Free Guide · Last updated October 20, 2026
How to Use ChatGPT, Copilot or Claude at Work
Night school for the assistant already on your work computer: short sessions, real tasks, your own pace.
This guide is for working professionals whose employer has given them an AI assistant, or who are choosing one. It covers OpenAI's ChatGPT, Microsoft 365 Copilot (Microsoft's assistant inside Microsoft 365 apps such as Outlook and Word) and Anthropic's Claude, and the method that works in all three. You get a step plan for this week, a starter path of free lessons, ready-to-paste prompts and an honest look at the limits.
The short answer on how to use ChatGPT, Copilot or Claude at work: use the assistant your employer already provides and approves, and apply the same four-step method in any of them. Give context, state the task and the format you want, iterate on the draft, and check the output before it reaches a colleague. Start this week with one real, low-risk task, such as a long email thread or a document you need to tighten.
Published
Free lesson · no account neededGetting an answer you can actually useYour first week with ChatGPT at work14 minStart the free lessonKey Takeaways
- The best AI assistant at work is usually the one your workplace already provides and approves, because approval answers the questions about what you may paste into it.
- The same four steps work in ChatGPT, Microsoft 365 Copilot and Claude: give context, state the task and the format, iterate, and check before sharing.
- You do not need to code. You write instructions in ordinary language, the way you would brief a capable new colleague.
- Practice on tasks you already have this week, in short sessions that fit around a full-time job, rather than on invented exercises.
- You stay responsible for anything you send. Check names, dates, numbers and commitments against the source every time.
What to give the assistant, and what to check, by task
| Work task | What to give the assistant | What to check before sharing |
|---|---|---|
| Summarize an email thread | The thread, your role, who will read the summary and the shape you want, such as five bullets with the decision first. | Every name, date and commitment against the original messages, and whether the summary missed a disagreement. |
| Draft a reply | The message you are answering, what you want to say, what you can and cannot promise, the length and the tone. | That every promise is one you can keep, that the tone suits the person, and that nothing was invented. |
| Tighten a document | Your own draft, the readers, a target length, what must stay unchanged and a request to list what was cut. | That figures, quotations and technical terms are unchanged and that nothing important was removed. |
| Prepare for a meeting | The topic, the attendees, your goal for the meeting and any background you already have. | Any figure or claim in the briefing, which you should confirm from your own sources. |
| Turn notes into actions | Your rough notes, the names of the people involved and the format, such as owner, task and deadline. | That each action has the right owner and date, and that nothing was assigned that nobody agreed to. |
Copy and Paste
Prompts You Can Use Today
Summarize a long email thread for someone who has not read it
I am a project coordinator at a logistics company. Below is an email thread with a supplier about a delayed delivery. Summarize it in five bullet points for my manager, who has not read it. Put the decision we need to make first, then the facts, then the open questions. Use plain, neutral language and do not add details that are not in the thread. [paste thread here]
Swap in your own role, the type of thread and who the summary is for. Remove names or sensitive details first if your workplace rules require it.
Draft a reply with a clear purpose, length and tone
Draft a reply to the message below. I am the account manager and the customer is upset about a late invoice. I want to apologize once, say the invoice will be corrected by Friday, and ask for their purchase order number. Keep it under 150 words, polite and direct. Give me two versions: one formal and one warmer. [paste message here]
Change the situation, the commitment and the deadline. Only promise in the reply what you know you can deliver.
Tighten your own draft without losing the facts
Here is my draft of a one-page proposal for an internal training program. The readers are department heads who have very little time. Tighten it to about 300 words, keep the headings, keep every figure exactly as written, and add a list of anything you removed so I can check it. [paste draft here]
Replace the document type and the readers. Asking for a list of removals makes your review much faster.
Prepare for a meeting by letting the assistant interview you first
I have a meeting tomorrow with the finance team about next year's budget for my department. Ask me up to five questions that would help you prepare a one-page briefing, and wait for my answers before writing anything. After that, write the briefing with these headings: Goal, Numbers I must confirm, Likely objections, Questions to ask.
Change the meeting and the headings to match your own situation. Answer the questions with real details; the briefing is only as good as what you give it.
Make the assistant check its own summary against the source
Review the summary you just wrote against the original text I pasted. List any claim in the summary that the original does not support, any number or date that differs, and anything important that was left out. Quote the line from the original for each item so I can verify it myself.
Use this after any summary. It does not replace your own reading of the original, but it shows you where to look first.
