AI Prompt Methodology (I): From Giving Identity to Giving Framework

Most people use AI like this: open the chat, directly ask "help me write an article," then frown at the response — "this isn't what I wanted."

The problem isn't that you didn't express yourself clearly. It's that you never told AI who it is, what you're doing, or what direction you want.

This article covers four techniques that fundamentally change how you converse with AI.

First, Give AI an Identity

Have you noticed that the same question, asked by different people, gets answers of vastly different quality? A professional journalist and an elementary school student asking "what is blockchain" will get answers of completely different depth.

AI is the same. By default, it's a "general assistant" — it knows a little about everything, but nothing deeply. If you don't tell it a role, it can only give you a mediocre generic answer.

So the first step is to give AI a clear identity.

Don't just say "help me write copy." Try this instead:

You are a tech product copywriter with 10 years of experience, skilled at explaining complex technical concepts in language ordinary people can understand. Your writing style is concise and persuasive, and you like to support your points with specific numbers and user scenarios. Now I need you to help me write an article recommending AI writing tools.

This identity assignment does three things: tells AI what professional capabilities it should have, tells it what tone to use, and tells it the specific context of this task.

The more specific the identity, the higher the quality of AI's output. This isn't mysticism — it's because AI's underlying mechanism is "predicting the most likely response based on context." The more precise the context you provide, the more accurate its predictions become.

A common mistake people make is giving a role but no personality. Saying "you are a copywriter" is fine, but saying "you are a copywriter who loves using unexpected metaphors and never writes more than two sentences in a row without a paragraph break" pushes the AI toward output that has a distinctive flavor rather than the default bland, neutral style most AI produces.

First Diverge, Then Converge

Most people's habit with AI is: think of a question → directly ask AI → get answer → done.

The problem with this process is that you treat your initial idea as the only answer, and AI is just helping you "execute" that idea. But AI's real value isn't execution — it's helping you discover things you didn't think of.

The correct approach is: let AI diverge first, then you converge.

Specifically, before asking AI for answers, don't rush. Instead, share your confusion, your initial ideas, your vague goals, and then ask:

What angles can you think of that I haven't considered?

Or:

If you were an expert in this field, how would you think about this problem?

Or more simply:

What are the common approaches to this topic? What are the uncommon but potentially better approaches?

At this stage, AI is your thinking partner, not your typist. It helps you broaden your vision and see more possibilities.

Once you've diverged enough, then pick the most valuable ideas and move to the next step.

I once needed to write a product launch plan. Instead of asking AI directly to draft the plan, I first asked it "what are the product launch strategies that most startups overlook?" Three of the suggestions — a pre-launch community teaser campaign, an exclusive beta tester program, and a founder story angle — became the backbone of the actual plan and were ideas I had never considered on my own.

First Plan, Then Define Details

After the divergence phase, you might have a bunch of ideas. The easiest mistake to do now is: pick one idea and start working on it, figuring out details as you go.

The result is often: halfway through, you realize the direction is wrong, or the details contradict each other, and you have to start over.

The correct order is: first define the direction, then define the details.

I call this the "harness theory" — before riding a horse, adjust the harness (reins, saddle, stirrups) first. The harness determines which direction the horse can run and how fast. If you rush before the direction is set, you'll either go off course or fall off.

How to define direction? Write down your core goals, constraints, and key elements. For example:

The target audience for this article is zero-basis users. The core goal is for them to be able to start immediately after reading. The constraint is within 1,500 words and no paid tools can be mentioned. Key elements must include screenshots and step-by-step instructions.

This is your "harness" — it defines the direction but doesn't yet define each step.

Once the direction is locked, the details won't go off course.

Give Framework, Not Answers

This is the most advanced technique, and the secret to getting maximum value from AI.

Many people give AI instructions like this:

Help me write an article about AI writing tools, including tool introductions, tutorials, and comparative analysis.

This seems specific, but actually you're telling AI "what to write," not "what standard to meet." AI will fill in content based on its own understanding, and the result often has gaps from your expectations.

A better approach is to give framework, not answers:

I'm writing an AI writing tool recommendation article for zero-basis users. My core argument is "free tools are already good enough." The article needs readers to decide which tool to use within 5 minutes. You decide the structure, but it must include a real usage case. The tone should be like chatting with a friend, not like a product manual.

See? You gave the core argument, target audience, reading experience requirements, and tone requirements, but didn't dictate specific content.

This is "giving framework, not answers" — you're the architect, AI is the executor. You set the standards and direction, AI fills in the flesh.

The benefit is: AI's output has both your soul and AI's creativity. You don't need to edit word by word, but every part of the finished product meets your standards.

Summary

Four techniques, from low to high:

• Give identity — let AI know who it is
• First diverge, then converge — let AI help you open up your thinking
• First plan, then define details — use the harness to set direction
• Give framework, not answers — you be the architect, AI be the executor

These four techniques aren't independent — they can be combined. In the next article, I'll cover how to use this methodology to guide actual AI-assisted development workflows — the Single Feature Loop Principle, and why "finish one before starting the next" is ten times faster than "spreading out simultaneously."

Next: AI-Assisted Development Workflow — Single Feature Loop Principle

Mastering these four techniques transforms your relationship with AI from "I ask, it answers" to "I direct, it creates." The key insight running through all four is that the quality of AI output is bounded by the quality of your thinking before you open the chat. Spending five minutes thinking through your identity, your direction, and your framework before typing your first prompt will save you hours of back-and-forth revision afterward. An additional dimension worth exploring in combination with these four techniques is the concept of "iterative refinement chains" — the idea that instead of expecting a perfect output in a single conversation, you deliberately structure your engagement with AI as a series of progressive refinements where each round builds on the last. This mirrors how human creators actually work: a rough sketch, then a rough draft, then a polished version. By mentally separating the "divergent exploration" phase from the "convergent polishing" phase and treating each as a distinct step in your workflow, you avoid the common trap of trying to get everything perfect in one exhaustive prompt, which paradoxically produces worse results than a series of focused, incremental requests. One practical exercise I recommend for internalizing this methodology is to take a task you have already completed using AI with mediocre results and re-do it from scratch using all four techniques in sequence. Compare the outputs side by side — the difference in quality and usefulness will be immediately apparent and will reinforce the value of investing those extra five minutes of thoughtful preparation before you send your first message. This comparison works best if you save both the original and the revised prompt-and-response pairs, so you can revisit them later and track how your prompt engineering skills develop over time.

The ultimate measure of prompt mastery is not the elegance of a single request but the consistency of high-quality output across hundreds of diverse tasks over weeks of iterative refinement.