A student pastes an AI-generated reference list into their thesis. Every citation looks perfect — real author names, real journal names, working DOIs. The student submits. Months later, an examiner runs a reference check tool and discovers that half the sources either don’t exist or have been swapped with unrelated papers.
That student’s thesis may still be accepted. Or it may be flagged during review. Or it may surface years later when someone audits published work that cited the thesis. The consequences of fake references in a thesis are not theoretical — they are real, documented, and escalating.
This article explains what happens when fake references are discovered in a thesis or dissertation, how universities treat the offense under academic integrity policies, how fabrication is being detected, and what you can do to protect yourself.
Fast facts
- Fake references have increased 12-fold since 2023 across biomedical literature alone (Lancet audit, 2025).
- A Dubai court confirmed student dismissal in November 2026 for an AI-written thesis with fabricated citations.
- Cambridge University explicitly states that submitting falsified references is academic misconduct “whether it results from AI use or not” — intent does not matter.
- One student’s 2026 thesis contained fabricated citations; the consequences extended beyond a failing grade to expulsion and a permanent record notation.
The Scale of the Problem
Fake references were once a rare mistake. They are now a documented crisis.
A 2025 audit published in The Lancet examined 2.5 million biomedical papers and found 146,900 fabricated references across approximately 110,000 papers — a 12-fold increase since a comparable 2023 study. One affected paper contained 18 fabricated references out of 30 total.
Even more concerning: 98.4% of affected papers received no publisher action at the time of the audit. No correction, no retraction. The fabricated references were embedded in published work that continued to circulate.
The rate of fabrication per paper shifted from 1 in 2,828 (2023) to 1 in 277 (early 2026). Detection pipelines are increasingly extending beyond journals into the post-graduate review process.
What Happens When Fake References Are Discovered?
The consequences of fabricated references in a thesis depend on when and how they are detected. This section maps the escalation pathway from thesis-level detection through post-graduation outcomes.
Consequence Severity Ladder
| Detection Stage |
Minimum Outcome |
Maximum Outcome |
What Happens to Your Record |
Real-World Examples |
| During thesis review (pre-defense) |
Failing grade on thesis chapter; mandatory resubmission |
Thesis rejected; academic misconduct investigation opened |
Failing mark on transcript; investigation record held by department |
UK university “Sam” case: 92% AI-score, 2 fabricated references, mark capped at 40%, investigation opened |
| During defense / oral examination |
Thesis accepted with mandatory corrections; notation in examiner report |
Degree withheld; resubmission required within set deadline |
Degree deferred; notation may appear on transcript |
Multiple cases documented where thesis accepted “pending reference verification” |
| Post-graduation (during employment or further study) |
Degree revoked; transcript notation of academic misconduct |
Permanent academic record notation; expulsion from subsequent programs |
Degree cancellation; permanent academic record with no statute of limitations |
Harvard’s Andrey Bezrukov — degree (PhD, 1989) revoked in 2011 for fabricated citations; Austrian minister Christine Aschbacher — doctorate revoked |
| In published work (paper or thesis published after graduation) |
Retraction of paper; funding ban; professional sanctions |
Ban from grant funding; professional license revocation; legal penalties |
Professional record damaged; possible legal liability for copyright infringement |
US copyright penalties for severe plagiarism/fabrication: $250,000 + up to 5 years imprisonment |
The key point most students miss: there is no statute of limitations. A fabricated reference discovered in a thesis five years after graduation can still result in degree revocation and a permanent academic record notation.
How Universities Treat Fake References
Universities now explicitly classify fabricated references as academic misconduct regardless of intent.
Cambridge University’s HPS policy (updated October 2026) states:
“HPS considers the submission of falsified information (such as fake references, quotes or grossly misrepresented source material) to be academic misconduct whether it results from AI use or not.”
This is not unique to Cambridge. The Open University’s GenAI policy (September 2026) classifies using generative AI as a “search engine to generate results, data, citations or scholarship” as cognitive off-loading (malpractice).
The critical insight: universities do not need to prove you used AI. They only need to prove the references are falsified. Intent is irrelevant — the act of submitting fabricated citations is the breach.
How Universities Detect Fake References
Detection has shifted dramatically. Universities are no longer relying solely on AI-detector scores (which have high false-positive rates for non-native speakers). They are checking references directly.
Turnitin and Reference Checking
Turnitin can flag unusual patterns in a reference list — mismatched formatting, inconsistent citation styles, references that don’t appear in databases. But Turnitin cannot automatically verify whether a source actually exists. This limitation has led to the development of dedicated reference-checking tools.
Dedicated Detection Tools
Several automated reference-checking tools have emerged as active fields:
- Hallucinator (hallucinator.science) — Scans bibliographies for fabricated references using Crossref, PubMed, and Google Scholar APIs
- Paper Reference Checker (zabbonat.github.io/References-Validation) — Validates every reference against multiple databases
- UB-Mannheim’s check-fake-references — Python script for bulk bibliography verification
These tools allow supervisors and examiners to check entire bibliographies at scale — something manual checking cannot achieve for theses with 100+ references.
Manual Examination Probing
The Open University’s guidance instructs examiners to probe student understanding rather than rely on detection scores. Examiners ask students to explain their sources, justify their choices, and defend their methodology. If a student cannot explain the origin of a reference, fabricated citations are likely to surface during oral examination.
