ChatGPT prompts for work only do their job once you include three things: what the task is about, what you want to do with it and what form you want the answer in. Leave those out and ChatGPT fills in the gaps itself, which rarely gives you exactly what you were looking for.
How much context you provide ultimately matters more than the exact words you choose.
Below are six categories that cover most office tasks people use ChatGPT for, each with an example prompt you can copy and adapt straight away. The last example adds a step that most prompt tips leave out: how to ask ChatGPT to flag what in its own answer needs checking.
Prompts for writing and rewriting
Writing and rewriting is the most common ChatGPT task, from an email to a product description. The quality depends mainly on how much context you give about who will read the text and what tone it should have.
Write an email to a customer whose delivery is a week late.
Tone: businesslike but not formal.
Length: no more than 100 words.
End with a concrete alternative.If you want to improve an existing text rather than write something new, include the original and say specifically what you want changed: shorter, more direct, less jargon.
These prompts work just as well if you write them in a language other than English.
Prompts for summarising and processing documents
Summarising works best when you say up front what you are going to do with the summary, because that decides which details matter and which do not.
Summarise this report in no more than 150 words, for a colleague who does not have time to read the whole thing.
Focus on the conclusions and the risks mentioned.
Leave out background information.
[paste the text here]With longer documents, feeding them in sections works better than pasting everything at once. Especially on the free version, you will otherwise hit a length limit sooner.
Prompts for analysing and structuring information
Analysing and structuring means asking ChatGPT to bring order to information you supply yourself. Do not ask for facts the model has no access to.
Compare these two quotes on price, delivery time and warranty terms.
Put the result in a table.
For each point, say which quote is more favourable and why.
[paste quote 1 here]
[paste quote 2 here]The same approach works for turning loose notes into an overview, or for sorting a long list of customer questions into categories.
Prompts for planning and organising
Planning and organising with ChatGPT works well for setting up a structure, and less well for filling in factual details that you have to supply yourself.
Draft an agenda for a 30-minute meeting about the Q3 planning.
Participants: 4 team members.
Goal: set priorities for the next two months.
Add an estimated duration for each agenda item.For bigger planning tasks, such as a project plan with several phases, it works better to build up the conversation step by step than to ask for everything in one prompt.
Prompts for learning and getting explanations
Learning and getting explanations from ChatGPT works best when you say what level you are already at. Without that context, you often get an explanation that is too basic or too technical.
Explain what a DPIA is.
I work at a marketing agency and have no legal background.
Use a concrete example from our sector.
No more than 200 words.Feel free to follow up with ‘explain that last point again, but more simply’ instead of asking the whole question again. ChatGPT keeps the context within the same conversation.
Prompts for getting your own work checked
Getting your own work checked is the category missing from most prompt lists, yet it is exactly the step that catches an AI error before it goes out the door.
Check this text for factual claims that I need to verify myself.
Mark every figure, every name and every reference to a source or law separately.
Do not comment on the tone, only on the factual support.
[paste the text here]This prompt works so well because it does not ask ChatGPT to confirm that something is correct. It asks it to point out what you still need to check yourself.
A language model can hallucinate: to the model, a wrong answer is just as plausible as a right one, and that is exactly why this verification step is needed.
What makes good ChatGPT prompts for work?
A good ChatGPT prompt contains four elements: a role for the model, a clear goal, the context the model cannot guess, and the form you want the answer in. The more explicitly you fill in those four, the less ChatGPT has to fill in on its own.
Assigning a role is the element people skip most often, even though it steers the tone of the whole answer.
Can you reuse a ChatGPT prompt?
Yes, you can reuse a ChatGPT prompt by saving the fixed part, such as the role and the format, and changing only the content each time. That way you gradually build your own library of prompts that have already proved themselves on your tasks.
For tasks with several steps, such as analysing first and only then writing, it works better to split them into separate prompts, one after the other.
Frequently asked questions about ChatGPT prompts
Beyond individual examples, ChatGPT prompts also depend on fixed settings and on how the model works.
Can you set standing instructions in ChatGPT?
Yes, you can set standing instructions in ChatGPT with custom instructions, which apply to every new conversation. You fill in two questions, each with up to 5,000 characters on the paid plans (since 15 July 2026) and 1,500 characters on Free, for example about your job and the tone you want. According to OpenAI, custom instructions are available on all plans. An ongoing conversation does not pick up the change, so start a new chat after you make a change.
Why does the same prompt give a different answer every time?
The same prompt often gives a different answer because language models like the one behind ChatGPT are non-deterministic by default. Microsoft writes in its Azure OpenAI documentation that if you ask a chat model the same question repeatedly, you are likely to get a different answer each time. A fixed structure with role, goal, context and format makes the answers more predictable, although the wording will still vary.
Does a ChatGPT prompt also work in Claude or Gemini?
A good ChatGPT prompt generally works in Claude or Gemini too, because Anthropic and Google recommend similar principles. Anthropic advises treating Claude like a new colleague who knows nothing about how you work, and being specific about format and constraints. For Gemini, Google recommends clear instructions with context, and mentions specifying the response format as a way to steer the outcome. Do test a prompt again after moving it, though, because each model responds differently to the same text.
How do you learn to write good prompts consistently?
You learn to write good prompts consistently by not stopping at individual examples, but building a fixed way of working around your own recurring tasks.
A list of ready-made phrases is not enough for that. You also need to know when to adjust which element and how to check the output before you use it.
You build that way of working step by step in LearnLLM’s ChatGPT Course, with the elements of a usable prompt and five standard checking questions that you apply to your own work rather than to a made-up example.
