A prompt is just a set of instructions
A prompt is whatever you give an AI tool to work from: a question, a task, some background, an example. The tool predicts a response that fits what you wrote. If the request is vague, the tool has to guess what you meant, and its guess may not match yours.
Think of how you would brief a capable helper who knows nothing about you. “Write something about dogs” gets you something generic. “Write a friendly 150-word note for new neighbors explaining our street’s rules about leashes” gets you something you can use. Prompt writing is mostly the habit of saying what you would otherwise leave in your head.
Better prompts improve the odds of a useful answer, but they do not make the output reliable. The tool can still be wrong, so the habits in checking AI output still apply.
The main ingredients
1. The task
Say plainly what you want done. Use a clear verb: summarize, compare, rewrite, list, explain, outline, proofread.
2. Context
Include the background the tool cannot know: who the audience is, what the purpose is, what you have already tried, and what the situation looks like.
- Weak: “Explain budgets.”
- Better: “Explain how a monthly household budget works to someone who has never made one. They are in their first apartment and earn a steady monthly paycheck.”
3. Constraints
Set boundaries so the result fits your needs: length, tone, reading level, format, what to include and what to leave out.
- “Keep it under 120 words.”
- “Use a friendly, plain tone with no jargon.”
- “Give three options in a bulleted list.”
- “Do not include anything that requires a purchase.”
4. Examples
Showing is often clearer than describing. If you want a certain style or format, paste a short sample and say “match this tone” or “use this layout.” If you want to avoid something, say so. When you include an example, mention which parts matter, otherwise the tool may copy details you did not intend.
5. The role or point of view, when helpful
You can ask the tool to take a perspective, such as “act as a patient tutor” or “respond as an editor reviewing a draft.” This can shape tone and structure, though it does not give the tool real expertise or authority.
6. The format of the answer
Say how you want the answer laid out: a table with three columns, numbered steps, a short paragraph followed by a list of questions. Without guidance, you get whatever the tool favors.
Before and after
Before: “Help me with my résumé.”
After: “I am applying for an entry-level office administration job. Below is my current work history. Suggest improvements to the wording of each bullet point so that they focus on results rather than duties. Keep each bullet to one line. Do not invent any experience or numbers I have not given you. Work history: [paste].”
The second version names the goal, the audience, the format and a limit, and it explicitly asks the tool not to make things up. That last instruction does not guarantee compliance, but it helps.
Before: “Summarize this article.”
After: “Summarize the article below in five bullet points for a busy reader who wants the main arguments, not the background. Then list any claims that are stated without supporting evidence. Article: [paste].”
Break big tasks into steps
Large, complicated requests produce muddy results. Split them up:
- Ask for an outline.
- Revise the outline.
- Ask for one section at a time.
- Ask for a final review for consistency.
This also makes it easier to spot where something went wrong. For tasks involving reasoning, you can ask the tool to go through the problem step by step, which often makes mistakes easier to see, though the shown steps are not proof of correctness.
Iterate instead of starting over
The first answer is a draft. Respond to it as you would to a colleague:
- “Shorter, and drop the second paragraph.”
- “More formal, and aimed at parents.”
- “Give me three different openings.”
- “What did you assume about my situation?”
- “Which parts of this are you least sure about?”
Keep notes on prompts that work well for you. A saved prompt with blanks to fill in can save time on repeated tasks.
Prompts for images and other media
The same principles apply to image generators: describe the subject, setting, style, lighting and mood, and add details one at a time. The limits of image tools are covered in the article on AI image generators, such as trouble with hands, counting and lettering.
Things prompts cannot fix
- Missing knowledge. If the tool does not know something, a cleverer prompt will not supply it, and it may produce a convincing guess. Pasting in reliable source material helps more than rephrasing.
- Built-in errors. Chatbots can state wrong things confidently, for reasons explained in the discussion of chatbot errors.
- Hidden skews. Prompts do not remove bias in the underlying tool. You can ask for balance, but still review the result.
- Unfair shortcuts. Using a tool for schoolwork or work tasks may be restricted by rules you should follow.
Keep private details out of prompts
Detail makes prompts better, but not all detail is safe to share. Leave out passwords, financial account numbers, health records, and private information about other people or your employer. Replace real names with placeholders where possible. The article on assistant privacy lists what to avoid and which settings to check.
A quick template
I want to [task]. The audience is [who]. The purpose is [why]. Here is the background: [context]. Please keep it [length, tone, format]. Avoid [things to exclude]. Here is an example of the style I like: [sample]. If you are unsure about something, say so instead of guessing.
Fill in only the parts that matter for the job at hand. The best prompts are not the longest ones. They are the ones that remove the guesswork, and they are followed by a careful read of whatever comes back.