What Should Students and Supervisors Agree on Before AI Becomes Part of Thesis Research?

What Should a Student and Supervisor Decide Before AI Is Used in Thesis Research?

A graduate student plans to use generative AI during thesis preparation. The proposed activities include literature exploration, statistical programming, language editing, and preliminary interpretation of findings.

The supervisor generally supports AI-assisted research but expects the student to demonstrate independent methodological competence and comply with institutional research requirements.

Both agree that AI may be useful. Yet neither has clarified which tasks are permitted, when prior approval is necessary, what information may be uploaded, or how substantive assistance should be documented.

Months later, the student submits an AI-assisted analysis that the supervisor considers insufficiently justified. The disagreement is not simply about the technology. It concerns expectations that were never established.

The practical question is what students and supervisors should agree on before AI becomes part of thesis research so that legitimate assistance supports scholarly development without obscuring intellectual responsibility or creating avoidable compliance problems.

A useful agreement should translate applicable research requirements into clear working expectations while recognizing that a thesis project may change substantially between proposal development and final submission.

What Should a Thesis AI-Use Agreement Cover?

Establishing a Practical AI Agreement for Thesis Research

Why an AI Agreement Is Different From a General University Policy

University policies may establish broad rules concerning academic integrity, research ethics, information security, and acceptable AI use.

A student-supervisor agreement addresses how those requirements will be applied within a particular thesis project.

For example, an institutional policy may permit AI-assisted writing under specified conditions, while the thesis agreement clarifies which writing activities the student may undertake independently and which require supervisory discussion.

The agreement should supplement rather than contradict binding institutional requirements.

Supervisors should distinguish their own expectations from formal rules and explain where students can obtain authoritative clarification.

1. Identify the Policies That Govern the Thesis

Before discussing individual AI tools, students and supervisors should identify the requirements that already apply.

These may include graduate school regulations, university AI policies, research ethics approvals, participant consent arrangements, confidentiality agreements, funder conditions, and disciplinary requirements.

Publication policies may become relevant when the thesis is prepared for journal submission.

The European Code of Conduct for Research Integrity identifies reliability, honesty, respect, and accountability as fundamental principles.

These provide a useful foundation, but they should not be represented as a universal set of detailed AI permissions.

The agreement should identify which requirements are binding and which expectations are specific to the supervisory relationship.

2. Define the Scope of AI Assistance

Students and supervisors should identify which research activities may involve AI.

These may include literature exploration, research design, instrument development, data cleaning, statistical programming, qualitative analysis, manuscript preparation, and presentation development.

Not every activity needs an identical rule.

For example, AI may be permitted to explain methodological terminology while substantive changes to the research design require consultation.

Defining the scope helps prevent a general statement such as "AI assistance is allowed" from being interpreted as permission for unrestricted intellectual delegation.

3. Establish Permitted, Approval-Required, and Prohibited Activities

A practical agreement should distinguish activities covered by standing permission from those requiring prior discussion or formal authorization.

It should also identify prohibited conduct.

The categories are illustrative and should be adapted to the actual project and institution.

The more detailed question of which AI activities supervisors should permit or prohibit should be resolved through task-specific requirements rather than product names alone.

4. Specify Which Intellectual Responsibilities Remain With the Student

AI may assist with research tasks without assuming responsibility for the thesis.

Students should remain able to explain the research question, justify the methodology, evaluate evidence, and defend consequential interpretations.

For example, a student may use AI to generate preliminary code while remaining responsible for understanding the analysis and ensuring that the implementation is verified.

However, a student who allows AI to determine the research design without independent evaluation may not demonstrate the methodological competence required by the degree program.

The agreement should identify the research competencies that must remain independently demonstrable.

These expectations should reflect the discipline and degree level.

5. Clarify the Difference Between Assistance and Intellectual Outsourcing

Students and supervisors should discuss what constitutes legitimate support and what would improperly replace the student's intellectual work.

For example, using AI to improve the wording of a student-developed argument may be different from accepting an AI-generated argument without understanding its evidential basis.

Similarly, generating statistical code may be legitimate when the student understands and verifies the analysis.

The distinction depends on the educational objective and the nature of the research contribution.

The agreement should explain how legitimate AI assistance differs from intellectual outsourcing without assuming that every generated sentence or computational operation is inappropriate.

6. Establish Verification Responsibilities

Students should know which AI outputs require independent checking and how that checking should occur.

For literature reviews, verification may involve examining original publications and confirming bibliographic details.

For analytical code, it may involve inspecting transformations, testing the implementation, and examining outputs.

For methodological recommendations, it may require consulting relevant disciplinary literature or qualified expertise.

