67. Troubleshooting AI Responses

The lesson

Problem: Output is too generic or vague. Cause: Instructions not specific enough. Fix: Add concrete requirements, examples, or constraints.

Problem: AI keeps making assumptions instead of using your data. Cause: Context not prominent enough. Fix: Put critical information in the Data section and explicitly say 'Use only the provided data'.

Problem: Wrong output format. Cause: Format section missing or ambiguous. Fix: Specify exact format with examples: 'Output as JSON with fields: name, age, score'.

Problem: AI includes information you know is wrong. Cause: Model may be hallucinating or using outdated training data. Fix: Use the Data section to ground responses in verified facts.

Problem: Response is too long or too short. Cause: Missing length constraints. Fix: Specify word count, bullet points count, or 'be concise' / 'be comprehensive'.

Problem: Code does not compile or has errors. Cause: Missing technical constraints or context. Fix: Include error messages, environment details, and explicitly ask for compile-tested code.

A network operations technician fixes a vague answer by adding exact requirements and constraints to the TPEE prompt, then runs the prompt again.

A waste services coordinator shortens a long response by setting word and bullet limits, and pastes the collection schedule into the Data section to stop the model guessing dates.

A registered nurse corrects a wrong output format by spelling out the exact fields the model must return for a shift-handover summary.

A site engineer stops invented contract details by placing the verified clauses in the Data section and instructing the model to use only that text.

An automotive mechanic fixes error-prone code answers by pasting the exact fault message and vehicle details, then asking for code that compiles.

A policy officer addresses unwanted assumptions by moving the key background into the Context section and telling the model to stay within it.

A retail store manager treats a too-generic product description by adding the audience, tone and a short example to the prompt.

A community pharmacist grounds a medication-leaflet draft by supplying the approved reference text and asking the model to rely only on it.

A robotics technician corrects a script that failed to run by adding the controller details and the exact error text, then asking for a corrected version.

An insurance underwriter shortens an over-long summary by setting a fixed number of bullet points and asking for concise wording.

Check yourself