? After the child completes a task, should the record just be kept on this phone, or should it be seen on other devices too? Can parents edit the child’s tasks and records? Does the app need daily reminders? If there is no internet, does the app still need to function?
These questions aren’t technical on the surface, but they will influence the technical implementation behind the scenes.
A good AI user isn’t someone who always orders AI around but someone who knows how to have AI clarify their questions.
05 Give Up Pretense: I Don’t Know Anything, Please Give Me the Simplest Explanation
When tackling complex things, you will surely encounter something beyond your knowledge or understanding range.
At this point, you can choose to “give up.”
This is not an attitude of giving up, but letting go of pretenses.
You can directly tell AI:
“I don’t understand anything about this field. Please explain it in the simplest way possible, without using any technical terms. If you must use technical terms, first explain in layman’s terms what it means and give me a real-life example.”
For instance, when I’m coding AI, I often run into errors. The error messages are filled with English and technical terms that I can’t understand.
I won’t pretend I understand. I’ll just copy the error to the AI and say:
“I don’t understand this error message at all. Don’t fix it yet, first explain it in a way non-programmers can understand: What’s it roughly saying? What could cause it? Is it serious? If I need to fix it, what information should I provide you?”
Asking this way, the AI’s job shifts from “writing code for me” to “helping me understand the problem first.”
Understanding the problem is important.
Because if you have no clue what the AI is doing, you can only trust it blindly. If it says it’s fixed, you believe it; if it says there’s no problem, you believe that too. This is actually quite dangerous.
I don’t understand code, but I need to at least try to understand a little bit of the problem’s structure.
This isn’t to become a programmer, but to judge whether the AI is messing around.
06 Start from user goals, not from the solution
Some people think AI programming is hard because it involves code, logical rules, various verifications, and tests, which is a complex process. Before AI, this did have a very high technical threshold; but in the AI era, everything is changing.
In the AI era, your task isn’t to first describe technical solutions, but to describe your intents and goals as accurately as possible.
For example, if I want to make a habit-forming app for kids.
I might describe it to the AI like this:
My kid is now 9 years old and has some bad habits. Despite our numerous reminders and requests, he still can’t correct them, and it’s worsened our parent-child relationship, making the child rebellious.
I hope for an app that can achieve the following goals:
- Help kids form good habits in their daily lives.
- Enhance interaction between parents and children, thus improving the parent-child relationship.
- When we and our child look back through this app, we can see the child’s growth, giving the child a sense of feedback and achievement about his development.
This is just an example of a broad requirement.
But it illustrates a point: Don’t just start by telling the AI “put a big button on the home page, a list below it, and then a popup.” That’s not entirely useless, but it jumps too quickly into a solution.
A better way to say it is:
“I want to make a habit-forming app for children aged 9. The main goal of the home page is for the child to know immediately what to do today upon opening it, and after completion receive immediate positive feedback. Please don’t rush to design the interface yet, start by analyzing the tasks the home page should undertake from a child’s usage scene, then provide 2 to 3 possible solutions.”
The former way tells the AI the interface in your head.
The latter way tells the AI the problem you’re trying to solve, allowing the AI to work with you to find solutions.
In my own work, I’m more and more inclined to first talk about the goal, then the constraints, and finally the ideas already in my head.
Because my ideas in my head might be wrong.
Real products are never just a stack of features but solve a real problem.
So my advice is: describe where you want to arrive first, then discuss how to get there.
07 Turn complaints and judgments into concrete questions
When AI does something wrong, people easily get angry and furious.
This is normal. Especially when you’ve gone back and forth making many changes, and it keeps making the same mistake, you can’t help but say:
“How did you mess up again?”
“Did you even understand my requirements?”
“This page is so ugly, redo it.”
These words can express emotions but don’t help much with improving AI results.
A more effective way is to turn complaints and judgments into concrete questions.
Don’t say:
“This page is too ugly.”
Instead, say:
“This page has three issues now: first, the button and background colors are too similar, making it hard for kids to notice; second, the feedback after completing tasks is too weak, with no sense of achievement; third, the parent and child perspectives are mixed, users don’t know what they should do. Please address these three issues first, and keep the interface simple overall.”
The same goes for writing articles.
Don’t say:
“This part isn’t well-written, help me fix it.”
Instead, say:
“This part feels awkward to me. Specifically: first, the first two sentences say the same thing, there’s repetition; second, the third sentence jumps to another point suddenly, lacking transition; third, readers finish and don’t know what exactly I want to emphasize. Please diagnose based on these symptoms first and then give me a minimal revision plan.”
