How to Write a Good AI Prompt for Work Tasks
What OpenAI and Anthropic's official prompting guides recommend, turned into a simple structure for clearer AI requests at work, with a before and after.
Most disappointing AI answers at work come from a request that was too short. "Write an update about the project" leaves the tool guessing about who will read it, what happened, how long it should be and what tone to use. It fills every gap with something generic. The official prompting guides from OpenAI and Anthropic say the same thing in different words: the clearer the request, the better the result. Here is what they recommend and how to apply it to everyday tasks.
What the official guides agree on
Be clear and specific. OpenAI's ChatGPT help article advises making prompts clear and specific, with enough context for the model to understand what you are asking, and avoiding ambiguity. Anthropic's guide makes the same point and suggests thinking of the AI as a brilliant but new employee who does not know your norms and workflows. Its "golden rule" is a good test: show your prompt to a colleague with minimal context on the task. If they would be confused, the AI will be too.
Explain why. Anthropic's guide says that giving the reason behind an instruction helps the model understand your goal and produce more targeted responses. "Keep it under 150 words" is fine. "Keep it under 150 words because it will be read on a phone between meetings" is better.
Show examples. Both companies describe "few-shot" prompting: including a few examples of the input and the output you want. Anthropic calls examples one of the most reliable ways to steer format, tone and structure, and recommends that examples be relevant to your real task and varied enough that the model does not copy an unintended pattern.
Separate the parts. OpenAI's prompting guide suggests using Markdown headings and lists, or simple tags, to mark where instructions end and reference material begins. Anthropic recommends the same, wrapping instructions, context and examples in their own labeled sections to reduce misinterpretation.
Name the tone. OpenAI suggests using descriptive adjectives such as formal, friendly, professional or serious to set the tone.
Iterate. OpenAI's help article describes prompting as an iterative process: start with a prompt, review the response, and refine the wording, add context or simplify the request based on what you got.
A simple structure to reuse
OpenAI's guide describes a common layout for instructions: an identity (the purpose and style of the assistant), instructions (the rules to follow), examples, and context (the supporting information, usually best placed near the end). For everyday work requests, that translates into five short parts:
| Part | What to write | Example |
|---|---|---|
| Role | Who the AI should act as | "You are helping a project manager write to executives." |
| Task | The exact deliverable | "Write a status update email." |
| Context | Facts, audience, purpose | "Launch moved from May 5 to May 19 because vendor testing found two bugs." |
| Format | Length, structure, tone | "Under 150 words, three short paragraphs, calm and direct." |
| Material | Text to work from, clearly separated | "Notes: [paste]" |
You do not need every part every time. A one-line question is fine for a one-line answer. The structure matters most when the output will be shared.
Before and after
Before:
"Write an update about the launch delay."
After:
"You are helping a project manager write to a group of executives who have five minutes to read email.
Task: write a status update email about the product launch.
Context: the launch is moving from May 5 to May 19. Vendor testing found two bugs in the payment step. Both have fixes in progress. No customer data was affected. The executives mainly want to know the new date and whether more delays are likely.
Format: under 150 words, three short paragraphs, calm and direct, no jargon. End with one sentence on the next checkpoint date.
Only use the facts above. If something is missing, write [NEEDS INFO] instead of guessing."
The second prompt answers every question a new colleague would ask. The last line also tells the tool what to do when information is missing, which reduces invented details.
Match the prompt to the model
OpenAI's guide notes a difference between model types. It compares a reasoning model to a senior coworker you can hand a goal and trust to work out the details, and a standard GPT model to a junior coworker who performs best with explicit, step-by-step instructions. If your workplace tool lets you choose, a quick goal statement may be enough for a reasoning model, while a detailed checklist helps a faster general model.
Anthropic adds that when the order or completeness of steps matters, write the instructions as a numbered list.
Common mistakes
- Leaving out the audience. The same facts need different writing for a customer, a manager and a technical team.
- Asking for "better" without saying how. "Shorter," "more formal" or "add a clear next step" gives the tool something to act on.
- Mixing instructions into pasted text. Put pasted material under its own label so the tool does not treat a sentence in the document as an instruction.
- Accepting the first draft. Refining once or twice, as OpenAI suggests, is normal and usually fast.
- Pasting sensitive information without checking your workplace AI policy. Good prompting does not change what you are allowed to share.
Key takeaways
- Official guides from OpenAI and Anthropic agree: clear, specific prompts with context produce better results.
- Test a prompt by imagining a new colleague following it; if they would be confused, add detail.
- Explaining why you want something helps the model aim its answer.
- A few varied examples are one of the most reliable ways to control format and tone.
- Use a simple structure (role, task, context, format, material) and refine after the first response.