Master Prompt Engineering: A 4‑Step Method to Make AI Give Exactly What You Want

Most people struggle with AI not because the models are unintelligent, but because they don’t know how to phrase their requests. The article argues that the quality of an AI’s answer is directly determined by the clarity and specificity of the prompt.

Why Asking the Right Question Matters

Imagine a super‑expert who can answer any question but has no mind‑reading ability. If you ask vague requests like “give me advice,” the answer could be about the Big Bang; if you ask “write something,” you might get a Tang‑poem. This illustrates the classic “Garbage in, garbage out” principle.

The Four‑Step Prompting Method

The author calls the method “Four‑Step Prompting” and claims it works for writing copy, planning trips, analyzing work problems, etc.

Step 1 – Assign a Role

Tell the AI which “persona” to adopt so it knows which knowledge domain to draw from. Examples:

"Please act as an experienced advertising planner."

"Please act as a kindergarten teacher with a gentle, playful tone."

"Please act as a HR consultant with 10 years of experience."

Step 2 – Provide Background

Give the AI context about who you are and the situation. Examples:

Job‑seeking email: "I just graduated and want to apply for an internet operations role; I’m outgoing but lack experience, so highlight my learning ability and enthusiasm."

Family trip: "We are a family of three, child is 5, have three days, departing from Beijing, budget ~5,000 CNY, want a low‑effort nature getaway."

Work summary: "I’m in sales, achieved 120 % of target this month, but two big clients fell through; my manager cares about data analysis."

Step 3 – State the Specific Task

Clearly describe what you want the AI to do, including length, style, and any constraints. Examples:

"Write a 300‑word cover letter, start with enthusiasm, list my college activities, and end with a request for an interview."

"Recommend three travel plans for my family, each with destination, transport, accommodation, and a brief daily itinerary."

"Trim this paragraph to under 100 words, keep the core meaning, and make the tone more assertive."

Step 4 – Specify the Output Format

Tell the AI how you want the result presented. Examples:

"Present the answer as a bullet‑point list."

"Generate a ready‑to‑post social‑media copy without extra explanation."

"Create a table with columns for option name, pros, and cons."

"Produce a concise PDF document."

Hidden Trick: Provide a High‑Quality Sample

Beyond the four steps, you can give the AI an exemplary piece of text and say, "Please use this style to write something similar." The model will pattern‑match the style, structure, and tone, often yielding better results than issuing dozens of separate instructions.

Full Example: From a Vague Request to a Precise Prompt

Two prompts are compared for generating a WeChat Moments post.

"Help me write a Moments post."

The AI replies with a generic, bland sentence.

"Please act as a humorous jokester (role). I just finished a workout and am proud of my new abs but feel shy about bragging (background). Write a self‑deprecating, concise post that lets friends know I’ve made progress without being pretentious (task). Keep it short, suitable for Moments, and include one or two emojis (format)."

The AI returns a witty, personalized message with emojis, clearly matching the user’s intent.

Iterative Refinement: Keep the Conversation Going

Even after using the four‑step method, the answer may not be perfect. The article suggests asking follow‑up questions such as “Can the second paragraph be funnier?” or “Can the budget be reduced?” Each iteration nudges the model closer to the desired output, similar to a collaborative polishing process.

Conclusion

Prompt engineering is essentially learning to “talk well” to a smart but literal assistant. No programming or deep‑learning knowledge is required—just clear, specific language that guides the AI’s massive knowledge base toward the exact answer you need.

Next, the author promises a discussion on how experts evaluate AI‑generated options, hinting at a second skill: “taste and discernment.”