Fake Papers, Real Problem: Detecting Hallucinated References from GenAI

Overview

Generative AI tools can be useful for writing and learning, but they can also produce convincing-looking academic references that do not actually exist. These hallucinated references create challenges for academic integrity, assessment, and feedback to students.

This project will explore whether we can build and evaluate a tool that automatically checks reference lists and identifies potentially non-existent or suspicious references. The project is connected to the broader goal of supporting responsible GenAI use in education and improving academic integrity processes.

You will help to develop a tool that takes a reference list as input and checks whether the listed sources appear to be real. The project will evaluate how design choices, such as prompt engineering, use of external databases, reference parsing, and verification strategies, affect the tool’s ability to identify false references. The student may also gather feedback from academic staff about how such a tool could be useful, where it might fail, and how it should be presented to both students and educators.

Research engagement

You will engage with applied research in responsible AI and education technology. This will include: a focused literature review on GenAI hallucinations and academic integrity, hands-on software prototyping, experimental evaluation of tool performance, and stakeholder-informed design. The project may also involve designing a small feedback activity with academic staff to understand how such a tool could be used responsibly in teaching and assessment contexts.

Research activities

You will work with the supervisor and, where appropriate, academic staff who have experience with assessment and academic integrity. Activities may include:

• Reviewing examples of hallucinated references produced by GenAI tools.
• Creating a small benchmark dataset containing real, incorrect, and non-existent references.
• Testing different approaches for checking references, such as prompt-based verification, DOI lookup, title and author matching, and scholarly metadata searches.
• Evaluating how system design choices affect precision, recall, and false positives.
• Analysing common failure cases, such as incomplete references, incorrect metadata, or real papers with small citation errors.
• Gathering informal or structured feedback from academic stakeholders on the usefulness, risks, and presentation of the tool.
• Preparing a short final report and demonstration.

Research skills

You will gain experience in applied GenAI research, Python-based software prototyping, experimental design, and evaluation using metrics such as precision and recall. You will also develop skills in data cleaning, benchmark construction, responsible AI thinking, and communicating technical findings to a non-technical audience.

Outcomes

Expected outcomes include a prototype reference-checking tool or workflow, a small evaluation dataset, quantitative results on tool performance, an analysis of common failure modes, and recommendations for how such a tool could be used responsibly by academics and students.

Skills and experience

You should have some programming experience, preferably in Python.

Start date

2 November, 2026

End date

19 February, 2027

Location

QUT Gardens Point or Online

Additional information

You will receive regular supervision and guidance throughout the project. Depending on progress, you may have access to an existing prototype, example reference lists, GenAI tools, scholarly metadata services, and support with experimental design and evaluation.  If stakeholder feedback is collected from academic staff, any required ethics or governance considerations will be discussed before the activity begins.

Keywords

• Generative AI
• GenAI
• Artificial Intelligence
• AI
• Education technology
• Reponsible AI

Contact

Dimity Miller

N/A

d24.miller@qut.edu.au