Shared Instructions and Standards
Apply description at team level: build a shared library of context and instructions, and define what 'good' looks like so AI-assisted work is consistent no matter who produces it.
Everyone Re-Explaining the Same Things
On most teams, each person who uses AI writes their own instructions from scratch, every time. They re-explain who the customers are, what tone the team uses, which products exist, and what format reports should follow, and each person explains it a little differently. The result:
- Inconsistent output. The same kind of document looks different depending on who made it.
- Wasted time. Everyone rebuilds context the team already knows.
- Hidden errors. One person’s outdated description of a product or policy quietly spreads into everything they produce.
Team description fixes this by turning the context and instructions that work into shared, maintained resources.
What Goes in a Team Description Library
| Resource | What it contains | Example |
|---|---|---|
| Team context brief | Who you serve, what you do, key terms, and what AI should never assume | ”We support small nonprofits. ‘Members’ means donor organizations, not individuals.” |
| Voice and style guide | Tone, reading level, words to use and avoid, formatting rules | ”Plain language, no jargon, sentence-case headings, no exclamation points.” |
| Task instructions | Tested product, process, and performance descriptions for recurring work | Instructions for weekly status reports, first-draft customer replies, meeting summaries |
| Examples | Approved samples of good work, with notes on why they’re good | Three strong past proposals, labeled for what to imitate |
| Boundaries | What AI should not do for this team | ”Never state pricing; use [PRICE]. Never promise delivery dates.” |
Many AI tools now let teams save shared instructions, projects, or custom assistants that carry this context automatically. Whether you use those features or a shared document, the principle is the same: write it once, test it, keep it current, and share it.
- 1 Draft
From what already works for someone
- 2 Test
On 3 to 5 real cases, by 2+ people
- 3 Approve
An owner signs off
- 4 Share
In one known place
Define What “Good” Looks Like
Shared instructions tell AI what to produce. Shared standards tell your people how to judge it. For each recurring type of AI-assisted work, write down a short definition of done:
- Accuracy: which facts must be verified, and against what source?
- Completeness: what must every version include?
- Fit: who is the audience, and what do they need?
- Voice: does it sound like the team, not like a generic AI?
- Sign-off: who has to approve it before it goes out?
Standards make review faster and fairer, because reviewers check against the same list instead of personal taste. They also give new team members a clear picture of quality from day one.
Common mistake: launching a shared prompt library with fanfare, then never updating it. Six months later it describes discontinued products and an old pricing model, and everyone who uses it spreads outdated information.
Give each resource an owner and a review date. An outdated shared instruction is worse than none, because people trust it.
- Pick the recurring task your team does most often with AI (use your AI use snapshot from Lesson 1.1).
- Collect the instructions two or three people currently use for it. Compare them: what’s the same, what’s different, what’s missing?
- Merge them into one tested instruction with product, process, and performance descriptions, plus a boundaries section.
- Write a five-point definition of done for that task.
- Have two people test the instruction on real work. Did they get consistent results? Revise.
- Assign an owner and a review date. Share it in one place everyone can find.
1. What's the main benefit of a shared team context brief?
Shared context improves consistency and saves time, as long as it's accurate and maintained.
2. Why does every shared instruction need an owner and a review date?
Shared resources multiply their effect, including errors. Maintenance keeps them trustworthy.
3. What does a 'definition of done' for AI-assisted work help with?
Clear standards make review faster and fairer, and show everyone what good looks like.
Up Next
Shared standards only matter if someone checks against them. The next lesson covers how teams review AI-assisted work, scaled to the stakes.