03 · What You Need to Know
Determining When Student AI Use Requires Supervisory Disclosure
Why Disclosure Should Be Connected to Research Significance
The purpose of supervisory disclosure is to ensure that important research activities can be evaluated and that responsibilities remain clear.
Disclosure can help a supervisor understand how an analytical decision was developed, whether generated information was verified, and whether the student remains capable of defending the work.
However, not every AI interaction affects the research in a meaningful way.
For example, using AI to clarify a general term may not require the same reporting as using AI to select a statistical model.
The distinction should be based on the activity's significance, not merely whether a chatbot was opened.
Students should also recognize that applicable institutional rules may require reporting of particular uses regardless of whether they personally consider them significant.
Disclosure, Documentation, and Permission Are Different
These concepts are related but should not be confused.
Disclosure
Informing the supervisor or another relevant authority that AI contributed to a research activity.
Documentation
Preserving relevant information about the AI contribution, verification, and resulting research decisions.
Permission concerns whether the activity is authorized under applicable rules.
A student may disclose an AI interaction without having obtained permission to undertake it.
Similarly, an activity may be permitted under a standing agreement without requiring a separate notification every time it occurs.
Students and supervisors should therefore distinguish what is allowed, what must be recorded, and what must be communicated.
The broader question of documenting student AI use in research concerns the records needed for transparency and reproducibility rather than disclosure alone.
What Counts as Material AI Assistance?
Material assistance is AI involvement that meaningfully affects the intellectual content, methods, evidence, interpretation, or reported procedures of the research.
For example, an AI recommendation that changes the analytical model may be material even if no generated text appears in the thesis.
Likewise, AI-generated questionnaire items may be material because they influence how research constructs are measured.
Materiality is not determined solely by the number of prompts, the amount of generated text, or the time spent using AI.
A single interaction can substantially affect a study, while many routine interactions may have little influence on its scientific conclusions.
The definition should be specified through applicable policies and supervisory expectations.
Which AI Activities Usually Deserve Supervisory Discussion?
| AI-assisted activity |
Disclosure consideration |
Reason |
| Explaining general research terminology |
May be covered by standing permission |
Usually supports learning without directly determining a research decision |
| Routine grammar correction |
Follow applicable rules; individual reporting may not be necessary |
May affect expression rather than substantive scientific content |
| Suggesting literature-search terms |
Discuss if suggestions materially shape the search strategy |
Search decisions affect evidence retrieval |
| Generating questionnaire items |
Substantive assistance generally warrants discussion |
Item content may affect measurement validity |
| Recommending research designs |
Disclose consequential recommendations |
Design choices affect scientific validity |
| Generating analytical code |
Document and discuss consequential implementation |
Code can alter data and results |
| Suggesting qualitative themes |
Disclose material interpretive contributions |
May affect how findings are constructed |
| Interpreting statistical findings |
Disclose substantive assistance |
Interpretations influence research conclusions |
| Drafting substantive thesis sections |
Follow explicit institutional and supervisory requirements |
Raises questions of intellectual contribution and attribution |
| Processing confidential participant information |
Obtain required authorization before use |
Involves research ethics and data-protection obligations |
The classifications are illustrative recommendations, not universal institutional rules.
A particular university may require broader or narrower disclosure, subject to its governing requirements.
Why Routine AI Assistance May Not Need Individual Reporting
Research students routinely consult dictionaries, documentation, statistical manuals, and other information resources.
AI can perform some comparable supporting functions, although its reliability and data-processing characteristics differ.
A supervisor may reasonably permit routine explanatory assistance under standing conditions.
For example, a student might use AI to explain the difference between two statistical concepts before consulting course materials.
If the interaction does not determine the research method or introduce confidential information, separate notification may add little supervisory value.
However, routine assistance should not be used as a label to conceal substantive AI contributions.
Students should understand when an interaction crosses from general support into consequential research decision-making.
Why One AI Interaction Can Be More Important Than Hundreds of Others
The significance of AI use depends on its consequences rather than frequency.
Consider a student who uses AI repeatedly for language practice and general explanations.
Another student consults AI once and accepts a recommendation to exclude observations from the final dataset.
The second interaction may have greater implications for research validity, even though it occurred only once.
Supervisory disclosure should therefore prioritize consequential influence.
Counting prompts alone provides little information about the quality of research decisions or the student's competence.
Should Students Disclose AI Use That They Ultimately Rejected?
Not every rejected suggestion requires reporting.
However, rejected AI recommendations may be relevant when they materially influenced the research process or explain a consequential decision.
For example, a student may initially consider an AI-recommended analytical model before rejecting it after examining the study's assumptions.
Recording this reasoning may support methodological transparency and supervisory discussion.
By contrast, a discarded suggestion about paragraph wording may have little relevance.
The appropriate level of disclosure depends on the research context and the purpose of the record.
When Should Students Inform Supervisors Before Using AI?
Some activities require prior authorization rather than disclosure afterward.
For example, a student planning to process identifiable participant records through an external AI service should confirm that the processing is authorized before uploading the material.
Similarly, an AI-assisted procedure that materially changes an approved research protocol may require appropriate review before implementation.
Supervisory approval alone may not be sufficient when institutional research ethics or data-governance requirements apply.
Students should distinguish activities that can be reported during ordinary supervision from those that require permission before proceeding.
Watch Out
Disclosure after an unauthorized upload does not retroactively authorize the processing. Where confidential or restricted information is involved, the required permission and safeguards should be established before using the AI system.
How Should Supervisors Establish Disclosure Thresholds?
Supervisors should explain which activities require immediate consultation, which can be reported at scheduled meetings, and which are covered by standing permission.
These expectations should be established early and reviewed as the project develops.
