Most disappointing ChatGPT results at work come from prompts that are too short to give the model anything to work with. "Write me an email about the delay" produces something generic because there's nothing specific to draw from. The fix isn't a magic phrase — it's including four things almost every good office prompt needs: context, audience, constraints, and format. Here's what that looks like for common tasks.
The four-part structure
- Context: what's actually happening — the situation, not just the topic.
- Audience: who's reading this and what they already know.
- Constraints: tone, length, what to avoid saying.
- Format: email, bullet list, table, slide outline — be explicit.
You don't need to label these parts in your prompt. You just need all four pieces of information present somewhere in it.
Example 1: Delay notification email
Weak prompt: "Write an email about the project delay."
Better prompt: "Write a short email to a client telling them the website launch is delayed by one week due to a third-party integration issue, not anything on our end. Keep it confident and brief, not apologetic or defensive. They're a long-term client who's generally easygoing about timelines. End with a new target date and an offer to call if they have questions."
The second version gives ChatGPT a reason for the delay, a relationship context (long-term, easygoing client), and a tone constraint (confident, not apologetic) — all of which shape word choice in ways the first prompt simply can't.
Example 2: Turning a messy meeting transcript into action items
Weak prompt: "Summarize this meeting."
Better prompt: "Here's a meeting transcript. Pull out only the concrete action items — things someone explicitly agreed to do — and list each with who's responsible and any deadline mentioned. Skip general discussion points that didn't result in a decision. Format as a simple list, no more than one line per item."
Without the instruction to separate "decisions" from "discussion," ChatGPT will often summarize everything that was talked about, leaving you to manually pick out what's actually actionable. Telling it exactly what to extract — and what to skip — does that filtering for you.
Tip: for meeting notes specifically, paste in speaker names if your transcript has them, and ask ChatGPT to attribute each action item to the right person. Generic "summarize this" prompts often lose that attribution.
Example 3: Writing a formula explanation for a spreadsheet
Weak prompt: "Explain this Excel formula."
Better prompt: "Explain this Excel formula in plain English, as if to someone who knows basic Excel but has never used VLOOKUP or IF statements before. Then tell me what would happen if the lookup value isn't found in the range."
Specifying the reader's existing skill level changes whether the explanation assumes too much or over-explains the basics. The second sentence in the prompt also asks for an edge case most people forget to check until something breaks.
Example 4: Turning rough notes into a status update
Weak prompt: "Make this sound professional."
Better prompt: "Turn these rough notes into a short status update for my manager. Keep my informal tone mostly intact — don't make it sound stiff or corporate. Three short paragraphs max: what's done, what's blocked, what's next."
"Sound professional" is vague enough that ChatGPT often defaults to overly formal, generic business language. Saying explicitly that you want to keep some personality, plus giving the exact structure (done / blocked / next), produces something closer to what you'd actually send.
A few habits that consistently help
- Paste in real examples of your own writing if tone matching matters — "match the tone of this email I wrote last month" works better than describing a tone abstractly.
- Ask for options, not one answer, for anything where phrasing is subjective: "give me three versions of this opening line, ranging from formal to casual."
- Iterate instead of starting over. If the first draft is close but not quite right, say specifically what's off ("this is too long" or "drop the second paragraph") rather than rewriting the whole prompt from scratch.
- State what to avoid, not just what to include. "Don't use the phrase 'I hope this email finds you well'" is sometimes more useful than any positive instruction.
The bottom line
A prompt that includes context, audience, constraints, and format will almost always beat a longer but vaguer one. The goal isn't to write an elaborate prompt every time — it's to make sure the model actually has the specific information it needs to produce something you can use without heavy editing.