Free first-week lessons
Start with a free lesson
A free AI course for work should let you try the method before you commit to anything. Take one lesson per tool, in this order, and use each on a real task the same week.
The method
The four-step method that works in any AI assistant at work
ChatGPT, Microsoft 365 Copilot and Anthropic's Claude look different on screen, but you talk to all of them the same way: in your own words, with a clear task. These habits carry over from one tool to the next, so what you learn is not wasted if your employer changes tools next year.
Give context
Say who you are, who will read the result and what the situation is. One sentence about your role and your reader usually improves an answer more than any clever wording.
State the task and the format
Name the job (summarize, draft, compare, rewrite) and the shape you want back: five bullets, a short email, a table with three columns, a one-page brief.
Iterate
Treat the first answer as a draft. Reply with what to change: shorter, more formal, add the deadline, remove the second paragraph. Two or three rounds is normal.
Check before it reaches a colleague
Compare names, dates, figures and commitments with your source. You are the author of whatever you send, so you are also its reviewer.
Save what worked
When a prompt gives a good result, keep it in a note with blanks for the parts that change. A small library of reusable prompts for work is where the time savings come from.
Your step plan
Steps 1 to 3: set up before you type anything
Step 1: find out what your workplace provides and approves. Many employers have already chosen an assistant, and some have written a short policy about what may be pasted into it. Look on the intranet, ask your manager or ask the IT or security team. If nothing exists, ask anyway; it is a normal question and the answer protects you.
Step 2: pick one real task that is low risk and repeats. Good first candidates are summarizing a long email thread, drafting a reply when you already know what you want to say, tidying the wording of your own document, or turning rough meeting notes into a list of actions. Leave anything involving personal data, legal commitments or sensitive financial figures until you know your employer's rules.
Step 3: write the first prompt using the method. Start with one sentence of context, then the task, then the format, then any limits such as length, tone and a reminder not to invent facts. A prompt of three or four sentences is not too long; it is what the assistant needs to do the job. The sample prompts further down this page are written this way, so you can copy the shape and change the details.
Weekly rhythm
Steps 4 to 7: iterate, check and build a rhythm you can keep
Step 4: iterate at least once. Read the first answer, decide what is wrong with it, and say so. 'Shorter, and start with the decision' or 'this is too formal for a colleague I know well' are complete instructions. The assistant works from what you tell it, so specific feedback gets a better second draft.
Step 5: check the output against the source before it leaves your screen. Open the original email or document beside the answer and confirm each name, date, number and commitment. If the assistant summarized something, ask it to quote the line each point came from, then verify those lines yourself.
Step 6: save the prompt that worked. Paste it into a note with square brackets where the details change. After a few weeks you will have a personal set of prompts for work, such as one for weekly status updates and one for preparing a meeting.
Step 7: set a rhythm that fits around a full-time job. A realistic pattern is one short session on a weekday evening or lunch break to take a lesson, one or two small real tasks during the week, and a few minutes at the weekend to note what worked. Learning at your own pace in small sessions beats a single long push, because the habit is what you are building.
If a week gets busy, drop the lesson and keep one real task. Using AI at work on something you already had to do counts as practice, and it is the kind that sticks.
Choosing
Which AI assistant at work should you use?
Use the one your workplace already provides and approves. That answer is less exciting than a ranking, but it is the right one, because approval covers the questions you cannot settle yourself: what may be pasted in, who is responsible for the tool and what your employer expects. A tool you chose on your own may be a good tool and still be the wrong one for your job.
The three tools in this guide are OpenAI's ChatGPT, Microsoft 365 Copilot (Microsoft's assistant inside Microsoft 365 apps such as Outlook and Word) and Anthropic's Claude. Each has its own interface, and these products change often, so this page does not list features, plans or prices. Check your own tool's current screens and your employer's guidance for those details.
If your employer has given you nothing and you are free to choose, choose by the work you do. If most of your day happens inside Outlook and Word, an assistant that lives in those apps may fit naturally. If you work with text, ideas and documents of many kinds, a general chat assistant may suit you. Where you are allowed, try the same real task in each and keep the one whose answers you find easiest to check and improve.
Do not let the choice stall you. Because the method is the same in all three, a week spent learning in one tool transfers to the others. Switching later costs you the menus, not the skill.
Starter path
Your starter path: one free first-week lesson per tool
The path follows the order ChatGPT, Microsoft 365 Copilot, then Claude. If your employer gave you only one of them, take that lesson first and the others when you are curious. Each step is something you can try on your own work the same week.
