# Give an AI workflow the inputs it actually needs A strong input brief gives the AI enough context to do the task without asking it to infer your facts or preferences. State the goal, audience, confirmed information, constraints, and expected output. Leave out private details that do not help the work. ## Write the outcome as an artifact “Help me with my career” is broad. “Create a requirement-to-evidence table for this job posting and my confirmed experience” names an inspectable output. It also makes it easier to tell whether the model did the work you asked for. ## A fictional brief | Field | Example | | --- | --- | | Goal | Prepare an onboarding-role application | | Audience | Hiring team | | Facts | Scheduled twelve customer handoffs | | Constraint | No retention measurements available | | Output | Evidence table and three revised bullets | | Review | Every addition tied to supplied facts | The brief is not a place to add guessed information. A missing outcome remains missing, even when the output would look more persuasive with one. ## Copy the structure ```text Goal: [what I am trying to do] Audience: [who will use the result] Confirmed facts: [information I can support] Unknowns: [information still missing] Constraints: [format, privacy, tools, timing] Expected output: [specific artifact] Review criteria: [what I will check] Ask before assuming facts that could change the result. ``` ## Keep the brief current during the task When you answer a follow-up question, update the fact list rather than relying on a long conversation to preserve every detail. If a requirement changes, say which earlier instruction it replaces. Contradictory directions can lead to a plausible but wrong compromise. After receiving the result, compare it with the output and review criteria. If it missed the purpose, revise the brief or the workflow before polishing the prose. A well-structured input does not guarantee correctness, but it gives both you and the model a clearer basis for asking questions and checking the work. --- SkillStall · 2026-10-02