Version 1.0 · May 2026

Ethical AI in
Academic Research

A practical framework for researchers and doctoral students navigating AI tools with integrity.

Dr. Sonal Katyal · Program Director, Faculty of Transformational Economics · James Lind Institute, Geneva

Explore the Framework → ⬇ Download Full Guide
3Non-Negotiable Rules
8Research Stages Mapped
4Prompting Formula Parts
8Self-Assessment Checks
Foundation

The Three Non-Negotiable Rules

Regardless of AI tool, research context, or institutional policy — these three rules apply universally.

01
Disclose

Document the name, version, and purpose of every AI tool used in your research. This disclosure belongs in your methodology section. Transparency is not optional — it is the foundation of academic trust.

02
Log

Maintain an audit trail of significant prompts and AI outputs from the first day of your research. Your research diary is your record of responsible use — and your evidence if your integrity is ever questioned.

03
Own

Every argument, conclusion, and intellectual claim in your dissertation must be traceable to your own judgment. If you cannot defend an argument in a viva without AI, it does not belong in your dissertation.

Section 2

AI Use Classification

Use this as a self-assessment tool before, during, and after using AI in your work. Not all AI use is equivalent.

🟢
Green — Ethical AI Use (Disclosure Required)
These uses are considered ethical when properly disclosed
Using purpose-built screening tools (e.g. Rayyan, Cadima) for abstract screening · Statistical analysis tools (e.g. JASP, jamovi) on your own original data · Literature synthesis tools (e.g. Consensus, Elicit) to identify and map research — outputs must be verified · AI writing tools (e.g. Trinka, Paperpal) for grammar and academic tone — intellectual content must remain yours · Using AI to explain a concept you then verify independently · AI to structure an argument you then develop in your own voice
🟡
Amber — Use With Caution (Disclosure + Judgment Needed)
Carries ethical risk — substantial original contribution required
AI-generated first draft of a section you then substantially rewrite — the rewrite must be genuine and traceable · AI summaries of papers — only if you have read the originals and verify accuracy · AI-suggested research arguments — only if you critically evaluate and independently develop them · AI-generated visualisations or tables — only if underlying data is original and outputs are verified
🔴
Red — Academic Integrity Violation
Academic misconduct regardless of institutional policy
Submitting AI-generated text as your own without disclosure · Including AI-generated citations without verifying they exist · Asking AI to write examination answers, coursework, or dissertation chapters submitted as unaided work · Sharing confidential research data or participant information with AI tools · Using AI to impersonate your research voice in ways that misrepresent your competence to examiners
Section 3

AI Tools by Research Stage

Specialist tools purpose-built for academic research, mapped to each stage of the research process.

Literature Screening
PRISMA-compliant abstract screening with full audit trail. Researcher sets criteria; tool executes sorting.
Free · Citable outputs
RayyanCadima
Statistical Analysis
Produces reproducible R-scripts and APA-ready tables. Outputs are citable in methodology.
Free · APA-ready outputs
JASPjamovi
Citation Mapping
Maps citation networks from verified academic databases. Does not fabricate references.
Free · Verified sources only
ResearchRabbit
Literature Synthesis
Draws from real peer-reviewed databases. Outputs show source papers. Verify all claims independently.
Free tier available
ConsensusElicit
Paper Summarisation
Turns papers into comparison tables and scaffolds. Use as synthesis starting point — not final output.
Free tier available
Scholarcy
Reference Management
Manages citations with full metadata. Use to verify all references before submission.
Free · Open source
Zotero
Academic Writing Quality
Trained on peer-reviewed corpora. Enforces academic register, hedging, and tense — not just grammar.
Free tier available
TrinkaPaperpalWritefull
Integrity Verification
Checks submissions for AI-generated content before final submission. Use as a self-check tool.
Institutional access for Turnitin
GPTZeroTurnitin AI
⚠ Critical Note on General-Purpose LLMs ChatGPT, Claude, Gemini, and similar tools can support tasks like explaining concepts or improving clarity — but must never be used as primary research tools for finding or citing literature. They do not search academic databases and regularly produce fabricated citations. Always verify any source provided by a general-purpose LLM before including it in your work.
Section 4

The 4-Part Precision Prompting Formula

The quality and reliability of AI output depends directly on the quality of the input prompt. Vague prompts produce vague — and potentially unreliable — outputs.

R
Role
Specify who the AI should act as.
"Act as an academic peer reviewer specialising in health economics..."
T
Task
State precisely what you want the AI to do.
"Identify the three main methodological limitations in the following abstract..."
C
Context
Provide the relevant background and purpose.
"This is for a PhD dissertation on maternal healthcare utilisation in South Asia..."
C
Constraint
Set clear boundaries on what AI should and should not do.
"Use only peer-reviewed sources. Do not fabricate citations. Clearly flag any uncertainty."
Why constraints matter: A vague prompt to a general-purpose AI tool is more likely to produce a vague, generalised, or fabricated response. Adding a CONSTRAINT that explicitly instructs the AI not to fabricate citations and to flag uncertainty significantly reduces — though does not eliminate — the risk of unreliable output. The researcher must still verify all AI output independently.
Section 5

AI Hallucination & Citation Verification

AI hallucination is particularly serious in academic research because fabricated citations can look entirely credible — with correctly formatted authors, journals, and page ranges — but not exist.

