Questions to help organisations work out how AI should be used, who remains responsible, and how they will learn as practice changes.
Whether your organisation calls it a policy, guidance, principles or an acceptable use framework, the underlying questions are similar. How should AI be used? Where is greater care needed? Who remains responsible? How will the organisation learn as practice evolves?
I have developed these questions as a starting point for smaller organisations without dedicated legal, risk or IT functions, including small NGOs and community organisations, research groups, consultancies and other professional teams working through existing governance, ethics and integrity processes.
AI use often begins before an organisation develops any formal guidance. Someone experiments with a tool, others follow, and gradually new ways of working emerge.
A policy can help make expectations visible by clarifying acceptable use, responsibilities and safeguards. For a small organisation, this need not be a large piece of work. Useful guidance may be a single page or a few paragraphs added to an existing staff handbook or research protocol. Much of the value lies in the conversation that produces it, including how the organisation understands current use, recognises where greater scrutiny is needed and learns from experience.
The questions below are offered as prompts rather than a template.
1. Why are we developing guidance?
- What do we hope it will achieve?
- What concerns or opportunities are prompting it?
- How will we know whether it is helping?
2. What is already happening?
- Where is AI already being used?
- What tasks does it support?
- What information is being entered, and by whom?
- Which existing policies already apply?
3. What should the guidance cover?
Think about questions such as:
- acceptable and restricted uses
- privacy and confidentiality
- checking and human review
- disclosure
- authorship and intellectual property
- responsibility and accountability
- uses with more serious consequences
4. Which uses deserve greater scrutiny?
Not every use carries the same consequences. Routine drafting, restructuring non-sensitive text and personal reflection usually need different oversight from uses that shape research findings, consultation, services or decisions affecting other people.
The useful distinction is rarely the tool. It is whether the consequences of getting it wrong fall mainly on the person using it, or on someone else.
5. Whose material is involved, and who is not in the room?
Much of the material organisations enter into these systems originated with other people: community submissions, interview transcripts, member records, partner correspondence, and the work of staff and volunteers.
Those people may not be party to the decision about which tool is used, or on what terms.
- Whose material would be entered, and did they consent to this use of it?
- What did we tell people when we collected it?
- Would we be comfortable explaining our practice to them?
- Where does data go, and who else can see or retain it?
Where Indigenous peoples, communities or customary knowledge are involved, policy development may need to address Indigenous data sovereignty and governance, collective rights, consent and existing ethical obligations. In Aotearoa New Zealand, this includes Māori data sovereignty and responsibilities that may already sit within research ethics and organisational arrangements. Many general policy templates give limited attention to these matters, if they address them at all.
6. Who should be involved?
Guidance works better when the people doing the work help shape it. They know where AI is already in use, where they are uncertain, and where disclosure might feel risky.
In a small organisation, this may simply mean a conversation across the whole team. Where particular expertise is missing, such as privacy, information security, cultural or ethical review, it is worth being explicit about that gap and about where the organisation would seek advice rather than assuming the question has been covered.
7. How will the guidance stay useful?
AI tools and organisational practice will continue to change. Consider:
- how the guidance will be reviewed
- how difficult cases will be discussed
- how experience will be shared
- how decisions will be made about where AI should continue, change or stop
Looking beyond policy
Guidance provides a shared starting point. What happens after that is a matter of everyday judgement: whether someone checks a summary against the original submissions, whether a group notices that a consultation report reads smoothly but no longer sounds like the people who wrote to them, and whether anyone asks what became better, for whom, and what the organisation has learned.
Related reflections explore these wider questions:
- Working with AI: where to begin
- AI can produce the record of participation without the participation
- How do we know whether AI is improving the work?
- What would better work with AI look like?
There is no single model AI policy that suits every organisation. The questions above are intended to help organisations develop guidance that reflects their own work, people and responsibilities.
For examples, frameworks and adaptable templates, see AI policy templates and guidance. For wider social, ethical and organisational perspectives, see AI in context: the wider picture. The AI for professional practice, research and collaboration hub provides the fuller collection of related essays, guidance and curated resources.
Support with developing AI guidance
I offer short, collaborative support to small organisations, NGOs, research groups and professional teams developing or reviewing AI policies and internal guidance. This may include:
- clarifying the purpose and scope of the guidance
- facilitating discussion about current use, uncertainty and responsibility
- reviewing and adapting existing policies or templates
- developing practical decision prompts and review arrangements
- linking policy development with organisational learning
Get in touch to discuss what would be useful.
[* Image by Jintana / Adobe Stock]