Guide1 min readUpdated June 29, 2026

How to write a better AI prompt

Short answer

Give the model the context it needs, state the task specifically, and describe the output you want. Vague prompts produce vague answers; a prompt that names the goal, constraints, and format gets far more useful results.

A structure that works

Give contextTell the model who it is helping and any background it needs, such as the audience or the source material.
State the task clearlySay exactly what you want done. 'Summarize' is vague; 'summarize in three bullet points for a beginner' is specific.
Define the outputDescribe the format, length, and tone: a table, JSON, 100 words, or a formal email.
Add constraints and examplesList what to avoid and, when it helps, show one short example of the result you want.

Specific beats longA focused prompt usually beats a long, rambling one. Extra words that do not add constraints or context mostly add noise.

Try it: Prompt OptimizerTighten a prompt's structure and clarity locally before you send it.Open tool

Common mistakes

Asking for several unrelated things in one prompt
Leaving the output format to chance
Assuming the model knows context you never provided
Not saying what to do when the model is unsure

Longer prompts also cost more tokens. If you are watching length, see What is a token in AI language models?.

References

Prompt engineering on Wikipedia — Wikipedia

Questions

Should I tell the model what role to play?

A short role or audience cue can help focus the answer, for example 'explain to a beginner'. Keep it brief and relevant rather than elaborate.

Do examples really help?

Often yes. One short example of the desired output format gives the model a concrete target and reduces guesswork, especially for structured results.

Does the prompt optimizer send my text to a model?

No. It restructures and cleans your prompt locally in the browser and does not call any model or server.