What is a prompt? How to write a good one, with 12 templates for work

What does prompt mean?

Cambridge Dictionary lists several senses of prompt, and one of them is the AI sense, “an instruction given to an artificial intelligence … by a human using natural language rather than computer language” (Cambridge Dictionary). Natural language means you write to the tool the way you’d write to a colleague, with no code involved. OpenAI describes a prompt as the text input that starts a conversation or triggers a response, and adds that it can also be an image or audio (OpenAI).

If words like token, context window or hallucination come up in the same conversation, my AI dictionary for HR explains 20 of them in plain English.

How do you write a good prompt?

A good prompt answers five questions. The split is my own proposal, built on what Microsoft, Google and Anthropic recommend. On my ChatGPT prompts for HR page I start from role, task, context, constraints and format. Here I swap the role for source material, because at work I think a missing document causes more bad answers than a missing role. The first question always belongs in the prompt. Add the others when the answer needs to get better.

1. Task. What should the AI do? Start with a verb, such as write, summarize, compare, suggest or review. “Write an email to all employees” gets you further than “email about expenses”.
2. Context. Why, and for whom? Describe the purpose, the reader and the situation. Without it, the answer stays general enough to fit any workplace.
3. Source material. What should it work from? Paste the text, attach the file or point to the document. Without a source, the AI fills the gaps itself, sometimes with things that aren’t true.
4. Format. What should the answer look like? Give the length, tone, language and form, for example a table, five bullet points or no more than 150 words. If you have a text that worked before, paste it as an example. Anthropic calls examples one of the most reliable ways to steer format, tone and structure.
5. Limits. What should it not do? Set boundaries and ask for checks, like “don’t invent numbers”, “flag anything you’re unsure of” or “ask me questions before you write”. Give the reason too. Anthropic recommends explaining why a rule exists, because the model can then apply it to cases you didn’t list.

Basic template to copy

Task: [write, summarize, compare or suggest] [what]
Context: [why you need it, who will read it and in what situation]
Source: Use only the text below.
[paste or attach]
Format: [length, tone and form, for example no more than 150 words in bullet points]
Example: [optional, paste a text in the style you want]
Limits: [what the AI should not do]. Write [CHECK] wherever you're unsure.

Microsoft’s list is goal, context, source and expectations, and Google’s is persona, task, context and format. Both say you don’t need every part in every prompt (Microsoft, Google). Google and Anthropic also suggest giving the AI a role, such as “You are an experienced recruiter.” It can help the tone, but I treat it as optional because it doesn’t replace the context.

Test your prompt on a colleague

Anthropic suggests thinking of the AI as “a brilliant but new employee who lacks context on your norms and workflows” and offers a test that works in any tool (Anthropic). Show your prompt to a colleague who knows little about the task and ask whether they could do it. If they’d be confused, the AI will be too. If I could keep only one piece of prompting advice, it would be this one. What you’d have to explain to that colleague is what the AI is missing.

Before and after for the same task

Say you work in HR and the rules for claiming work expenses change next quarter. A typical first prompt looks like this.

Write an email about the new expense rules.

The AI doesn’t know what’s changing, who’s reading or how long the email should be, so it guesses. You get a generic email that you’ll have to rewrite. The second version answers the five questions.

Task: Draft an email to all employees saying that the rules for claiming expenses change on [date].

Context: We have [number] employees and many read email on their phones. Last time the rules changed, HR got a lot of questions about which receipts count.

Source: The new rules are below. Use only what's in them.
[paste the rules]

Format: A subject line and no more than 150 words. Start with what changes and from when. End with who to ask. Plain language, written to "you".

Limits: Don't promise anything that isn't in the rules. Write [CHECK] wherever you're unsure.

The second prompt is longer, but everything in it is something you’d have to tell a new colleague who was writing the email for you anyway. The line about questions to HR gives the AI a reason to explain which receipts count, and [CHECK] shows you where to read extra carefully.

How much the prompt matters showed up in my own test of ChatGPT, Claude, Gemini and Copilot on HR tasks. For one task I used a deliberately loose prompt, the way most people write. Claude built an interactive dashboard, while Gemini produced small images where the text overflowed. My guess is that a better prompt would have narrowed the gap. The whole test is in Which AI tool is best for HR?

Follow up instead of starting over

Treat the first answer as a draft. OpenAI and Microsoft both tell you to expect a few rounds, and Microsoft notes that the same prompt can give different responses each time. Keep these follow-ups at hand.

• “Cut this in half and keep what matters most.”
• “Rewrite this for someone who doesn’t work in HR.”
• “Read this as a skeptical employee. What’s missing?”
• “Give me three other ways to open.”
• “Which claims in your answer should I check?”
• “Turn this into a table with one row per action.”

