What should a good brief actually describe?
Three things worth describing, and most people only describe one.
In "AI can't read your mind" we covered the basics: purpose, audience, what good looks like.
Anthropic's AI Fluency course splits a good description into three parts. Each one fixes a different kind of bad answer.
This is the part most people already do, at least a little. "Write a summary." "Make a chart."
The detail that usually goes missing is the shape:
In Anthropic's study of real conversations, fewer than 1 in 4 people said what tone they wanted.
This is the part almost nobody mentions, and it's often the one that matters most.
Without it, it takes the most common route, which might not be the one you'd take.
This is about the conversation itself, not the output.
Fewer than 1 in 3 conversations set out how they wanted to work with Claude like this. Without it, you get its default: helpful, agreeable and thorough, which isn't always what you need (see "Why AI agrees with you so much").
Product: A one-page plan for moving our team's files to the new drive. Plain English, no jargon, under 400 words.
Process: First list what could go wrong. Then plan around those risks. Put the steps in order of what has to happen first.
Performance: Point out anything in my notes that doesn't make sense. Ask before assuming.
Made-up example.
An example of what you want covers a lot of the product in one go: format, length and tone, all at once (see "AI copies what you show, not what you say").
People who do this are in good company. It was one of the more common habits Anthropic saw: about 4 in 10 conversations included an example of what good looks like.
The other 6 in 10 left the model to guess.
Before you send a request that matters, check you've said one thing about each:
It's three sentences, and it saves the five rounds of back-and-forth.