If you only express dissatisfaction, AI might know you’re not satisfied, but it doesn’t know how to improve.
If you can point out specific issues, impacts, priorities, and expected results, it’s easier for AI to adjust.
08 Clarify task, context, and trying process
I used to think, talking too much to AI might be annoying?
Later I realized, many times it’s not that I talked too much, but that I said too many unimportant things and left out the truly important ones.
For instance, you say:
“Help me optimize this article.”
AI can optimize it, but it doesn’t know what you mean by optimize. Is it fixing typos? Making the article more engaging? Making the structure clearer? Is it suitable for a WeChat public account? Or for Xiaohongshu?
A better way to say it is:
“Please help me optimize this article. My target readers are those interested in AI, but without a technical background. Keep my first-person tone, don’t change it to a formal tutorial. Focus on three things: is the logic smooth, are the examples specific enough, and can readers follow it after reading. Provide a diagnosis first, don’t directly alter the original text.”
That’s much clearer.
AI is smart, but it doesn’t know what’s happening in your head.
It doesn’t know why you’re making this product, doesn’t know what you’ve tried before, doesn’t know what you like or dislike, and doesn’t know who your users really are.
So I usually provide AI with several types of information:
First, who I am.
Second, who I’m serving.
Third, in what scenario this is happening.
Fourth, what I’ve already tried.
Fifth, what I care about most.
Sixth, what should not be done.
After sharing this information, AI’s responses are noticeably different.
For example, I wouldn’t just say:
“This article isn’t good enough.”
I would say:
“I’ve tried adding a few method points, but the latter part still feels like an outline rather than an article. I hope you can help me expand each method into specific examples while keeping my conversational tone, don’t write it like a paid knowledge course.”
When you clarify the task, context, and trying process, AI can understand where you’re stuck now.
AI’s ability is often not limited by the model, but by the context.
09 Correct AI’s working method over delivering results
This is a realization I’ve increasingly come to.
Often, the problem with AI isn’t just that a particular result is wrong, but its working method is wrong.
For example, it starts modifying without first understanding the context.
For example, it sneaks around the real problem to make the result look complete.
For example, it always fixes surface issues without finding the root cause.
For example, you ask it to keep your writing style, but it changes the article to look nice, but nothing like yours.
At this time, don’t rush to have it continue correcting results.
First, correct its method.
You might say:
“Hold on. The problem wasn’t a particular sentence being poorly written, but you turned my article into a generic tutorial. Please review: where did you change my tone? Where did you add judgments I didn’t express? Next, please make only the minimal necessary changes, keeping my first person, hesitations, conversational tone, and rhythm.”
Or in AI programming, you might say:
“Don’t continue writing code yet. First, analyze the root cause of this failure. Don’t bypass it with a temporary solution. Please identify the links where the problem occurred, state which assumptions you want to verify, and then propose a fix.”
This is very much like managing a newcomer.
If a newcomer makes a mistake, you can say, “Change this part.” But if they always use the wrong method, you should correct their working method.
AI is no different.
Truly advanced AI use is not about constantly urging it to deliver, but consistently calibrating its working methods.
10 Let AI set its own validation standards
I’m becoming increasingly skeptical of one phrase:
“I’ve already fixed it.”
It’s not that AI is deliberately trying to trick me, it’s that it often confuses “looks complete” with “actually complete.”
So I ask AI to set its own verification standards.
For example, when making a product, I’ll say:
“Before you start implementation, please list the acceptance criteria for this task. Which can be automatically verified, like linting, tests, builds; which ones need human judgment, like visual effects, natural language, user comprehension. Once done, tell me for each item whether it passed, failed, or needs human review.”
Same with writing articles.
“Set completion standards for this piece first: is the outline spot filled, is the logic coherent, does every method have examples, does it keep my tone, and is there nothing new I didn’t mention? After revising, please self-check against these standards.”
This step is crucial.
Because once there are standards, you’re not just relying on feelings to judge whether AI did well or not.
Especially in AI programming, verification standards are more important.
If it changes code, you have to make it run tests, check for errors, open page screenshots, and compare before and after differences. You can’t just let it tell you, “It should work.”
“It should work” is not evidence.
Evidence should be: what checks it ran, what the results were, where it still needs a human look.
This is also how I protect myself as someone who doesn’t understand code.
I can’t read code, but I can ask AI to give me verifiable results.
11 Ask AI to Find Best Practices and Sources
Now when I do things I’m not familiar with, I often ask AI to first find mature practices in the industry.
