How to Write a Good ChatGPT Prompt for Work: The 4-Part Formula

A reliable work prompt has four parts: role, task, context, and format. Instead of "write an email," say: "You are my ops manager. Draft a 120-word reply to this late-delivery complaint (pasted below). Tone: apologetic but firm. Bullet the next steps." Naming the role, the exact job, the real details, and the output shape is what turns a generic answer into something you can paste straight into work.

Type "write an email about the late order" into ChatGPT and you get bland filler. Not because the AI is weak, but because you gave it almost nothing to work with. It doesn't know who you are, which order, how angry the customer is, or how long the reply should be, so it guesses the safe average. The fix is a habit, not a trick: give it four things every time. Role, task, context, format. Vague in, vague out. Specific in, usable out.

Every good work prompt answers four questions. Role: who should the AI act as? "You are my ops manager." Task: what exactly should it do, including length? "Draft a 120-word reply." Context: the real details it can't guess, usually pasted in. "Here is the customer's complaint: ..." Format: the shape of the answer. "Apologetic but firm, and bullet the next steps." Miss one and the answer drifts. Hit all four and the first draft is usually close enough to paste with light edits.

Start the prompt by telling the AI who to be and the one job to do. The role sets the vocabulary and judgment; "you are a careful bookkeeper" answers very differently from "you are a punchy copywriter." Then state the task as a single clear verb with a size on it: draft, summarize, rewrite, compare, list. Add the length here so you don't get an essay when you wanted three lines. Example opener: "You are my ops manager. Draft a 120-word reply to an unhappy customer." That one sentence already beats most prompts people send.

This is the part beginners skip, and it's the one that matters most. The AI can't read your inbox, your CRM, or your head. Whatever it needs to know, paste it in: the actual email thread, the order number, the dates, the customer's exact words, the product details. More real context means fewer made-up details and less back-and-forth. A good move is to label it so the model knows what it's looking at: "Here is the complaint, pasted below:" and then the text. If a fact matters to the answer, it has to be in the prompt.

Tell the AI what the finished thing should look like before it writes, not after. Specify length, tone, and structure in one line: "Under 120 words, apologetic but firm, bullet the next steps, no greeting." This is the difference between an answer you reshape for ten minutes and one you paste straight into Outlook. Formats worth asking for by name: a bulleted list, a short table, a 5-line email, a tweet-length summary, "just the subject line," or "three options I can choose from." The more precisely you describe the output, the less you edit.

The weak version: "write an email about the late delivery." The strong version, all four parts in one go: "You are my ops manager. Draft a 120-word reply to this late-delivery complaint (pasted below). Tone: apologetic but firm. Bullet the next steps, no greeting." Same AI, same model, wildly different result. The second one reads like you wrote it on a good day. Notice you didn't use fancy words or prompt-engineering jargon. You just refused to make the AI guess. That's the whole skill.

You don't rewrite this from scratch every time. Turn your best prompts into fill-in-the-blank templates and reuse them. Keep one that reads: "You are my [role]. [Task with a length]. Here is the context: [paste]. Format: [tone, structure, length]." Save it in a note, or in ChatGPT, Claude, or Gemini as a saved prompt. For a week, every time an answer comes out great, paste that prompt into the note. Soon most of your daily writing is a 20-second swap of the brackets, and the blank-page part of the job is gone.

The formula works the same in ChatGPT, Claude, Gemini, and Microsoft 365 Copilot, because it's about clear instructions, not a single tool. One rule before you paste, though: only put confidential company data into an AI account your employer approved. Personal and free accounts may use what you type to train the model. For sensitive replies, use the work-sanctioned tool, or swap real names and numbers for placeholders and drop the details back in yourself.

Pick one thing on your plate right now: a reply you're dreading, a summary you owe, a post you keep putting off. Write it using all four parts in a single prompt: role, task with a length, the pasted context, and the exact format. Read the first answer. If it's off, the fix is almost always more context or a tighter format, not a longer prompt. Get one win today, save the prompt that earned it, and you've started a template library that pays off every week.

Take the last prompt that disappointed you — it is still in your history — and rebuild it with the four parts. Same task, same tool, new brief. The before-and-after is how the formula sticks.