Guide · 3 min read
Turn Rough Notes Into a Reusable Claude Project
The biggest time sink when using AI is re-explaining context. Every new conversation, you're back to square one — who you are, what your company does, what tone you prefer. Claude Projects eliminate all of that.
A Project is a persistent workspace where Claude retains context across every conversation. Set it up once, and every session starts with Claude already knowing everything it needs about your work.
What a good Project contains
Most people create a Project and leave the instructions blank, or paste a one-liner. A well-built Project has four components, each doing a specific job. The first is a role definition — not "I work in marketing" but a precise description of your function and decision-making authority that shapes how Claude frames every response.
The second component — the context brief — is where most people leave the most value on the table. It's not a company description. It's a structured summary of the specific things Claude needs to know to be useful: your team's constraints, the current priorities, the stakeholders involved in the work, and the things that are true about your situation that Claude would have no way to know otherwise.
The third component is behavioral rules — the standing instructions that prevent Claude from defaulting to generic, sanitized output. And the fourth is at least one uploaded reference document that Claude can access in any conversation within the Project.
The template
Everything you already know about your own job is the raw material — the constraints you'd mention on a call, the history a new hire would need, the two people whose opinion decides things. Rough notes are fine. This is where they go.
Paste this into Project Instructions and fill it in once.
ROLE
I am [job title] at [company], a [what it does, who it sells to].
I decide: [the calls you actually get to make]
I don't control: [what's decided above or around you]
CONTEXT BRIEF
Priorities this quarter: [the two or three things that actually matter]
Constraints: [budget, headcount, timeline, tooling — the real limits]
Stakeholders: [who approves this work, and what each one cares about]
Things you'd never guess: [the history and politics that change the right
answer — e.g. "we tried this in 2024 and it failed because sales was
never consulted"]
BEHAVIORAL RULES
- Be direct. No preamble, no "Great question!"
- Prose by default. Bullets only when I ask.
- If I'm wrong, say so plainly and say why.
- If a task is ambiguous, ask one clarifying question instead of guessing.
- [your own standing rule]
OUTPUT FORMAT
Default: [e.g. "under 300 words, no headings"]
For documents: [e.g. "H2 sections, prose, no bullet lists"]
The line that does the most work is "Things you'd never guess." Role and rules change Claude's tone; that line changes its answers. It's the only part of the template Claude cannot infer from anything else you give it.
Then upload one document to the Project — a style guide, a product spec, a client list, last quarter's plan. Whatever you'd otherwise paste in more than twice. It's available in every conversation in the Project from then on, and you never paste it again.
For teams
Want us to build this for your whole team?
We run AI workshops and full implementations for teams of 5–200. Custom prompts, workflows, and training built around your actual work.