⚠ Warning Signs of a Fabricated Citation

  • The DOI does not resolve to a real paper when searched at doi.org
  • The journal name exists but volume, issue, or page numbers are inconsistent with its publication history
  • The author exists but has no record of publishing on that topic
  • The paper cannot be found on Google Scholar, Semantic Scholar, PubMed, or SCOPUS
  • The paper appears only on AI-generated content aggregator sites, not in academic databases

Verification Protocol — Before Citing Any AI-Generated Reference

Search the exact title in Google Scholar (scholar.google.com)
Search the DOI at doi.org if one is provided
Search the author name combined with the topic in Semantic Scholar or SCOPUS
If found, access the actual paper and confirm the claim the AI attributed to it is accurate
If not found through any of these methods — do not cite it
Section 6

Disclosure Requirements

In the absence of a specific institutional policy, the following guidance represents good academic practice.

What to Disclose

  • Name and version of each AI tool (e.g. ChatGPT-4o, Claude 3.5, Trinka, JASP 0.19)
  • Purpose for which it was used (e.g. grammar checking, literature screening, data analysis)
  • Stage of research at which it was used (e.g. literature review, write-up)
  • Any limitations of the tool relevant to your research findings

Where to Disclose

  • In the Methodology section of your dissertation — this is the standard location
  • In a dedicated AI Use Statement if your institution requires one
  • In the Acknowledgements for minor uses such as grammar checking
  • In the data availability statement if AI was used in data processing
Sample Methodology Section Disclosure This research used the following AI tools in the conduct and writing of this dissertation: Rayyan was used for abstract screening during the systematic literature review; the researcher set all inclusion and exclusion criteria independently. Trinka AI was used for grammar and academic tone checking of the final draft; all intellectual content and arguments are the author's own. ChatGPT-4 was used to clarify specific statistical concepts, which were subsequently verified against primary sources. No AI tool was used to generate research findings, conclusions, or cited references.
Section 7

Data Privacy & Confidentiality

Data privacy is a distinct and serious concern that sits alongside academic integrity when using AI tools in research.

🚫 Never Share With AI Tools

  • Personal data of research participants — names, contact details, identification numbers
  • Verbatim interview transcripts or focus group recordings without participant consent for AI processing
  • Confidential institutional data, internal reports, or proprietary organisational information
  • Unpublished research findings of colleagues shared in confidence
  • Patient data or clinical records in health research

Why This Matters

Most commercial AI tools process inputs on external servers. Data shared with these tools may be used to train future models, stored in ways outside your control, or subject to data breaches.

Sharing participant data with commercial AI tools without explicit consent may violate data protection regulations including GDPR in Europe and equivalent legislation in other jurisdictions.

Practical Rule: Before entering any data into an AI tool, ask: would I be comfortable if this data appeared in a future AI training dataset? If the answer is no — do not enter it. Use anonymised or synthetic data for any AI-assisted analysis.
Section 8

Pre-Submission Self-Assessment

Use this checklist before submitting any research work in which AI tools were used. Check off each item honestly.

1 Have I disclosed every AI tool used, its purpose, and the stage at which it was used in my methodology section?
2 Have I personally verified every citation in my reference list — including any suggested by AI tools?
3 Can I defend every argument and conclusion in my work without relying on an AI tool to explain it?
4 Have I maintained an audit trail or research diary documenting significant AI interactions?
5 Have I ensured that no personal data, participant information, or confidential data was shared with any AI tool?
6 Have I read and complied with my institution's current AI use policy?
7 Is the intellectual contribution in this work genuinely mine — or have I allowed AI to substitute for my own thinking?
8 Have I used specialist, citable tools (not just general-purpose chatbots) for methodology-critical tasks such as screening and statistical analysis?
0 / 8 Tap each item to check off
Expert Agent Prompt

Use ResearchMind as Your AI Auditor

Paste this prompt into ChatGPT, Claude, or Gemini to activate an AI tuned to this ethics framework.

You are the "ResearchMind AI Agent," a specialized academic auditor based on the framework by Dr. Sonal Katyal (James Lind Institute, Geneva). Your goal is to audit AI use in academic research and doctoral dissertations. When a researcher describes their AI use, you will: 1. Apply the Traffic Light System: 🟢 GREEN – Ethical use with proper disclosure 🟡 AMBER – Use with caution; requires substantial original contribution 🔴 RED – Academic integrity violation 2. Enforce the Three Non-Negotiable Rules: DISCLOSE – Document every AI tool, version, and purpose in methodology LOG – Maintain an audit trail of significant prompts and outputs OWN – Every argument must be traceable to the researcher's own judgment 3. Apply the Precision Prompting Formula (ROLE → TASK → CONTEXT → CONSTRAINT) when helping researchers construct effective academic prompts. 4. Guide users through the Citation Verification Protocol: – Search exact title in Google Scholar – Verify DOI at doi.org – Cross-check on Semantic Scholar or SCOPUS – Confirm the cited claim in the actual paper – Do not include any reference not personally located and read 5. Help draft Methodology Disclosure Statements following the framework's sample. 6. Run the Section 8 Self-Assessment Checklist as a final pre-submission audit. Tone: Professional, academic, and strict regarding integrity. Never excuse a RED-category use. Be constructive but uncompromising on the core principles.
How to use: Copy the prompt above → open ChatGPT, Claude, or Gemini → paste it as your first message. The AI will then respond as the ResearchMind Auditor and apply Dr. Katyal's framework to evaluate your AI use. No subscription or setup required.

Download the Complete Guide

All 8 sections, the full tools table, sample disclosure statements, and the self-assessment checklist — in a single Word document for your research files.

⬇ Download Word Guide (Free)