If you’re not sure what you want, let the AI help you write the prompt.

I want help with [task]. Ask me questions, one at a time, until you know what you need. Then write the prompt you would want to receive, and wait for my OK before you do the task.

12 prompt examples for work, ready to copy

You can use these templates in Copilot, ChatGPT, Claude or Gemini. Replace what’s in [brackets] and paste your material where the template says so. Use the AI tool your organization has approved, sign in with your work account, and read what to keep out of a prompt before you paste anything about a specific person. For job ads, survey comments, interview scorecards and policy drafts, use the 13 patterns in ChatGPT prompts for HR. The HR templates below cover other tasks.

Prompt templates for managers

1. Turn meeting notes into decisions and next steps

Here are my notes from [the meeting]. Summarize them in three parts.
1. Decisions we made
2. Actions, with owner and date where the notes say so
3. Questions that are still open

Use only what's in the notes. Write "missing" where an owner or a date isn't stated. No more than half a page.

[paste the notes]

2. Write a status update for leadership

Turn my notes below into a status update on [project] for [the leadership team or my manager]. They have two minutes to read it.

Use this structure.
1. Status in one sentence (on track, at risk or off track) and why
2. What we've done since [date]
3. Risks, and decisions I need from them
4. Next steps with dates

No more than 200 words. Don't make the status sound better than my notes support.

[paste your notes]

3. Delegate a task with a clear brief

I'm handing a task to [a colleague or a role]. Help me write a short brief they can work from without coming back with ten questions.

The task: [what needs doing]
Why it matters: [the purpose]
Deadline: [date]
What good looks like: [the result I expect]
What they can decide on their own: [scope]
What they should check with me first: [limits]

Write the brief in plain language, no more than 200 words. Then list the questions a new person would still have after reading it.

4. Find the weak spots in a proposal

You're a critical but constructive colleague on the leadership team. Here is my proposal on [topic].

[paste the proposal]

Look for weaknesses before I present it. Give me
1. the three strongest objections and who is likely to raise them
2. assumptions I make without evidence
3. questions I should be ready to answer
4. what would make the proposal easier to say yes to.

Be concrete and skip what already works.

Prompt templates for HR

5. Write interview questions for a role

Here are the requirements for [role].
[paste the requirements]

Write an interview guide for a [number]-minute interview. For each requirement, give
1. two questions about situations the candidate has been in
2. what a strong answer includes
3. one follow-up question.

Every candidate should get the same questions. Leave out questions about age, family plans, health, religion or anything else unrelated to the job.

AI can help you prepare the interview, but people assess the candidates. I go through where AI fits in hiring in AI in recruiting.

6. Turn a policy into a one-page summary for employees

Here is our policy on [topic].
[paste the policy]

Write a short version for employees, no more than one page, in plain language and written to "you". Start with what the policy means for them day to day. Then add the five questions employees are most likely to ask, with answers based only on the policy.

Don't change the meaning. Where the policy is unclear, write [UNCLEAR IN POLICY] instead of guessing.

7. Turn a process into a checklist for managers

Below is our process for [for example a resignation, parental leave or a new hire's first week].
[paste the process description]

Turn it into a checklist for line managers. Group the steps by when they happen, mark which steps HR owns and which the manager owns, and keep each step to one line.

Use only what's in the process. At the end, list any step where the description is unclear or the owner is missing.

8. Write a respectful rejection email to interviewed candidates

Write an email to candidates who interviewed for [role] but didn't get the job. They met [who] and spent [how long] in the process.

The tone should be warm, direct and respectful. Thank them for their time, give the decision in the first two sentences and say [whether they're welcome to apply for future roles]. No more than 120 words.

Write [first name] where the name goes, and I'll add the names myself. Don't give reasons for the decision and don't promise feedback unless I add that here.

Prompt templates for office work

9. Reply to an email

Here is an email I received.
[paste the email, without personal details that aren't needed]

Write a reply where I [the message, for example decline but suggest another date]. The recipient is [the relationship, for example a client I know well]. The tone should be [for example friendly and direct]. No more than [number] sentences. Reply in the same language as the email.

10. Summarize a long document for a specific reader

Summarize the attached document for [the reader, for example the leadership team, who have five minutes before the meeting].

Give me
1. the main message in two sentences
2. the three most important facts, with page numbers
3. what requires a decision or action from [the reader].

Write "not stated" if something is missing from the document. Don't add your own conclusions.

