AI-Assisted Research Charter

Project:
Researcher / team:
Date:
Discipline / field:

1. Substantive research problem

What problem matters independently of generative AI?

2. Existing researcher contribution

What substantive knowledge, theory, prior method, data access or research programme already exists before AI assistance begins?

3. Analytical object

What exactly is being studied?

Include unit of analysis, relevant population/corpus/source base, time period, main constructs or outcomes and proposed inferential target.

4. Non-negotiable constraints

What should not be silently changed by AI-assisted exploration?

Consider inclusion/exclusion rules, aggregation decisions, required caveats, disciplinary distinctions, ethical/confidentiality rules and other project-specific constraints.

5. Pre-use data and disclosure permission

Before any material enters a specific AI service, what do ethics approvals, participant consent, institutional policy, confidentiality, collaborator agreements, contracts and the service’s data-handling terms permit? Identify excluded data, any approved environment, and an alternative procedure when permission is absent or unclear. Pseudonymisation alone does not establish that disclosure is safe.

6. Permitted AI roles

7. Verification boundary

For each permitted role, what must happen before the output can influence the evidence base or final claim?

8. Consequential decisions to record

Which decisions would materially change the research trajectory and therefore deserve a provenance record?

9. Escalation rule

When must the project return to substantive/analytical judgement rather than accept an AI-generated solution?


Completed example

The PISA illustration is a retrospective completion based on the mature project record and predates the addition of the explicit pre-use permission prompt. Do not infer historical permission decisions from a blank field or reconstruct them as contemporaneous records.

See the completed PISA example.