Study: ChatGPT Fabricates 1 in 5 Academic Citations

The Research

A 2025 study from Deakin University systematically tested ChatGPT's citation accuracy and found disturbing results:

• 1 in 5 citations (roughly 20%) were completely fabricated
• Half of all citations contained some form of error or hallucination
• Fabricated citations often included fake authors, non-existent journals, and invented DOIs

The Anatomy of a Fake Citation

What makes AI-generated fake citations so dangerous is how plausible they look. A typical fabrication might include:

• Realistic author names
• A plausible-sounding journal
• Proper volume, issue, and page numbers
• A correctly formatted DOI

The problem? The paper doesn't exist. The authors are invented. The journal might be real, but this issue isn't. The DOI leads nowhere.

Real Consequences

• Students have submitted papers with fabricated sources, facing academic integrity violations
• Researchers have cited non-existent papers, damaging their credibility
• Journalists have referenced invented studies in published articles
• Lawyers have cited fake cases in court filings

Protecting Yourself

Always verify every citation from AI:

1. Check the DOI — paste it into doi.org to see if it resolves
2. Search the title — Google Scholar, CrossRef, or the journal's website
3. Verify the authors — Do they exist? Do they work in this field?
4. Check the journal — Is this a real publication? Does it cover this topic?

Trust nothing. Verify everything.