For qualitative interpretation, it may involve returning to the original material and examining methodological coherence.

The agreement should identify who is responsible for verification and when supervisory or specialist review is necessary.

Permission to use AI does not establish that its outputs are scientifically correct.

7. Establish Research Data and Confidentiality Rules

Thesis research may involve participant interviews, institutional records, unpublished manuscripts, or restricted datasets.

Students and supervisors should identify which information may be processed through particular AI systems.

Authorization should be based on the actual processing environment and applicable research requirements.

For example, an external AI service may be suitable for exploring public literature but not authorized to receive identifiable participant transcripts.

De-identification may reduce risk without necessarily eliminating confidentiality obligations.

The agreement should establish how confidential student research will be protected during AI-assisted processing and feedback.

Where authorization is uncertain, processing should not proceed until the necessary clarification is obtained.

8. Identify Approved AI Tools and Processing Environments

A thesis agreement may identify which tools or deployment arrangements are authorized for particular activities.

However, tool approval should not be interpreted as universal permission for every data category or research task.

Relevant considerations include institutional access controls, provider terms, retention arrangements, and permitted processing purposes.

Some projects may use locally deployed systems or institutionally managed services.

These arrangements still require appropriate security and governance.

The agreement should establish how new tools will be evaluated before they are introduced into consequential research workflows.

9. Define What AI Use Must Be Documented

Documentation should help reconstruct consequential AI contributions and research decisions.

For example, a student may record that AI suggested an analytical approach, identify how the recommendation was evaluated, and explain the final decision.

For generated code, the student may preserve the verified script, relevant tests, and computational environment.

However, saving every prompt is not universally necessary.

The agreement should identify which activities require records, what those records should contain, and where they will be stored.

The question of documenting student AI use should be distinguished from general interaction monitoring.

10. Establish Disclosure Expectations

Students should understand when AI assistance must be communicated to the supervisor, thesis committee, or publication venue.

Routine assistance may be covered by standing permission, while consequential methodological or interpretive contributions may require specific discussion.

Publication disclosure should follow the intended journal's requirements.

The International Committee of Medical Journal Editors provides recommendations concerning AI-assisted technologies and human author accountability.

Other publishers may establish different requirements.

The agreement should distinguish supervisory disclosure from formal publication reporting.

The issue of which AI interactions students should report to supervisors should be addressed before substantial assistance occurs.

11. Clarify Whether AI Prompt Histories Will Be Reviewed

Students and supervisors should know whether consequential AI interactions must be preserved and when records may be requested.

Complete conversation histories may contain unrelated personal information, confidential material, and preliminary research ideas.

For many supervisory purposes, targeted excerpts or research decision records may provide sufficient evidence.

However, some approved research protocols or formal institutional processes may require more extensive records.

The agreement should establish a proportionate approach consistent with privacy, confidentiality, and institutional authority.

Requirements should not be introduced retrospectively as though they had always applied.

12. Define the Supervisor's Responsibilities

The agreement should not place every AI-related responsibility on the student.

Supervisors are responsible for providing appropriate academic guidance, evaluating consequential research decisions, and applying institutional requirements within their authority.

If supervisors use AI to prepare feedback, they should also verify substantive recommendations and protect confidential research material.

For example, a supervisor should not forward unverified AI-generated methodological criticism as an established finding.

The question of AI-assisted supervisory review of student drafts should be addressed as part of reciprocal accountability.

Student and supervisor responsibilities may differ, but both should remain transparent and defensible.

13. Specify How AI Will Be Used During Assessment

Thesis research is both a scholarly activity and evidence of degree-level competence.

Some assessments may require students to demonstrate particular skills independently.

For example, an oral examination may assess whether the student can justify methodological choices without relying on live AI assistance.

Other activities may explicitly evaluate the student's ability to critique automated recommendations.

The agreement should distinguish research assistance from assessment conditions.

Formal assessment rules must remain consistent with graduate school requirements and the authority of the relevant examiners.

14. Establish What Happens When AI-Generated Work Is Incorrect

AI-generated references, code, and interpretations may contain errors.

The agreement should establish that scientifically inaccurate work must be corrected regardless of whether the error originated from AI or a human.

For example, an AI-generated statistical script may run successfully while incorrectly merging records.

The student should investigate and correct the procedure, with appropriate supervisory or specialist assistance.

The correction should be documented when it affects consequential research results.

Scientific error should not automatically be equated with academic misconduct.

15. Establish How Possible AI-Use Violations Will Be Handled

Students should know which institutional procedures apply when an AI restriction may have been violated.

For example, unauthorized assistance during an independent assessment may be governed by academic integrity rules.

