A "prompt" is just what you type to an AI tool. The quality of what you get back depends heavily on how clearly and specifically you ask. You don't need to become a "prompt engineer" or study any technical frameworks — a few simple habits make a real difference to the results you get for everyday tasks.
The single biggest improvement: add context
Most weak prompts fail because they give the AI too little to work with. The AI has no idea who you are, what you're trying to do, or what a good answer looks like for you. Adding a few sentences of context is the single highest-impact change you can make.
Compare these two prompts:
❌ Too vague
"Help me prepare for a difficult conversation."
✅ With context
"I need to have a conversation with my flatmate about the fact that they consistently leave shared spaces messy and don't contribute to cleaning. We've lived together for 6 months and things have been getting worse. I want to raise this without it becoming an argument or making the living situation awkward. Can you help me think through how to approach this conversation, and suggest some opening lines I could use?"
The four elements of a strong prompt
For most everyday tasks, a good prompt includes:
The task — what you want the AI to do ("write", "explain", "suggest", "summarise", "help me think through")
The context — relevant background information about your situation
The format — how you want the answer structured (bullet points, a short paragraph, step-by-step, a list of 5 options)
The tone — formal or casual, detailed or brief, direct or gentle
You don't need all four every time — but including more usually gets a better result.
Specifying what you want AI to produce
Being explicit about the output format saves a lot of back and forth. If you want a list, ask for a list. If you want a short paragraph, say so. If you want options to choose from, ask for options.
💡 Format examples
"Give me 5 different ways I could open this conversation, as a numbered list." "Summarise this in 3 bullet points — one sentence each." "Write a single paragraph, no more than 80 words." "Give me the pros and cons in two separate lists."
Giving AI a role to play
Sometimes it helps to tell the AI what kind of expert or perspective to adopt. This can improve the quality and relevance of the response for specific tasks.
💡 Role-framing prompts
"Act as an experienced HR manager and advise me on how to handle this situation: [describe situation]." "You are a plain-English legal explainer. Summarise what this clause means in everyday language: [paste clause]." "As a personal finance coach, help me think through whether I should pay off my overdraft or start a savings account first, given: [your details]."
Refining as you go
The best prompts are often the second or third version, not the first. After you see an initial response, you can refine it with follow-up instructions:
- "Make this shorter — aim for half the length."
- "The tone is too formal. Make it more conversational."
- "The first option is best — can you give me three variations of that one?"
- "Add a section at the end explaining what to do if they say no."
Saving prompts that work
When you find a prompt structure that works well for a recurring task — writing a type of email, planning a weekly review, brainstorming ideas — save it. You can store these in your notes app and adapt them each time rather than starting from scratch.
✅ One-sentence rule
If your prompt is one sentence with no context, it will probably produce a generic result. Add at least three to four sentences of detail — who you are, what you're doing, and what a good answer looks like. That's usually enough.
A few patterns worth knowing
"Before you answer, ask me any clarifying questions you need" — useful for complex tasks where missing context would produce a generic response. The AI will ask what it needs before generating the output.
"Here's what I've already tried: [...]" — helps the AI avoid repeating advice you've already considered.
"Check your answer for accuracy" — prompts the AI to review its own response, which can catch obvious errors (though it's not a substitute for your own verification).