First, the course Your first week with ChatGPT at work, lesson 'Getting an answer you can actually use', which is 14 minutes long. It covers the thing this whole page is about: turning a vague request into an answer you can put in front of someone. The same week, apply it to one email or paragraph you were going to write anyway. If you want more, the second free lesson in that course is 'Keeping a project's chats and files together'.
Second, the course Your first week with Microsoft 365 Copilot at work, lesson 'Clear your morning inbox before your next meeting', which is 11 minutes long. Try it on a morning when you have a meeting coming up and a full inbox, and check each summary against the message it came from. The next free lesson, 'Draft and tighten a document without starting from a blank page', moves on to drafting in Word.
Third, the course Why Claude?, lesson 'Meet Claude', which is 6 minutes long. Afterwards, hold a first conversation about a real task from your week; the next lesson in that course is 'Your First Conversation'. When you want to practice the method more deliberately, 'Context Is Everything' in Strategic Prompting and 'The Iteration Loop' in Master the Essentials cover giving context and improving a draft step by step.
Notice that the lessons are about different tools but ask the same things of you: give context, say what you want back, improve it and check it. That is why the path works even if you only ever use one of them at your job.
Tasks
What tasks can AI help with at work?
You do not need to know how to code to use AI at work. You write instructions in ordinary language, the way you would brief a capable new colleague, and you read the result critically.
Summarizing long text
Email threads, reports and meeting notes can be reduced to the points a busy reader needs. Ask for a specific shape, and ask for the source line behind each point.
Drafting and rewriting
Replies, updates and first drafts of documents, or a rewrite of your own text for a different reader. You supply the facts and the intent; the assistant supplies a starting version.
Preparing for meetings
Agendas, likely questions, objections to expect and a short briefing. Letting the assistant interview you first usually produces a more useful result.
Turning notes into actions
Rough notes become a list with an owner, a task and a date for each item. Check that nothing was assigned that nobody agreed to.
Explaining and learning
Ask for a plain-language explanation of a concept, a term or a process in your field, then ask for an example from a job like yours. Verify anything you plan to rely on.
Working with tables and numbers
Ask for help structuring a table, explaining a formula or describing a set of figures in words. Always recheck the arithmetic yourself.
Practice
How to practice on real work, at your own pace
Practice works best on tasks that already exist in your week. A lesson on its own gives you a pattern; a real email, report or agenda gives you a reason to remember it. Keep a simple list of the tasks you tried with an assistant and one line on how each went.
Here is a four-week shape that fits around a full-time job. In week one, take the first lesson for your tool and use it on one low-risk task. In week two, repeat that task type and add a second, such as meeting preparation. In week three, save your best prompts and share one with a colleague who asks how you did it. In week four, pick a task that usually drags and see whether the method makes it lighter.
Use the assistant to question your work as well as to produce it. Paste a draft and ask what a skeptical reader would object to, or what is missing for someone who was not in the project. This is often more valuable than asking for a new draft, and it keeps you as the author.
When an answer disappoints, change one thing at a time. Add the missing context, tighten the format, or split the job into two smaller jobs. Most weak answers trace back to a prompt that left out who the reader is or what a good result looks like.
Safety and judgment
Is it safe to use AI with work documents?
The honest answer depends on your employer's rules and on the tool you use, and this page makes no claims about how any tool handles data. What you can do is follow a simple discipline: use the assistant your workplace approves, read the policy if one exists, and ask when it does not.
Before pasting anything, ask three questions. Is this tool approved for this kind of material? Does the text contain personal data, customer information, contracts, unreleased results or passwords? Could I do the task with a shortened or anonymized version, replacing names with roles such as 'the client' and 'the supplier'?
If you are unsure, begin with material that is already public or that you wrote yourself and that has nothing sensitive in it. Move to more delicate work only when you know the rules. Asking once, early, is simpler than untangling a doubt later.
Responsibility stays with you. If an assistant helps with a message to a client, you are still the sender. Read it as if you had written it yourself, because as far as the recipient is concerned, you did.
Limits
Honest limits: what a few weeks of practice will and will not give you
A few weeks of regular, self-paced practice will give you a dependable method, a handful of prompts that fit your job and a better sense of which tasks suit an assistant. It can make routine writing, summarizing and preparation faster, though not for every task, and it will make you quicker at spotting weak answers.
It will not make you an expert in AI, and it will not make the assistant right. Assistants can state something wrong in a confident tone, invent a detail that sounds plausible, misread a table or miss the one sentence that mattered. The risk is highest with names, numbers, dates, quotations and references, so those are the first things to check.
