Generative AI tools like ChatGPT are changing student writing – and with it, the reality of academic integrity. Many schools and universities are asking the same questions: How much AI cheating is happening? Is this plagiarism? How can we detect it reliably?

Here’s a clear, measurable signal that often survives into final submissions: hallucinated references (fake or wrong citations). Unlike “AI-sounding” text, a fabricated source is an objective error. And that makes it one of the most actionable indicators for cheating, plagiarism, and weak research practice.

What our study found (DACH student papers)

We analyzed 368 bibliographies from undergraduate and graduate submissions in Germany, Austria, and Switzerland. The results show the problem is already mainstream:

• 36.4% of texts contained at least one invalid reference

• In affected papers, 12.7% of listed sources were false on average

• 18.2% of all papers had high-error bibliographies with more than 20% wrong references

A single fake citation can undermine trust in the entire paper. If one source doesn’t exist, educators can reasonably ask: Were any sources actually read? This is why hallucinated references are increasingly viewed as an indirect but strong indicator of unchecked AI use and a practical focus area for academic quality assurance.

Why citation verification beats “AI detection”

Traditional detection tools often rely on guessing whether writing looks machine-generated – and they unavoidably produce false positives. Writing process analysis is a great way to identify personal contribution, but often enough no version history data is available anymore, because the writinng is only being checked ‘post hoc’. Reference verification is different: it checks whether evidence is real. It supports academic integrity without turning assessment into a “style trial.”

What educators and decision makers can do now

• Add citation checking to grading rubrics (fast, fair, defensible)

• Use scalable tools to reduce workload and standardize quality control

• Teach AI literacy: AI can draft text – but it cannot replace research and originality

Use Mentafy’s Reference Verifier for faster detection

If you want a practical way to identify hallucinated citations and prevent AI-driven cheating and plagiarism, Mentafy’s Reference Verifier helps educators quickly flag non-verifiable references and thereby turning an invisible integrity issue into a measurable workflow.

–> Download full study (PDF)
–> Try Mentafy’s Reference Verifier