A good AI prompt states a role, a task, the context and the format you want back. Cover those four and ChatGPT, Claude, Gemini and Copilot will usually return something useful on the first try.
Most disappointing answers are a briefing problem, not a model problem. A vague input produces a vague output every time, and the fix is a set of learnable patterns, not a dark art. This guide covers them, with examples that work across all four tools.
Why do most prompts fall flat?
Most people type into an AI the way they would run a Google search: short, vague, no context. That works for “weather in Manchester” and fails for a tool that can write, analyse, plan and summarise. ChatGPT, Claude, Gemini and Copilot are capable assistants waiting to be briefed, and the clearer the brief, the better the output.
The most common prompting mistakes
Too short and vague. “Write me an email” gives the AI almost nothing. Who is it to, what is it for, what tone?
No context. The AI does not know who you are or what you are trying to achieve unless you say.
No format guidance. If you want bullets, a table or two paragraphs, say so.
Treating it like a search engine. Full sentences beat keywords.
Giving up after one attempt. You can follow up, refine and ask for changes.
What is the anatomy of a great prompt?
A great prompt has four parts: Role, Task, Context and Format (RTCF). You do not need all four every time, but knowing each one helps you build better requests by instinct.
Role: tell the AI who it is
A role sets the tone for everything after it. “Act as an experienced HR manager” or “You are a plain-English editor” tells the model how to approach the request.
Task: be specific about what you want
“Write a short email” is fine. “Write a two-paragraph email declining a meeting request, keeping the tone warm but firm” is much better. Specificity is not over-engineering, it is being clear.
Context: tell it what it needs to know
Context is anything the AI could not reasonably guess: your audience, the situation, the background, the constraints. The more relevant context you include, the less the model has to improvise.
Format: say how you want the output
A bulleted list, a table, a numbered plan or a paragraph: say which, and add a length limit if you need one. Anthropic’s prompt engineering guidance names output format as one of the most useful things to specify.
A quick example
Weak prompt: “Write a LinkedIn post about our new product.”
Stronger prompt: “You are a B2B copywriter. Write a LinkedIn post announcing a new project management tool for small teams. The tone should be conversational and confident. Keep it under 150 words. End with a question to encourage comments.”
Same task, completely different output quality.
Which prompting techniques are worth knowing?
Ten techniques cover most situations. Try a few, see which fit your workflow and build from there. They work across all major AI tools.
Give it a role first. “Act as a financial analyst” sets the register immediately.
Specify the audience. “Explain this to someone with no accounting background” changes the output completely.
Ask for a specific format. “Format this as a table with three columns: task, owner, deadline.”
Set a length limit. “In under 100 words” stops the padding.
Give examples. Paste in writing you like and say “match this tone”. Examples are one of the fastest ways to steer quality.
Ask it to think step by step. For anything analytical, this tends to produce more accurate, reasoned responses.
Use positive instructions. “Write in a warm, direct tone” is clearer than “don’t be too formal”.
Ask for alternatives. “Give me three versions” opens up options you had not considered.
Follow up and iterate. If the first response is 80% there, say what to fix: “Make it shorter” or “The second paragraph needs to be more assertive.”
Use a prompt improver. The ivee app sits inside ChatGPT, Claude, Gemini and Copilot. Press Shift+Command+Space on a draft and it rewrites the prompt and suggests which model to use, before you send.
The OpenAI prompt engineering guide covers several of the same principles in more detail. The fundamentals transfer between tools.
How does prompting differ between ChatGPT, Claude and Copilot?
The principles are the same everywhere, but each tool has its own character.
ChatGPT
ChatGPT handles a wide range of tasks well, including writing, summarising, coding and brainstorming. It responds well to step-by-step instructions and gives longer, more structured answers by default, so ask for brevity if you want it.
Good for: drafting, ideation, Q&A, explaining complex concepts
Tip: use custom instructions to set a persistent role or style
Claude
Claude, made by Anthropic, is strong at nuanced writing, long documents and careful reasoning. It handles large pastes of text well and tends to be more cautious about making things up, which helps when accuracy matters.
Good for: analysis, long-form writing, editing, research summaries
Tip: give it detailed context, and do not be afraid to include background
Microsoft Copilot
Copilot is built into Microsoft 365, across Word, Excel, Outlook and Teams, which makes it useful in a work context because it can reference your actual documents and emails. We covered the most useful Copilot forward-slash prompts for a practical shortcut list.
Good for: work tasks inside Microsoft 365, meeting summaries, drafting in Outlook
Tip: type “/” in any Copilot chat to get ready-made commands
How do you get a whole team prompting well?
A team prompts well when the best prompts are shared, owned and safe to use. Left alone, every person improvises, and the quality of work depends on who happened to learn the trick.
Share the prompts that work. Structured, reusable prompts beat everyone starting from a blank box.
Give someone ownership. An AI champion finds, tests and shares the good ones. See what the AI champion role involves.
Set the data rules first. Start with whether you can put company data into ChatGPT and what AI governance covers.
Train for the real work. How to train your team in AI covers the rollout, and ivee runs practical, hands-on AI training tailored to how your team works.
The ivee app supports this with team access, so the prompt that works for one person becomes the way the whole team does it.
FAQs: writing prompts for AI
What is a prompt in AI?
A prompt is any input you give an AI tool: a question, an instruction or a request. The quality of your prompt directly affects the quality of the response.
How long should an AI prompt be?
There is no perfect length, but more context usually helps. A one-line prompt works for simple tasks. For a report or an analysis, a few sentences covering context, format and tone will get far better results.
What is prompt engineering?
Prompt engineering is the practice of crafting inputs to AI models to get the best outputs. In practice it means being clear, specific and giving the model enough context. No coding required.
Is there a tool to help me write better prompts?
Yes. The ivee app is a Mac app that sits inside ChatGPT, Claude, Gemini and Copilot. Press Shift+Command+Space on a draft prompt and it rewrites it and suggests which model to use.
How do I get my team to write consistent prompts?
Share the prompts that work, agree who owns them, and set simple rules on what data can go in. An AI champion usually keeps the shared set current, and team access in the ivee app makes the best prompt everyone’s prompt.
Put it into practice
Writing better prompts is one of the fastest ways to get more from any AI tool, and the skill compounds. The quickest way to see the difference is to send two versions of the same prompt and compare.