Because AI can easily give you a seemingly reasonable plan out of thin air.
Looking reasonable doesn’t mean it’s reliable.
So I ask it:
“I’m not familiar with this. Please first research the official documentation, industry best practices, and excellent open-source projects, don’t just answer from memory. Please list sources, summarize common points and differences, then give advice based on my situation.”
If it’s AI programming, I’d say:
“First, search this project for similar implementations. Prioritize reusing existing code patterns, don’t invent a new style. If none, then find official documents or mature open-source projects as references.”
If it’s product design, I’d say:
“First analyze common user flows and interaction patterns in this field. Don’t rush to draw interfaces. First tell me: which patterns are mature, which areas tend to go wrong, which parts need redesigning with my user context in mind.”
This method is especially important for non-techies.
Because we don’t know if a solution is a conventional practice in the industry or if there are pitfalls behind it.
AI can help fill this gap in perspective.
But first, you have to clearly ask it to find sources, explain reasoning, make comparisons, instead of just giving a confident answer.
Don’t just let AI give answers, let it give reasoning.
12 Set Boundaries and Constraints for AI
AI is good at generating, but it doesn’t naturally know where your boundaries are.
So you have to give it standards, and constraints too.
For example, when writing, I’d say:
“Please retain my original intent, don’t add new viewpoints, don’t turn colloquial speech into formal text, don’t use abstract words to appear advanced. Every method point should have examples, but don’t fabricate my real experiences.”
When making products, I’d say:
“This app is for kids and parents to use together. Don’t use mechanisms that create stress, like shaming, punishment, ranking, or reminders of continuous login failures. The overall direction should be encouragement, companionship, and visible growth.”
When letting AI code, I’d say:
“Don’t change irrelevant files. Don’t introduce new tech stacks. Don’t delete tests just to pass them. Don’t hide errors. After completion, you must tell me what was changed, why, and what it might impact.”
These words may seem verbose, but they are very useful.
Because AI can sometimes be too eager.
It’ll do a lot of things you didn’t ask for just to complete the task. It’ll turn a small problem into a major overhaul. It’ll add judgments you don’t agree with just to make the result look complete.
So you have to tell it the boundaries.
Boundaries don’t curb creativity. Boundaries keep creativity from going astray.
13 Establish Rules and Processes for AI
When you just occasionally ask AI a question, one prompt is enough.
But if you collaborate with AI long-term, just relying on temporary inputs isn’t enough.
You need to establish processes for AI.
Like now, I would ask AI to follow a process when writing articles:
First, diagnose, don’t change directly.
First, state where it thinks the article’s problems lie.
Then give a revision plan.
After I confirm the direction, then it can modify the text.
After revising, it should still tell me what was changed, why, and what might still need my attention.
The same goes for product design and programming.
I set similar processes for AI:
First, read the context. Then list acceptance criteria. Then propose a plan. Then execute. Run checks after execution. Finally, tell me the results in plain language I can understand.
That’s why I gradually value things like Rules, Skills, AGENTS.md.
They’re essentially not technical documents, but work rules I write for AI.
I can’t explain every time: I’m someone who doesn’t read code, please tell me in plain language what was changed; don’t touch important files directly; don’t skip verification; don’t make my articles unlike me.
If these are always up to me saying on the fly, I’ll surely miss some.
So I turn them into rules.
The significance of rules is to let AI know my work style, my boundaries, my preferences, and my bottom line before every task.
At the end of the day, it’s like managing a team.
You can’t expect someone to rely on impromptu every time. You have to give him processes, standards, feedback, and also reviews.
AI is the same.
14 Finally
If you just use AI as a search engine, it’s already very useful.
But if you want to go further and turn AI into a true collaborator, you can’t just ask:
“What’s the answer?”
You have to start asking:
“What problem do I really want to solve?”
“Is my information sufficient?”
“Has AI understood my goal?”
“How to validate this result?”
“If it’s wrong, is it the result that’s wrong, or the working method that’s wrong?”
This is what I understand about AI handling techniques.
It’s not about learning a bunch of magical prompts.
It’s about learning to articulate your intentions clearly, breaking down problems, supplementing context, setting standards, and then adjusting AI in collaboration.
I don’t understand code, but I understand users, experience, and also the process of a vague idea becoming clear.
And AI just gives me an opportunity to start putting those formerly only in my mind ideas into reality bit by bit.
So I increasingly feel that the most important skill in the AI era might not be whether you can write code or remember prompt templates.
Rather, it’s whether you can ask AI the right questions.