11. Make a text shorter and clearer

Make the text below shorter and easier to read for [the audience]. Keep all the facts and don't change any numbers, dates or names. Replace jargon with everyday words where you can. Show the new text first, then a list of what you removed.

[paste the text]

12. Build an Excel formula

I have a table in Excel. Column A has [dates], column B [department] and column C [amount]. The headers are in row 1 and the data runs to row [number].

I want to [for example total the amount per department and month]. Write the formula, explain step by step what it does and tell me what to check if the result looks wrong. I use [Excel for Microsoft 365, Excel 2019 or Excel for Mac].

Name your version. Microsoft notes, for example, that XLOOKUP isn’t available in Excel 2016 or Excel 2019, so a formula built on it won’t work there (Microsoft).

How to build a prompt library for your team

A template that only lives in your chat history helps nobody else. As a manager, you can turn your prompts into something the whole team shares.

1. Pick three tasks that come up every week. Choose ones that take time but don’t need hard judgment calls, like weekly updates, meeting notes and answers to common questions.
2. Write one template per task that answers the five questions, and test it on a real task without personal data.
3. Save the templates where everyone can find them. A shared document works in any tool. Saved prompts in Copilot are only visible to you, and sharing a prompt link takes the paid Microsoft Copilot (work or school) license. In ChatGPT you can share a project, with its instructions and files, on every plan from Free to Enterprise. In Claude, sharing projects takes a Team or Enterprise plan.
4. Review the library once a month. Remove what nobody uses and add what someone on the team has figured out. If the library is there but the team still doesn’t use it, my one-page ADKAR checklist for AI helps you see where they’re stuck.

To practice prompts like these on HR tasks at your own pace, try my free AI course for HR. It takes about 60 minutes and ends with a certificate of completion.

Does prompting differ in ChatGPT, Claude, Copilot and Gemini?

The five questions apply in all four tools, so the prompts themselves change little. The differences are in where each tool gets your material and where you save the instructions you use often. This is how it looks as of November 2026.

Based on each vendor’s help pages, checked 1 November 2026. Links under Sources.

Claude also has Cowork on paid plans. On desktop it can read and write your local files and work through a task in several steps, so I’d write the prompt as a task brief. Describe the finished result, where it should be saved and what Claude should check with you before it changes anything. I explain how it works in What is Claude Cowork?

If you’re choosing a tool for the whole company, start with Copilot vs ChatGPT.

Common prompting mistakes

• Cramming everything into one prompt. Break big tasks up. Ask for an outline first, then for one section at a time.
• Passing on numbers, sources and legal references unchecked. AI answers sound convincing even when they’re wrong, and Microsoft tells users to review and verify the responses. What you send on is your responsibility.
• Using prompts from the internet as they are. Long ready-made prompts look impressive but weren’t written for your situation. Use them as a starting point and add your own context. That goes for the twelve on this page too.

What to keep out of a prompt

Whatever you type goes to the vendor. My rule of thumb is that personal data about colleagues, candidates or customers stays out of personal AI accounts entirely, names included. A work account with a contract protects more than a personal one, but privacy and employment law apply in both. In your work tool, the categories below go in only if your organization has explicitly decided how. Passwords and login details stay out of every tool.

• Special category data. Article 9 of the UK and EU GDPR covers health, racial or ethnic origin, political opinions, religious or philosophical beliefs, union membership, sex life or sexual orientation, and genetic or biometric data used to identify someone, and prohibits processing it unless an exception applies (UK GDPR, EU GDPR). In HR it’s usually sick leave, accommodations and union membership.
• Employee medical information in the US. The ADA covers employers with 15 or more employees. It requires them to keep medical information from disability-related inquiries or medical exams, and anything an employee discloses voluntarily, as a confidential medical record that may be shared only in limited circumstances (EEOC, ADA.gov).
• Criminal records and allegations. Under Article 10 of the UK GDPR, data about criminal convictions and offenses may be processed only under official authority or when the law authorizes it with safeguards (UK GDPR). Internal investigations often contain it.
• ID and account numbers. Social Security numbers, National Insurance numbers and bank details. A prompt almost never needs them.
• Pay and assessments of named people. Salaries, performance ratings, warnings and conflicts between colleagues.
• Confidential business information and customer data. Contracts, non-public financials, customer records and strategy.

What to do instead

• Remove the name and any detail that points to one person. In a team of eight, a job title or a sick-leave date can be enough.
• Use the tool your organization has approved and sign in with your work account. If your rules say nothing about AI, ask before you try.
• Take cases involving health, accommodations or suspected misconduct to HR or legal before any of it goes into an AI tool.

Say a manager is preparing a return-to-work meeting. A first draft of the prompt might look like this.

