Our AI policy
Thirteen decisions, answered in one meeting, out loud. Every question comes with answers to pick from, so nobody has to guess what it means. Tap what is true for you, write your own where the options miss, then print and sign. The result is the policy.
Answers save in this browser as you go.
Organization
Date adopted
Who may use AI for our work?
Pick oneDecide who this policy covers. Anyone not on the list does not use AI for organization work, on any account.
Which accounts do they use?
Pick oneA personal account can train on what gets typed, and you cannot switch it off when somebody leaves. An organization seat does neither.
Which data never goes into any AI tool?
All that applyVague rules fail. Check every category your organization holds. Checked means banned in every tool, on every plan, no exceptions.
The rule that makes this workable: describe the situation, never the person. “A tenant facing eviction with arrears and two children” is fine. The name, address and case number are not. Same rule for spreadsheets: strip the name and contact columns before any analysis. Patterns never need the person.
What must a human read before it leaves the building?
All that applyNo AI tool has an approval queue or a reviewer role, so review exists only if this page names a person. Check each output type, then write who reviews it, by role.
The real exposure is copy and paste, not a share button. Somebody pasting a drafted grant narrative into an email to a funder touches no setting anywhere.
Which tools are approved today?
All that applyA short named list beats a careful sentence. If a tool is not listed here, it is not approved.
Who approves a new tool or integration?
Pick oneOne approver. Nobody connects anything new to our email, files or donor database without them.
Where does the saved time go?
Pick oneAI frees real hours. Decide where they go before they dissolve into more email. This is the question every policy skips, and the one that decides whether staff feel helped or squeezed.
Set expectations per role, out loud. The hours are only real if the calendar shows where they went, and nobody’s workload silently doubles because AI made them faster.
Do we tell funders we use AI?
Pick oneFunders are split and most have not published a policy. The National Institutes of Health bans AI-written applications outright.
Do we tell donors?
Pick oneDonors mostly want honesty. A plain statement usually settles it.
Do we tell the people we serve?
Pick oneThis one can be binding rather than optional. If AI touches legal, health, housing, employment or financial advice, the person must know at the start of the session, and a qualified professional must review the output before it reaches them.
The rule that settles all three disclosure questions: if knowing AI was involved would change how the reader sees the work, they need to know. That is the threshold.
Where does AI never go near a decision?
All that applyThese carry real liability. Anything you leave unchecked you have decided to allow, so decide out loud.
When something goes wrong, what happens?
All that applyTwo incidents are predictable, so the response gets written now, not during the incident.
Who owns this page, and when do we review it?
Pick oneA policy with no owner and no review date is decoration.
The owner, by name:
Approved by
Signature and date
Template prepared by Applied AI Advisor © 2026 · 3a4ai.com