Most bad AI output is a briefing problem. The model did exactly what it was asked; the ask was just vague. Description is the skill of telling AI what you actually want, with enough context that it can get there. It is the part of fluency that improves fastest once people take it seriously.
What this means for your team
Treat a prompt like a brief you would give a sharp new starter. The same things that make a good brief make a good prompt:
- Context. What it is for, who it is for, what has come before. The model knows nothing about your situation unless you tell it.
- Specifics. What good looks like, what to avoid, the format you want back. 'Write something about onboarding' and 'draft a 200-word welcome email for a new warehouse hire, warm but not gushing' give very different results.
- Examples. One sample of the style or structure you want is worth a paragraph of description.
- Role and tone. Tell it who to be and how to sound. A blunt reviewer and a supportive coach give different feedback on the same work.
Most teams get an immediate lift just from briefing properly rather than firing off one-liners.
Try this
Take your team's most common AI task and write a proper brief for it together: context, specifics, one example, the tone you want. Save it as a shared template so nobody reinvents it each time.
Common questions about briefing AI
How do you brief AI clearly to get useful work back?
Treat a prompt like a brief you would give a sharp new starter. The same things that make a good brief make a good prompt:
- Context: what the work is for, who it is for, and what has come before, because the model knows nothing about your situation unless you tell it.
- Specifics: what good looks like, what to avoid, and the format you want back.
- Examples: one sample of the style or structure you want is worth a paragraph of description.
- Role and tone: tell it who to be and how to sound, since a blunt reviewer and a supportive coach give very different feedback on the same work.
Most teams get an immediate lift just from briefing properly rather than firing off one-liners.
Why does AI give vague or unhelpful answers?
Most bad AI output is a briefing problem. The model usually did exactly what it was asked. The ask was just vague. AI cannot read your mind, so 'write something about onboarding' and 'draft a 200-word welcome email for a new warehouse hire, warm but not gushing' give very different results. Description, telling the AI what you actually want with enough context to get there, is the part of fluency that improves fastest once people take it seriously.
What context should you give AI when you ask it to do something?
Give the AI the context it could not possibly know: what the work is for, who the audience is, what has come before, and any specific data or constraints it should draw on. The model already has broad training, so your job is to add what is specific to your situation. You can also guide how it works through the task, not just the end goal, much as you might want a colleague to use your method rather than their own.
What is the difference between describing what you want and how the AI should work?
The lesson splits description into three parts. Product description is defining clearly what you want the AI to create or provide, the end result. Process description is guiding how the AI approaches the request, for example a particular method, order, or technique. Performance description is setting how the AI should behave, such as whether to challenge your assumptions or follow your lead, and whether to be detailed or concise. The point underneath all three is that AI tools are not databases or vending machines. They are interactive systems that behave differently in different contexts, much like people, so it pays to say how you want them to behave.