Write down what AI is for in your organization.
Ask five people on your team what AI is for here and you will probably get five answers. One thinks it is for drafting emails. Another thinks it is banned for anything involving clients. A third heard in a hallway that leadership wants everyone using it, and is quietly worried about what that means for their job.
The hallway version is already running.
Every organization has a working answer to "what is AI for," even if nobody has written it down. It lives in a passing comment the director made at a staff meeting and in whatever the office manager saw a colleague do last week. People act on that version every day, because it is the only one they have.
The trouble with the hallway version is that it drifts. It changes depending on who you ask and how recently they talked to someone in charge. A new hire at a clinic gets a different story from the front desk than from the billing lead, and both are guessing. Nobody can be held to an expectation that was never stated, and nobody can be told they are doing well against one either. A written page gives everyone the same answer, which you need before you can test whether that answer is any good.
Most AI statements are too vague to use.
When organizations do write something, it often reads like a values poster. "We will use AI responsibly while keeping people at the center of everything we do." Nobody disagrees with that sentence, and nobody can tell from it whether they should run a grant application through an assistant tomorrow morning.
A useful statement is specific enough that a manager could look at what someone did last week and say whether it matched. That is a higher bar than it sounds. It means naming actual work: the monthly board report, the intake form, the first draft of a proposal, the reply to a parent's email. It also means naming the things AI stays out of, and saying what you expect people to do, in terms they could check on a Friday afternoon.
If a manager can't hold last week's work up against the page, the page won't change anything.
What belongs on the page.
Keep it to one page. A longer document becomes a policy manual, and people stop reading at the second heading. Your data rules and approved tools belong in the policy; this page is about purpose and expectations, and it should point to the policy rather than repeat it.
Start with why you are doing this at all, in one or two sentences tied to your own work. A town office might say it wants permit questions answered faster without adding staff. A small manufacturer might say it wants quotes out the same day. If a line further down doesn't connect back to that reason, cut it.
Then list where AI helps. Be concrete and use your own nouns: "first drafts of client letters," "summarizing inspection notes," "turning meeting notes into action items." Next, list where it doesn't belong yet, and say why, because a reason lets people reason about the cases you didn't list. A nonprofit might keep AI out of final donor communications until someone has reviewed a few months of drafts. A professional firm might keep it away from anything filed with a court.
The last part is what you expect from people, and it is the part most statements skip. That might be time set aside each week to try something, a habit of sharing what worked at the team meeting, or an expectation that managers ask about AI in one-on-ones. Whatever you choose, it should be something you could notice happening or not happening.
Keep it alive by using it.
A statement that sits in a shared drive goes stale quickly. Tools change, and so does your sense of what the team can handle. What felt ambitious in spring may be routine by fall, and what you ruled out may turn out to be fine once you have seen it done carefully.
Put a review date on the page, and treat it as a real meeting. Quarterly works for most teams. At the review, ask whether the "helps" list still matches what people actually do, whether anything in the "stays out" list has earned a trial, and whether the expectations were met. Bring a few real examples from the last few months, good and bad, rather than impressions.
Between reviews, the statement should show up in ordinary places. Mention it when you onboard someone. Refer to it when a manager asks whether a task is fair game. In our work with teams, the statements that change behavior are the ones leaders quote back in staff meetings, while the ones that only get read at launch tend to fade.
A one-page template.
Fill in the blanks, cut anything that doesn't apply, and share the draft with three or four people before it goes to everyone. Their questions will tell you which lines are still vague.
What AI is for at [organization name]
- We use AI because: [one or two sentences tied to your own work, for example "we want every customer quote out the same day"].
- AI helps with: [three to six specific tasks, named the way your team names them].
- AI stays out of, for now: [specific tasks or decisions], because [the reason, in one line each].
- A person always reviews: [outputs that go to clients, the public, regulators or payroll before they leave the building].
- We expect everyone to: [one or two habits you can observe, such as "try one new task a week and bring it to the Monday meeting"].
- We expect managers to: [what they do differently, such as "ask about AI in every one-on-one this quarter"].
- Questions go to: [a named person], and the rules on data and tools are in: [link to your policy].
- We review this page on: [date], and the next version will be shared at: [meeting or channel].