It will also not replace your knowledge of the work. The assistant does not know your client's history, your team's habits or the unwritten reason a process exists unless you tell it. Your job is to bring that judgment and let the assistant handle the first draft of the routine parts.
Finally, tools and their menus change often. Learn the method first and the buttons second, and expect to relearn the buttons from time to time. If a lesson screen differs from what you see at work, trust your workplace's current version and apply the same method.
Learn to use AI at work, one real task at a time
Reading about an assistant and being good with it are different things. Nightschool AI is free to start, and the lessons are hands-on from the first one, so you can try the method on your own work this week. Begin with the ChatGPT lesson, then continue through Microsoft 365 Copilot and Claude at your own pace.
Frequently Asked Questions
How do I use ChatGPT at work?
First confirm that your employer allows it and what you may paste into it. Then pick one real, low-risk task, such as summarizing an email thread or drafting a reply. Write a prompt with context (your role and the reader), the task and the format you want, then ask for changes until it fits, and check names, dates and numbers before sharing. Nightschool AI's course Your first week with ChatGPT at work starts with the lesson 'Getting an answer you can actually use', which is 14 minutes long.
How do I use Copilot at work?
Microsoft 365 Copilot is Microsoft's assistant inside Microsoft 365 apps such as Outlook and Word. Check with your IT team how your workplace provides it, then apply the same method: give context, state the task and format, iterate and verify. A good first task is a busy morning inbox before a meeting, checking each summary against the original message. The lesson 'Clear your morning inbox before your next meeting' in Your first week with Microsoft 365 Copilot at work is 11 minutes long.
How do I use Claude at work?
Claude is Anthropic's AI assistant. If your workplace approves it, start with a conversation about one real task, giving your role, the reader, the task and the format you want, then refine the answer in follow-up messages. Verify facts and figures before they reach a colleague. The lesson 'Meet Claude' in the Why Claude? course is 6 minutes long, and 'Your First Conversation' follows it.
Which AI assistant should I use at work?
Use the one your workplace already provides and approves. Approval settles questions you cannot answer alone, such as what material is allowed and who supports the tool. If you are free to choose, pick by the work you do and test the same real task in more than one tool where permitted. The method (context, task and format, iterate, check) is the same in ChatGPT, Microsoft 365 Copilot and Claude, so your skills transfer between them.
Is it safe to use AI with work documents?
It depends on your employer's rules and the tool, so check your workplace policy or ask IT before pasting anything. Avoid personal data, customer information, contracts and passwords unless the tool is approved for them. When in doubt, anonymize by replacing names with roles, or practice on public material or your own non-sensitive text. Whatever the tool does, you remain responsible for what you send, so review every output.
Is there a free course to learn AI tools for work?
Yes. Nightschool AI is free to start, and it has first-week lessons for ChatGPT, Microsoft 365 Copilot and Claude that you can read without an account. The ChatGPT and Copilot courses each have six lessons, and their first two lessons are free to read. The Why Claude? course has six lessons and begins with 'Meet Claude'. Each lesson is built around a real work task you can try the same week.
What tasks can AI help with at work?
Common ones are summarizing long emails and documents, drafting and rewriting messages, preparing for meetings, turning notes into action lists, explaining unfamiliar concepts and structuring tables or plans. These are tasks where a first draft helps and where you can check the result against a source. Be more careful with anything involving numbers, legal commitments or personal data, and always review the output yourself.
Do I need to know how to code to use AI at work?
No. You work with an AI assistant by writing instructions in ordinary language, much like briefing a capable new colleague. What helps most is clarity: who the result is for, what you want done and what shape it should take. Those are writing and thinking skills you already use at work, not programming skills.
Is Nightschool AI affiliated with OpenAI, Microsoft or Anthropic?
No. Nightschool AI is independent and is not affiliated with or endorsed by OpenAI, Microsoft or Anthropic. ChatGPT, Microsoft 365 Copilot and Claude are products of their respective companies and are named here only to describe what they are. Our lessons teach working methods for using AI assistants on everyday tasks, and the products' own interfaces and features are described by their makers.
How long does it take to get comfortable with an AI assistant at work?
Many people get a first useful result in a single session, once they try the method on a real task. Becoming comfortable usually takes a few weeks of small, regular practice rather than one long study block. A pattern that suits a full-time job is one short lesson session, one or two real tasks during the week and a quick weekend review of what worked.