When an AI answer is not useful
The most useful “trick” is finding the cause: an unclear question, missing material or a capability the system does not have.
1. Break the task down
Large tasks often benefit from smaller, verifiable questions.
Divide my goal into three useful subtasks. Complete only the first one initially, and state what result is needed for the next step.
2. Ask what information is missing
Repeating the same vague question with different wording rarely resolves the underlying gap.
What are the three most important pieces of information you need to answer reliably? Ask specific questions before making assumptions.
3. Supply the material
If the AI cannot access a file or page, provide the relevant excerpt that you are permitted to share.
Answer using the following excerpt. Separate statements in the text from your interpretation. If the text does not contain the answer, say so explicitly.
4. Show what needs to change
“Make it better” is vague. Name the problem and show the desired structure.
The previous answer is too general. Replace general advice with five concrete steps. Each step must include an action and a verifiable result.
5. Make uncertainty useful
A partial answer and a research plan can be more useful than unsupported certainty.
What can be established from the available information? What remains uncertain? Which primary sources or measurements would resolve the gaps?
6. Clarify a legitimate goal after a refusal
If a harmless question was misunderstood, provide honest context. You can ask for general principles, prevention or a safe alternative.
My goal is: [state your actual goal]. Which parts can you answer? Explain the general principles and suggest a safe approach.
Invented permissions, deception or roleplay to bypass safeguards are not reliable working methods. Rewording a request does not remove a legitimate boundary.
7. Match the tool to the task
A longer prompt does not supply internet access, file access or a computation environment. Provide suitable material or use an available tool that supports the task. Check substantive results.
Improve an unsuitable answer with a focused follow-up
A dedicated page offers twelve repair templates for issues such as missing requirements, altered facts and incomplete code. Each card explains the inputs, success criteria and limits. Choose a template, add your original task and the previous answer, then create a follow-up prompt instantly.
These original templates have not yet been comparatively tested for answer quality. They do not perform an automatic content check.
Open the repair library