For example, routine language editing may be permitted without individual notification, while substantive changes to analytical procedures may require prior discussion.
The expectations should be consistent with institutional requirements.
This is part of setting clear supervisory expectations for student AI use.
Students should not have to infer disclosure thresholds from a supervisor's personal reaction to previous work.
Can Students Be Required to Report Every AI Interaction?
An institution or supervisor may establish detailed documentation requirements for particular educational or research purposes, subject to applicable authority and policies.
For example, a study investigating how students interact with AI may legitimately require comprehensive interaction records under an approved protocol.
However, universal reporting of every minor interaction may be disproportionate for ordinary thesis supervision.
Such requirements can create administrative burdens and expose information unrelated to consequential research decisions.
They may also encourage compliance-oriented record collection rather than meaningful discussion of research reasoning.
Supervisors should be able to explain why the requested information is necessary and how it will be handled.
What Should Students Say When Disclosing Material AI Assistance?
A useful disclosure should identify the activity, nature of the assistance, and its influence on the research.
It should also explain how consequential outputs were verified.
For example, a student might report that AI generated preliminary statistical code, which was subsequently reviewed, tested, and modified before use.
Another might explain that AI suggested alternative qualitative interpretations, but the final analysis was developed through examination of the original transcripts.
The disclosure should be accurate and sufficiently specific to support supervisory evaluation.
It need not reproduce every unrelated interaction unless applicable requirements demand that level of detail.
How Can Students Disclose AI Use Without Oversharing?
Students should avoid unnecessarily including confidential participant information, private conversations, or unrelated personal material in AI-use disclosures.
A concise description of the consequential assistance may be sufficient for many supervisory purposes.
Where detailed records are required, they should be stored and shared through authorized arrangements.
Data minimization is relevant because collecting additional information can introduce privacy and confidentiality risks.
Supervisors should establish who may access records and how long they should be retained when institutional procedures require such decisions.
Does Disclosure Prove That the Student Understands the Work?
No.
A student may disclose extensive AI assistance while remaining unable to explain the resulting analysis.
Another may use AI appropriately and demonstrate strong independent understanding.
Disclosure provides information about the research process, but it does not directly measure competence.
Supervisors should assess whether students can explain the methods, evidence, and interpretations represented in their work.
The distinction is central to assessing independent research competence when AI is permitted.
How Does Disclosure Relate to Research Integrity?
The European Code of Conduct for Research Integrity identifies reliability, honesty, respect, and accountability as fundamental principles.
Accurately representing research procedures and intellectual contributions is consistent with these principles.
However, the Code should not be interpreted as establishing a universal obligation to report every AI prompt to a supervisor.
Specific requirements depend on the institution, research project, and scholarly context.
Students should understand that permitted AI assistance may still require verification and that undisclosed assistance may raise concerns when disclosure was explicitly required.
Whether a particular failure constitutes misconduct must be determined under applicable definitions and procedures.
Should Students Disclose AI Use to Coauthors as Well?
Supervisory disclosure and coauthor communication serve related but different purposes.
Coauthors may need to know about consequential AI assistance because they share responsibility for verifying and approving the scholarly work.
For example, AI-generated analytical code or substantive interpretations may affect the evidence that all authors must evaluate.
However, coauthor expectations should reflect the project's agreements and applicable publication policies.
Students should not assume that telling a supervisor automatically satisfies every obligation to collaborators.
The question of coauthor awareness of AI-assisted research requires attention to collective accountability.
How Does Supervisory Disclosure Differ From Journal Disclosure?
Supervisory disclosure supports research guidance, competence assessment, and project accountability.
Journal disclosure communicates relevant AI assistance according to the publication venue's requirements.
The International Committee of Medical Journal Editors provides recommendations concerning disclosure of AI-assisted technologies and human author responsibility.
Other journals and publishers may establish different requirements.
Students should consult the current policy of the intended journal before submission.
A disclosure sufficient for a supervisory meeting may not satisfy a journal's instructions, and a journal statement may not provide all the information needed for thesis supervision.
What If the Student Forgot to Disclose Earlier AI Use?
The student should review the applicable requirements and provide an accurate account of relevant assistance where possible.
They should avoid inventing prompts, dates, or verification records to reconstruct an incomplete history.
The supervisor should determine whether the omission concerns a misunderstanding, inadequate documentation, or a possible violation of an established requirement.
These possibilities should not be treated as equivalent.
Where a formal compliance concern exists, the institution's procedures should be followed.
Correcting an incomplete record may be appropriate, but it does not automatically resolve every possible issue involving unauthorized processing or misrepresented scholarly work.
Why Trust Should Not Depend on Constant AI Monitoring
Research supervision requires both accountability and the development of student autonomy.
Requiring continuous reporting of every AI interaction may shift attention toward monitoring behavior rather than evaluating research quality.
However, a complete absence of disclosure expectations can make consequential AI contributions difficult to reconstruct.
A proportionate approach establishes clear boundaries and uses ordinary supervisory discussions to examine important decisions.
Students should feel able to raise uncertainties about AI assistance without assuming that every question will be treated as evidence of misconduct.
Supervisors should remain attentive to research validity while avoiding surveillance practices that lack a clear educational or institutional justification.
What If the Supervisor and Student Disagree About Disclosure?
Disagreements may concern what counts as material assistance, whether a particular record is necessary, or which requirements apply.
The supervisor should explain the purpose and authority of the requested disclosure.
The student should have an appropriate opportunity to clarify the nature of the assistance and raise legitimate confidentiality concerns.
Where institutional rules are unclear, the responsible graduate school or research governance office may need to provide guidance.
New supervisory expectations should be communicated clearly and applied consistently with existing requirements.
Fair disclosure practices depend on transparency from both parties.
