Documenting AI use in a research ethics protocol

Usage context

A university researcher or academic who is preparing a human-subjects research protocol and must declare AI tool use to their institution's Research Ethics Committee (REC), a national funding agency, or an accreditation panel. Since 2025, many universities require ethics forms to include an explicit AI section specifying what data were processed, with which tool, and what safeguards were applied. The prompt helps draft that section in formal prose at the level of detail committees require.

Originally written for Chilean classrooms.

Paste before

Before opening the model, have ready:

• Your role in the project (principal investigator, co-investigator, thesis student)
• The name of your institution and the type of committee you are submitting to (internal REC, national funding agency, accreditation panel)
• The project name or a brief description of its objective
• The list of AI tools you used or plan to use, with the study phase in which each was used
• A generic description of the type of data processed with AI (no real data or identifiers)
• The safeguards you applied (anonymization, local instance, human review of outputs, etc.)
• If your institution has an internal policy on AI in research, have it on hand to reference

Privacy rule: Describe data types generically ("transcribed and anonymized interviews coded with a participant identifier"). Do not paste real transcripts, survey responses, field records with names, or any data that could identify participants. Submitting identifiable data to an external model may violate applicable personal data protection law and jeopardize the ethical approval of your protocol.

Act as a research ethics advisor with experience in ethics forms from universities, national funding agency applications, and accreditation panel submissions. I need your help drafting the artificial intelligence use section of my ethics protocol. The resulting text must be technically precise, address the questions committees typically ask, and be written in the formal register appropriate to an institutional ethics document.

**My role in the project:** {{researcher_role}}

**Institution and committee I am submitting to:** {{institution_name}} — {{committee_type}}

**Project name:** {{project_name}}

**AI tools used or planned in the study:**
{{tool_list}}
(For each one, indicate: tool name, provider, study phase in which it was used, and specific task. Example: "Claude Sonnet, Anthropic, analysis phase, AI-assisted thematic coding of anonymized transcribed interviews")

**Type of data processed or to be processed with AI:**
{{data_type}}
(Describe the generic type, without real data or identifiers. Example: "anonymized individual interview transcripts coded with an alphanumeric identifier")

**Safeguards implemented:**
{{safeguards}}
(Example: "full anonymization before entering the model; API use with no-data-retention option per provider policy; review of outputs by two researchers before incorporation into analysis")

**Purpose of AI use in the study:**
{{ai_purpose}}
(Example: "reduce bias in qualitative coding through AI-assisted thematic analysis; accelerate audio transcription during the fieldwork phase")

Please draft an AI use declaration section that includes:

1. A technically precise description of which tools were used, in which study phase, and for which specific task.
2. Specification of the type of data entered into the model and justification for why it does not represent additional risk to participants — or, if risk exists, what it is and how it is mitigated.
3. The safeguards implemented to protect participant privacy and ensure the integrity of the analysis.
4. A statement on human oversight: who reviewed the model's outputs, how discrepancies between AI-assisted analysis and the research team's judgment were handled, and how that review process is documented.
5. If applicable, a note on internal institutional or funder policies regarding AI use in human-subjects research.

Write in formal prose, in English, aligned with the register of institutional ethics forms. Length: 250–400 words. Mark with [PENDING: complete with...] any aspect you cannot resolve from the information I provided.

Expected output

Example (condensed):

Declaration on the use of artificial intelligence in the research process

This project involves the use of AI in three phases. Design phase: ChatGPT-4o (OpenAI) for literature synthesis; no participant data were entered. Fieldwork phase: Whisper (OpenAI) running on a local instance with no internet connection for automatic transcription, reviewed manually by the team. Analysis phase: Claude Sonnet (Anthropic) via API with no-data-retention option enabled; materials entered consist exclusively of anonymized transcripts coded with alphanumeric identifiers, containing no data that could identify participants.

Emergent categories proposed by the model are validated independently by two researchers on the team; discrepancies are resolved by consensus and recorded in the project's analytical decision log. No output is incorporated into the final analysis without explicit human validation.

[PENDING: complete with reference to current institutional policy on AI use in human-subjects research, or your national funding agency's guidelines, and cite the relevant instrument.]

Watch out for

• The model may propose safeguards that sound robust but are vague. Phrases like "data are encrypted" or "the provider guarantees confidentiality" will not satisfy a committee asking for a concrete mechanism. Verify that what you declare matches exactly what you applied: if you used the API with the no-data-retention option enabled, say so explicitly; if you anonymized before entering data, describe the procedure.
• The generated text may omit human oversight, which is the element ethics committees most frequently flag as insufficient in AI protocols. If the draft does not include who reviewed the outputs and how that review was documented, ask the model to add that section before including it in the form.
• Do not enter participant data into the model to "illustrate" the type of analysis. Even if the temptation is to paste a transcript excerpt so the model better understands the context, doing so could compromise participant confidentiality and jeopardize protocol approval. Use only generic descriptions of the data type.

Suggested iteration

If your committee uses a form with predefined fields: "Here are the exact questions from the ethics form I need to answer: [paste questions]. Adapt the declaration you drafted to respond directly to each field, without additional free text between them."

If you need to add a section on ethical risks of AI use: "Add a paragraph on the potential risks that AI use introduces into the analysis process — algorithmic bias, automatic transcription errors, loss of qualitative nuance — and explain how they will be addressed methodologically in the study."

If the protocol is for a national funding agency and you want to align with current guidelines: "Revise the text noting that [your funding agency] requires an explicit declaration of the AI model, the phase of the research cycle in which it is used, and the provider's privacy policy. Indicate whether any aspect needs to be added or reformulated to meet those criteria."