When an AI assistant gives you something vague, most people assume the model is not good enough. Usually the model was capable and the question was not specific enough to aim it. A prompt is a brief, and a thin brief produces generic work from a person too.
You do not need a course in prompt engineering to fix this. Five habits cover most of the gap.
1. Say who it is for and what happens next
"Write about our new feature" gives the model no way to choose a register, a length, or a level of detail. "Write a short in-app announcement for existing customers who have never used the reporting section, ending with a link to the setup guide" gives it all three.
Audience and purpose do more work than any clever phrasing. State both.
2. Name the format you want
If you do not specify a shape, you get prose, because prose is the safe default. Ask for what you actually need: a table with named columns, five bullets of at most fifteen words, a numbered checklist, a single paragraph under eighty words.
A quick test: if you cannot tell in advance roughly what the answer will look like on screen, you have not described the format.
3. Give an example of good
Showing beats describing. Paste one example of the output you want and say "match this tone and structure." One good example communicates more than a paragraph of adjectives, and it removes the guesswork about how formal, how technical, or how long you meant.
4. Provide the raw material
An assistant that is not given your notes will invent plausible-sounding substitutes. If the answer depends on your data, your policy, or your document, paste it in. The difference between a prompt that references your actual material and one that does not is usually the difference between something you can use and something you have to rewrite.
This is also the single most effective way to reduce made-up details: give the model the facts rather than hoping it recalls them.
5. Treat the first answer as a draft
The most underused feature of every assistant is the follow-up message. You do not have to rewrite the prompt and start again. Say what is wrong: "too formal," "cut it in half," "you invented the second statistic, remove it," "keep the structure but rewrite for someone non-technical."
Two or three rounds of correction reliably beat one very long prompt, and they take less time to write.
A reusable skeleton
When you are stuck, fill in these five slots and you will have a workable prompt:
- Role: who the assistant should act as
- Task: the single thing you want done
- Context: the material, constraints, and audience
- Format: the shape of the output, with a length
- Check: what to do when something is missing, such as asking rather than guessing
That last slot matters more than it looks. Adding "ask me for anything you need rather than assuming" turns a confident wrong answer into a useful question surprisingly often.
The AI Guide prompt library has ready-made prompts built on this structure for each tool it covers, so you can start from a working example instead of a blank box.