Most teams buy AI seats and stop there. Everyone writes their own prompts, nobody knows which sources are allowed, and good results are hard to repeat. This playbook sets up the shared layer — projects, instructions, approved sources and rules — around the work the team already does.
Step 1 — Inventory the real work
Goal: Find the tasks worth standardizing before touching any settings.
- Ask each person for the three tasks they already use AI for, or wish they could.
- Group them into recurring tasks with a clear input and output.
- Pick three that happen weekly and matter to a customer, a sale or a decision.
- Note which information each task needs and where it lives.
You finish with: A shortlist of three tasks, each with its inputs and owner.
Step 2 — Decide the rules
Goal: Agree what may go into the assistant and what must come out reviewed.
- List the data the team may use with AI, and the data it may not.
- Decide which outputs need a human review before they leave the company.
- Choose one workspace (your ChatGPT or Claude business plan) as the standard.
- Write the rules on one page and share them with the team.
You finish with: A one-page usage policy the whole team has read.
Step 3 — Build the shared setup
Goal: Give each shortlisted task a reusable home.
- Create a shared project or workspace per task, with written instructions.
- Add the approved reference files: templates, style guides, product facts.
- Connect only the sources each task needs, with the narrowest access that works.
- Test each setup on three real examples and fix the instructions until results are consistent.
You finish with: Three shared, tested setups the team can reuse.
Step 4 — Teach, measure, expand
Goal: Make the setup the default way of working.
- Run a short, task-based walkthrough for each setup — no generic AI training.
- Measure time per task before and after for two weeks.
- Collect the prompts and fixes people add, and fold the good ones into the instructions.
- Add the next three tasks only once the first three are in regular use.
You finish with: Measured adoption and a list of what to standardize next.
Common pitfalls
- Buying more seats before anyone has a shared setup to use.
- Generic AI training instead of walkthroughs on the team’s own tasks.
- Connecting every data source on day one.
- No owner for the instructions, so they go stale.