Write talking points for my meeting with [full name] in payroll, who has been off with burnout since March and is back on Monday.

A safer version does the same job.

Write talking points for a return-to-work meeting with an employee who has been on leave for several months. Keep it supportive and focused on how we plan the first few weeks.

The second version leaves out the name, the team, the dates and the diagnosis, and the AI can still write the talking points.

Can your chats be used to train the models?

It depends on the tool and on whether you use a personal or a work account.

Based on each vendor’s help pages, checked 1 November 2026. Links under Sources.

If people already use personal accounts for work, that’s shadow AI, and I go through what leaders can do about it in Shadow AI. If your organization has no rules yet, the AI governance framework includes a one-page AI policy outline you can start from.

Frequently asked questions

What is a prompt in AI?

A prompt is the instruction or question you give an AI tool such as ChatGPT, Copilot, Claude or Gemini to get a response. It can be a short question or a long instruction with attached documents. Cambridge Dictionary defines it as an instruction given to an artificial intelligence by a human using natural language.

What does prompt mean outside AI?

As a verb, to prompt means to make something happen or to make someone say or do something, and to remind an actor of a forgotten line. As an adjective, prompt means done quickly or on time. A prompt is also the sign on a computer screen that it’s ready for a command, and a writing prompt gives you a topic to write about.

How do you write a good prompt?

Start with what the AI should do, using a verb. Then add why and for whom, what material it should work from, what the answer should look like and what it should not do. If you have a text in the style you want, paste it as an example. Follow up on the answer instead of starting over.

What is a prompt template?

A prompt template is a ready-made prompt with gaps, like [recipient] or [paste the text], that you fill in each time you do the same kind of task. Save the templates that work where your whole team can find them. This page has a basic template and 12 templates for work.

What is prompt engineering?

OpenAI describes prompt engineering as the process of designing and optimizing prompts to guide a language model’s responses. At work it mostly means writing clear instructions, giving context and source material, and following up on the answers.

How long should a prompt be?

As long as the task needs. Microsoft says a clear goal is all that’s required. Add context, source material, format and limits when the first answer comes back too generic. Google’s advice for Gemini is to keep the request brief but specific.

What should you never put in a prompt?

In a personal AI account, leave out all personal data about colleagues, candidates or customers, names included, and confidential company information. In the tool your organization has approved, follow its rules, and add sensitive data such as health information, union membership or ID numbers only if the organization has decided how. Passwords never go in, whatever the tool.

Does the same prompt work in ChatGPT, Claude, Copilot and Gemini?

Mostly, yes. The parts of a good prompt are the same in all four. The answers won’t be identical, and Microsoft notes that the same prompt can give different responses each time. The main difference is how each tool gets your material, for example by typing / to pick a work file in Copilot or @ to tag a Drive file in Gemini.

Sources

• Cambridge Dictionary: prompt
• OpenAI Help Center: Prompt engineering best practices for ChatGPT
• Microsoft Support: Get started writing prompts in Microsoft Copilot
• Google Workspace: Writing effective AI prompts for business
• Anthropic: Prompting best practices
• OpenAI Help Center: ChatGPT Custom Instructions
• OpenAI Help Center: Projects in ChatGPT
• OpenAI Help Center: Connected apps in ChatGPT
• Claude Help Center: What are projects?
• Claude Help Center: Understanding Claude’s personalization features
• Claude Help Center: Use Google Workspace connectors
• Microsoft Support: Customize how Microsoft Copilot responds to you
• Microsoft Support: Save your favorite prompts to use again
• Microsoft Support: Share your best prompts with others
• Google: Use Gems in Gemini Apps
• Google: Save info and reference past chats in Gemini Apps
• Claude Help Center: Get started with Claude Cowork
• Microsoft Support: XLOOKUP function
• OpenAI Help Center: How your data is used to improve model performance
• Microsoft Learn: Microsoft Copilot Chat Privacy and Protections
• Anthropic Privacy Center: Is my data used for model training? (consumer plans)
• Anthropic Privacy Center: Is my data used for model training? (commercial products)
• Google: Gemini Apps Privacy Hub
• Google Workspace: Generative AI in Google Workspace Privacy Hub
• legislation.gov.uk: UK GDPR, Article 9 (special categories of personal data)
• legislation.gov.uk: UK GDPR, Article 10 (criminal convictions and offences)
• EUR-Lex: Regulation (EU) 2016/679 (GDPR), Article 9
• EEOC: Enforcement guidance on disability-related inquiries and medical examinations of employees under the ADA
• ADA.gov: Introduction to the Americans with Disabilities Act