Template · Giving good context
AI context brief template
If you use AI on the same project every week, you explain the same background every week. Each time, you leave out something different. A context brief fixes that. You write the background once, keep it current, and paste it at the start of each new conversation.
Keep the brief to one page. It holds what is true for the whole project. The details of a single task belong in the request that follows it.
When to use it
- You return to the same project or account in AI conversations more than once a week.
- Answers drift because the model does not know your terms or your fixed constraints.
- A colleague is taking over AI-assisted work on the project and needs the same starting point.
The template
Project and goal
What is the project, and what does finished look like?
Current status
Where does the work stand today?
Key terms
Which words does your team use in a specific way?
Fixed constraints
What cannot change, whatever the request?
Sources in play
Which documents or data does the work draw on?
Deliberately left out
What have you excluded, and why?
How the work is split
What are the steps, and where do you check the output?
A filled example
A data analyst writes a standing brief for the quarterly customer retention review, a project worked on with AI across several weeks.
- Project and goal
- Quarterly retention review for the subscription product. Finished means a ten-slide summary for the leadership meeting on March 28.
- Current status
- Data pulled through December 31. Cohort tables done. Churn reasons not yet coded.
- Key terms
- “Churned” means no renewal within 30 days of term end. “Account” means a paying organization, not a user.
- Fixed constraints
- Leadership wants cohort figures by signup quarter only. All figures come from the warehouse tables, never from the sales dashboard.
- Sources in play
- Retention tables export, attached. Last quarter’s summary deck, attached. Survey free-text responses, to be attached when coding starts.
- Deliberately left out
- Customer names and contract values are removed. Last year’s pricing study is out of scope and should not be cited.
- How the work is split
- Step 1: code churn reasons. I check a sample of twenty. Step 2: summarize by cohort. Step 3: draft slides from the checked summary.
What each field is for
Project and goal
Skill: Giving good context
A stated goal lets the model judge what is relevant. Without it, every detail in the material looks equally important.
Current status
Skill: Giving good context
Status stops the model from redoing finished work or assuming steps that have not happened. Update this field each time you use the brief.
Key terms
Skill: Giving good context
Teams use ordinary words in specific ways. A model reads them in their ordinary sense unless the brief defines them.
Fixed constraints
Skill: Giving good context
Constraints stated once in the brief apply to every request. That matters in long conversations, where early instructions get crowded out.
Sources in play
Skill: Giving good context
Listing sources tells the model what it can cite and tells you what to attach. Anything not on the list should not appear as a source in the output.
Deliberately left out
Skill: Keeping it safe
Exclusions are part of context. Writing them down keeps sensitive material out of the conversation and keeps old analysis from creeping back in.
How the work is split
Skill: Putting it together
A brief that names the steps keeps each conversation inside one step. The checkpoint stops an early error from reaching the slides.
Common mistakes
Letting the brief go stale
A brief with last month’s status gives the model a false picture. Update the status field before you paste the brief, or delete it.
Turning the brief into an archive
A five-page brief buries the facts that matter. If a detail applies to one task, put it in that task’s request instead.
Pasting it into an unapproved tool
A good brief describes your project in detail. Check that the tool is approved for that material before you paste it anywhere.