PRACTICAL AI AT WORK · CHATGPT · CLAUDE · GEMINI
A vague request usually produces a vague answer. Ask an AI assistant to “write something about our service” and it has to guess the audience, purpose, evidence and tone. A clear brief removes much of that guesswork. The skill transfers between OpenAI’s ChatGPT, Anthropic’s Claude and Google Gemini, even though their features and interfaces differ.
Think of prompting as briefing a capable colleague who has just joined your organisation. They can help immediately, but they do not automatically know which figures are approved, what a customer was promised or how your team defines a good result.
Give the assistant five things
Task: describe the work to be done. “Draft a follow-up email” is more useful than “help with sales”. Context: explain who will read it and what they already know. Evidence: supply the approved facts or source material. Output: specify the length and structure. Checks: identify anything it must not assume.
For example, a customer update might need a short explanation, three agreed actions and a clear next step. It may also need an instruction to flag missing dates rather than invent them. Those details are more valuable than a long introduction telling the assistant to be an award-winning expert.
A PRACTICAL WORKFLOW
A prompt you can use today
Draft a customer update from the notes below. The reader is a busy operations manager who attended our last meeting. Explain the current position in plain English, list agreed actions and finish with one clear next step. Keep it under 180 words. Use only the supplied facts. Do not invent dates, prices, commitments or names. Put anything missing in a separate “Questions to resolve” list. Notes: [insert approved notes].
This is a starting brief, not a magic formula. Try it with a routine task your team already understands. Compare the answer with the original notes and change the brief where the assistant misunderstood you.
Improve the answer in focused passes
First check the facts. Then check whether the response answers the real question. Only then refine the writing. Asking for a warmer tone before noticing a wrong delivery date is polishing the wrong result.
Useful follow-ups include “Which statements are unsupported by the notes?”, “What would the reader still need to know?” and “Shorten this while preserving the three agreed actions”. An assistant’s self-check is another aid to review; it does not independently prove its own answer is correct.
Choose a task before choosing a platform
Test the same brief in your organisation’s approved tools, using the same non-sensitive sample. Judge the results on factual accuracy, usefulness, editing effort and how comfortably they fit your workflow. Avoid deciding from one impressive answer. A tool that works well for a campaign draft may need a different setup for a spreadsheet or research task.
Do not assume that naming a document lets an assistant read it. Attach the approved file or use an authorised connection, and check that the response refers to the right version. Availability, limits and connected tools vary by subscription and administrator settings.
Turn good prompts into team habits
Keep a small shared collection of tested briefs with an example input, an acceptable output and the checks a reviewer should make. Give each brief an owner. When a process changes, update the example as well as the wording.
In a short team exercise, ask two colleagues to use the same brief on different sample cases. Their questions often reveal what the first author took for granted. That conversation is where prompting becomes a repeatable business skill.
Sources and further reading
Product information checked 4 November 2026. Features and availability depend on your plan and administrator settings. Examples and suggested workflows are illustrative.