Milan Bogojevic Blog

A One-Page AI Policy for a Small Nonprofit: What to Include and What to Leave Out

7 min read updated 10 October 2026

A single-page AI policy with four checkboxes beside a laptop and pen on an office desk.
A single-page AI policy with four checkboxes beside a laptop and pen on an office desk.

A small nonprofit’s AI policy should state four things staff can act on: which data must stay out of unapproved tools, which outputs need human approval, who approves a new tool and when the rules will be reviewed. Start with the work people already do. Keep the policy short, then link to the detailed procedures needed for sensitive or higher-risk tasks.

What happens when the policy is bigger than the practice?

Consider this hypothetical NGO reality. A board asks for an AI policy. Someone finds a template. The template has thirty clauses and covers model training, biometric data and automated decision-making with legal effect.

Your organisation has eleven staff and uses a chatbot to draft newsletter copy.

The policy is now longer than the practice it governs. Staff cannot tell which clause applies to their next draft, so they continue using the tool as before. The document exists, but the useful decisions remain informal.

The answer is not to ignore serious uses. It is to distinguish the ordinary workflow from uses that require a separate assessment. A newsletter assistant, a case-note summariser and a tool affecting access to services should not enter the organisation through the same approval shortcut.

Where should the first draft begin?

Ask the team what it already uses before writing rules. Invite specific answers: tool and account, task, input data, output, reviewer and storage location. Include browser assistants and tools embedded in existing software. Explain that the inventory is a starting point for safer work, rather than a request to hide experiments.

The 2026 Virtuous nonprofit AI adoption report found that 47% of surveyed organisations had no AI policy. Its survey covered 346 nonprofit professionals in December 2025. That is context from one sample, not a measurement of every nonprofit. Your own inventory tells you which gaps need attention locally.

Mark tools as approved for specified tasks, awaiting review or unsuitable for the proposed use. An approved newsletter use does not automatically approve uploading beneficiary records. Use the data levels in Gate 3 to make that distinction explicit.

Which four rules deserve space on the page?

Rule What the policy should say Practical example
Protected inputs Specify data prohibited in unapproved tools and the route for sensitive uses. No identifiable beneficiary notes, unpublished financials or funder-confidential documents in a public chatbot.
Human approval Name the output types that require checking and the responsible reviewer. A programme lead checks service facts before a newsletter is sent.
New tools Name the approver and backup; require a defined purpose and data check. The operations lead approves a bounded trial before a team starts a new subscription.
Review date Set a date and earlier triggers for reassessment. Review after three months, or sooner when a tool, task or permitted data changes.

Use job roles with current names where helpful. “Management” is too vague if nobody knows which manager can make the decision. A backup matters when the named person is absent.

What can a one-page policy actually look like?

Copy this template and replace the brackets. It is an operational starting point for routine drafting, not a complete policy for every legal or safeguarding context. Confirm it against your organisation’s existing obligations before adopting it.

AI use policy: [organisation]

Owner: [name and role]. Backup: [name and role]. Effective date: [date]. Next review: [date].

Purpose: Staff may use the tools in [approved-tools register] for [specified routine tasks]. Approval applies to the recorded task and data, not to every feature of a tool.

Inputs: Do not enter identifiable beneficiary information, confidential donor or partner material, unpublished financials, credentials or restricted documents into unapproved tools. Request assessment from [contact] before proposing a sensitive use. Use only the minimum approved information needed.

Outputs: AI-assisted material leaving the organisation must be checked by [responsible reviewer]. Verify factual claims, sources, figures, permissions and confidential details against authorised records. A tool cannot approve its own work.

New tools: Ask [approver] before adopting a tool or expanding its use. Record purpose, account, data, owner, review step and exit route in [register location].

Problems: Pause the affected workflow and report mistakes or possible exposure to [incident contact and channel]. Follow [incident procedure].

Review: The owner reviews this policy on [date], and earlier after a significant change or incident. Store the current version at [location] and communicate changes to staff.

Keep the completed template on one readable page. Put the approved-tools register and detailed procedures behind links, so the policy remains usable without pretending those records are unnecessary.

What should stay outside the short policy?

Do not fill the page with a glossary of model architecture, every possible future use or a long list of values without an operating rule. Staff need to know what they may do next, whom to ask and when to stop.

Leave detailed retention schedules, contract reviews and incident investigation instructions in their established procedures. Link to them. The policy should not replace safeguarding, privacy, financial controls or donor commitments. If an AI use affects those areas, involve the responsible specialist before approving it.

Avoid a blanket “AI is prohibited” rule when people already use it for permitted routine work. Equally, avoid “AI is allowed” without a task and data boundary. Both leave the next real decision unresolved.

How does a small team put the policy into use?

Give staff a short walkthrough using three fictional tasks: rewriting a public event description, drafting a donor update from approved aggregate figures, and summarising an identifiable case note. Ask which tool is approved, what can be entered and who checks the result. Resolve disagreements before calling the policy implemented.

Apply the Signature Test to external claims: the responsible person must be able to stand behind the content. Keep the original source material available so checking is possible. Reviewing tone alone will not catch an invented date or a changed budget commitment.

Add the policy to onboarding and tell staff where the current version lives. Have the backup approve one trial request during the rollout. That tests whether the process works when the owner is away.

My view

A page gets read when it answers a decision people actually face. I would judge an AI policy by whether two colleagues reach the same answer about a proposed task, not by the number of clauses it contains.

Shortness is useful only when the boundaries are clear. A small nonprofit may need substantial work behind the page, particularly for sensitive information. The short policy should point to that work and identify who owns it.

What to do next

Collect the tool inventory, complete the four rules and fill the template with names, locations and dates. Test it against the three fictional tasks with the people who will use it. Ask the appropriate privacy or safeguarding contact to review the boundaries for sensitive uses.

Then rehearse the AI incident response plan. Staff should know how to stop a workflow and report a mistake before one reaches a donor. Schedule the first review while the policy is fresh, and use that meeting to compare written rules with actual practice.

FAQ

Does every small nonprofit need a long AI policy?

The length should follow the uses and risks. Routine drafting can start with a short operational policy. Sensitive or consequential uses need the relevant specialist review and supporting procedures, even if the staff-facing summary remains one page.

Who should approve a new tool?

Name a person with authority to assess purpose, data and ownership, plus a backup. They should involve privacy, safeguarding, IT or other specialists when the proposed use requires it.

Can staff use a personal account for nonprofit work?

Only if the organisation has approved that account and use. A personal subscription does not establish acceptable data handling, organisational access or an exit route.

When should the policy be reviewed?

Set a date in the policy itself. Review sooner when the tool, permitted data or workflow changes, or an incident reveals a gap. Record and communicate the revised rules.

Sources

About the author

Milan Bogojevic works on practical AI adoption in nonprofits, with a focus on workflows, human accountability and organisational capability.

Last updated: 10 October 2026.

Next in AI Transformation