Is a bad AI answer often our fault for not knowing what we need?
How often is a disappointing AI result caused by the model, and how often by us not knowing clearly what we wanted? Not prompt tricks, but understanding the result well enough to give the right question, context, requirements and limits. Has anyone found the real problem was their own unclear ask?
Often, yes. A vague ask gets the most average answer, because the model fills every gap with the most common guess. When the result disappoints, the first question to ask is whether you could have described a good answer before you saw a bad one.
The simplest fix is to make the model ask first. Tell it not to answer yet, but to ask the questions it needs, one at a time, until it is confident about what you want. This brings up assumptions on both sides, including ones you did not know you were making.
For bigger tasks, write down two things before you start: why you need this, and what done looks like. If you cannot fill in either, the task is not ready for an AI or for a person.
Keep the questions and your answers. Next time you do the same kind of task, paste them in as context and you skip the interview.
Listings mentioned
- Interview me · skill by addyosmaniHas the model interview you one question at a time until it understands what you actually want.
- Requirements clarity · skill by davila7Clears up a fuzzy request with focused questions before any work starts, starting with why it is needed.
Answers by the AgentAlley team, drafted with AI and checked against the listings they link to. Not a real-person reply from the original thread.