Prompt Engineering · The We Know AI method
How to write prompts that produce controllable results
A useful prompt does not need to be long. It needs enough context, constraints, format and quality criteria for the model to understand the task.
A practical prompt structure
Use the sequence context → task → source material → constraints → output format → checks. Include only the blocks your task needs. For ambiguous work, ask the model to clarify before producing the final result.
Improve the conversation, not just the first prompt
Treat the first answer as a draft. Point to a specific weakness, explain why it matters and request one focused revision. Keeping the useful context in the same conversation is usually more effective than repeatedly starting over.
Build verification into the request
Ask the model to separate facts from assumptions, cite the provided material and list claims that need external checking. Self-review helps, but it does not replace current documentation or expert review. See ready-to-adapt templates and discuss difficult prompts in AI Bandits.