A generative assistant produces a proposal, not proof. To get useful output, define the work expected and how you will verify it.
1. Define the task and its context
“Help me with my project” leaves too many decisions implicit. Specify the audience, goal, available information and constraints. Provide only necessary data and avoid confidential information in an unauthorised service.
2. Request verifiable output
Choose a format and criteria: a decision list, a table with assumptions or a draft of a defined length. Ask the assistant to distinguish supplied information, assumptions and missing information.
3. Check before using
Fluent text can contain errors. Check figures, claims and references against reliable sources. For code, read the changes and test a concrete behaviour; a convincing response does not replace execution.
TASK: Prepare a short project brief
AUDIENCE: A non-technical team
INPUT: An anonymised description
OUTPUT: Goal / deliverables / risks / open questions
CHECK: Separate facts from assumptionsYour turn to build
Write two requests for the same need: a vague request, then one with context, format and verification criteria. Identify what you still need to validate yourself.
Your discovery progress is saved on this device. You can clear it from My learning.