Unauthorized processing of confidential participant data may require a separate information-security or research ethics response.

These concerns should not automatically be treated as equivalent.

The agreement should refer to applicable institutional processes rather than invent its own disciplinary framework.

Questions concerning student AI-use violations require relevant evidence, fair procedures, and appropriate authority.

16. Define How Disagreements Will Be Resolved

Students and supervisors may disagree about whether AI assistance is methodologically suitable, ethically appropriate, or permitted.

The agreement should identify how such disagreements will be discussed and which institutional authorities can provide clarification.

For example, uncertainty about participant data processing may require consultation with the research ethics or data-governance office.

A disagreement about independent thesis competence may require graduate program guidance.

Students should not be expected to resolve conflicting institutional instructions independently.

The agreement should support respectful discussion without implying that either party's preference automatically overrides binding requirements.

17. Review the Agreement at Research Milestones

AI use may change as a thesis progresses.

During proposal development, AI may be used primarily for literature exploration and methodological learning.

During data collection, confidentiality and research ethics requirements may become more prominent.

During analysis, generated code and interpretation may require additional verification.

During manuscript preparation, publication disclosure and authorship obligations may become relevant.

The agreement should identify when expectations will be reconsidered and how revisions will be documented.

What Should a Written AI Agreement Actually Contain?

A concise agreement can organize the preceding expectations into a practical reference.

This is a proposed structure rather than a mandatory institutional form.

Some institutions may already provide an approved template or require additional information.

Why the Agreement Should Be Proportionate

A thesis involving public literature and nonconfidential computational exercises may require different safeguards from a clinical study involving identifiable participant records.

Likewise, routine language editing may not justify the same documentation burden as AI-assisted changes to a primary analytical model.

The NIST AI Risk Management Framework emphasizes evaluating risks in context.

For thesis supervision, proportionality means aligning requirements with the significance of the activity and the potential consequences of error or unauthorized processing.

However, proportionality does not permit supervisors to disregard binding institutional restrictions.

Why an Agreement Should Not Become a Surveillance System

The purpose of a supervisory agreement is to clarify responsibilities and support research development.

It should not automatically require continuous monitoring of every student interaction with AI.

Excessive documentation may create administrative burdens and introduce privacy concerns without providing meaningful evidence of research competence.

Supervisors should identify the information genuinely necessary to evaluate consequential decisions.

Students should remain able to explore ideas, ask preliminary questions, and develop independent scholarly judgment within the agreed boundaries.

Accountability should be established through appropriate evidence rather than the quantity of interaction records collected.

Why an Agreement Cannot Guarantee Research Integrity

A written agreement may reduce uncertainty, but it cannot guarantee that every AI-generated output will be accurate or that every research decision will be appropriate.

Students may still misunderstand methodological advice, and supervisors may overlook errors.

Research integrity depends on actual practices of verification, accurate reporting, and responsible oversight.

The agreement should therefore be treated as a working governance instrument rather than evidence that compliance has automatically been achieved.

Its effectiveness should be evaluated through the quality of research decisions and the clarity of responsibilities.

Creating an AI-Use Agreement Before a Mixed-Methods Thesis

The example illustrates how a relatively concise agreement can clarify responsibilities across different research phases.

Its purpose is not to predict every future AI interaction, but to establish defensible principles and procedures for decisions that materially affect the thesis.

Misconceptions About Student-Supervisor AI Agreements

How to Establish a Workable Thesis AI Agreement

Students and supervisors should develop the agreement before substantial AI-assisted research begins, using applicable institutional requirements as the foundation.

The agreement should clarify practical responsibilities without creating unnecessary monitoring or duplicating existing formal policies.

Start With an Explicit Conversation

Discuss anticipated AI activities, independent learning expectations, confidentiality, and research accountability.

Identify which questions require institutional clarification before the agreement can be finalized.

Record the Agreement in Accessible Form

Use the institution's approved template where available.

Otherwise, a concise written record can identify the agreed permissions, responsibilities, and review procedures.

Make Verification Responsibilities Specific

Explain how AI-generated references, methods, code, and interpretations should be checked.

Assign responsibilities according to expertise and the research context.

Clarify Both Student and Supervisor AI Use

Address how students may use AI in producing research and how supervisors may use it when providing feedback.

Apply appropriate confidentiality and transparency standards to both activities.

Review the Agreement at Meaningful Milestones

Revisit expectations when research enters data collection, analysis, thesis examination, or publication preparation.

Document material changes and communicate them before the affected work begins.

Is Your Student-Supervisor AI Agreement Ready?

Common Questions About Thesis AI-Use Agreements

Agree on Responsibilities Before AI Influences the Thesis

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.