Briefing AI like a new colleague

Context in, quality out

After this, you'll be able to write a conversation opener that gives an AI model enough context to produce output specific to your role, audience, and goal.

Before you start

Complete Ground Floor first; this lesson builds on understanding that AI generates by prediction, not lookup.

The idea

A new colleague doesn't know your industry, your clients, or what 'good' looks like on your team. You brief them once at the start, then correct course as the work unfolds. AI works the same way. The brief gets you started. What makes you fluent is knowing how to steer after the first response.

Here is the before and after: 'Summarize this article' returns a 5-sentence generic summary that could have been written for anyone. 'I'm a product manager preparing a weekly update for an engineering team. Summarize this article in 4 bullet points: what happened, why it matters to us, what we should watch, and one question to raise in standup' returns something you can paste directly into your update.

Now try it: write a chat opener for a task you do weekly using those four fields. Then, after you get the first response, practice one specific correction: name exactly what is wrong and ask for just that one change. Mid-conversation steering is the skill this lesson adds to the brief you learned in L0.

Try it (8 min)

Watch out for

Describing your audience as 'my team' instead of naming their role and what they care about
Writing a task that is really two tasks; split it and brief each one separately
Skipping the 'done looks like' field; the AI will pick its own definition of success
Pasting raw notes without a one-sentence framing of what they are and why they matter

Paste this into Claude

I am a [job title or role]. I need to [specific task, one sentence] for [audience: describe who they are and what they specifically care about]. The output should be [format: bullet points, paragraphs, or table; specify length and structure exactly]. The output is successful when [one thing you can verify is true without reading the whole output]. Here is the context: [paste your actual content, document, or situation]. Additional constraints: [any specific thing to include or avoid, e.g., no jargon, keep under 200 words, include a concrete next step].

What good looks like

The output is specific to the audience you named, not a generic response
You can verify the success criterion you wrote without reading the full output
The format matches exactly what you specified in the prompt
After one follow-up that names a specific change, the output improves without you re-explaining the full context

When this breaks

Breaks when the audience field is missing because the AI writes for a generic reader, not the person who will actually use the output
Breaks when the task is compound because the AI trades off between sub-tasks instead of completing both fully
Breaks when there is no success criterion because the AI has no way to know when to stop adding content

You can now

Read the output and confirm it is specific to the audience you named; a stranger reading it could tell exactly who it was written for.

Key takeaways

AI has no context about your work until you provide it. A four-part brief turns generic output into a usable one. Knowing how to correct and redirect without starting over is what makes you fluent.

AI has broad knowledge but zero context about your role or audience until you give it directly
Four fields make a brief: who you are, what you need, who it's for, and what done looks like
The audience field is the one most people skip and the one that changes the output most
When the output drifts, name the specific thing that is wrong and ask for one targeted correction instead of starting over

Was this helpful?

← Back to lessons

I am a [job title or role]. I need to [specific task, one sentence] for [audience: describe who they are and what they specifically care about]. The output should be [format: bullet points, paragraphs, or table; specify length and structure exactly]. The output is successful when [one thing you can verify is true without reading the whole output]. Here is the context: [paste your actual content, document, or situation]. Additional constraints: [any specific thing to include or avoid, e.g., no jargon, keep under 200 words, include a concrete next step].