Prevention: How to Avoid Fake References in Your Thesis
The good news: fabricated references are preventable. You do not need to trust AI-generated reference lists. You do not need to check every DOI by hand (though that helps). You need a systematic verification process.
Before You Write
- Never copy-paste AI-generated reference lists into your thesis. AI (ChatGPT, Claude, Gemini, Perplexity) consistently hallucinates plausible-looking but fabricated citations.
- Build your reference list yourself using your university’s library databases, Google Scholar, or your institution’s subscription databases.
- Use a reference manager (Zotero, EndNote, Mendeley) to import citations directly from databases — not from AI output.
While You Write
- Exclude your bibliography from AI editing prompts. If you use Grammarly, ChatGPT, or similar tools for editing, do not include your reference list. AI tools will not verify sources.
- Verify every source you add. Check that the paper exists on the journal’s website or in a trusted database.
- When in doubt, verify the DOI. If a reference has a DOI, resolve it at doi.org and confirm the title, authors, and journal match your citation.
Before You Submit
- Run a reference check using an available tool (Hallucinator, Paper Reference Checker, or your university’s library tool) if available.
- Cross-check every reference against Google Scholar by title. If no result appears, the source is likely fabricated.
- Confirm retraction status for every paper you cite. A retracted paper is not valid evidence. See our guide on checking retraction status for detailed steps.
- Exclude your bibliography from AI editing prompts — AI tools edit text but do not verify sources. If you need your thesis edited, include your references as a separate step after editing so your editor can verify them.
How to Cite AI Tools Themselves (APA 7, MLA 9, Chicago 18th)
If your university allows the use of AI tools in your research, you must cite them correctly. AI cannot be listed as an author — the company behind the AI (OpenAI, Anthropic, Google) is the author.
URLOpenAI.ChatGPT Literature Search [GPT-4o]. https://chat.openai.comMLA 9AI Company, Chat Title [Model Name], Day Month Year.URLAnthropic, Research Question [Claude 3 Opus], 15 Aug.2025. https://www.anthropic.comChicagoAI Company.URLGoogle.
|
| Style |
Format |
Example |
| APA 7 |
AI Company. |
| (Year). Chat Title [Model Name]. |
(2025). |
|
|
| Year. Chat Title [Model Name]. |
2025. Essay Outline Help [Gemini Pro]. https://gemini.google.com |
Key points from the APA Style Blog (September 2025 update):
- AI cannot be an author. Treat it like a tool you used, not a person you interviewed.
- Include the chat title (the topic prompt or your own title for the session).
- Use the share link if the tool provides one; otherwise use the company’s home page URL.
- Disclose use in your methods section if required by your university’s policy.
Using AI Research Tools Safely (Perplexity, Deep Research, Elicit)
Answer engines like Perplexity AI are increasingly popular among students for literature search. They are not inherently dangerous — but they carry specific failure modes you need to understand.
The Perplexity Failure Mode
Perplexity and similar tools can produce “confident-sounding citations that do not actually say what they are credited with saying.” The AI may:
- Invent a paper that does not exist
- Attach a real paper’s citation to a claim the paper never made
- Paraphrase a real source so loosely that it becomes unrecognizable
The CASRAI/NISO guide on Perplexity for research recommends treating these tools as discovery and scoping instruments, not as citation sources. Every link provided by Perplexity should be opened and verified independently.
Safe Usage Checklist
- Use Perplexity, Elicit, or Consensus to identify potential sources — not to build your reference list.
- Open every link the tool provides. Read the abstract. Confirm the source actually addresses your claim.
- Cite the primary source, not the AI tool. Never list Perplexity or ChatGPT as a citation.
- When your university requires AI disclosure, follow the citation formats above.
What Should You Do If You Already Have Fake References in Your Thesis?
If you discover you have already submitted a thesis with fabricated references, do not panic — but do act quickly.
- Replace the fake references. Find legitimate sources that address the same claims. Use Google Scholar, your university library, or database search to locate real papers.
- Verify all other references. Run a full reference check on the remaining sources to ensure they are genuine.
- If the thesis has been submitted but not yet assessed: Contact your supervisor and explain the situation honestly. Most supervisors value transparency over deception.
- If you have already graduated: You cannot control whether a fabricated reference is discovered post-graduation. The best protection is not to submit fabricated references in the first place.
Fake references in a thesis are not a minor formatting error. They are classified as academic misconduct at leading universities worldwide — regardless of whether they resulted from AI use or human error. Consequences range from a failing grade and investigation to degree revocation years after graduation.
The risk is growing: fabricated references have increased 12-fold since 2023. Detection tools are becoming more sophisticated. Universities are checking references at scale.
The most important rule: Never trust an AI-generated reference list without verification. Treat every AI-sourced citation as a claim that requires independent proof. Check that the paper exists, the DOI resolves, and the source actually supports your claim. This takes minutes per reference — far less than it takes to face the consequences of a fabricated citation.
- Lancet audit of 2.5 million biomedical papers (reported via LinkedIn expert community source)
- The Lancet audit (2025) — 98.4% of affected papers had received no publisher action
- Rate shift from 1 in 2,828 (2023) to 1 in 277 (early 2026) — Lancet audit comparison
- GradCoach — academic misconduct overview; Compilatio — degree revocation examples
- Cambridge HPS Policy on generative AI and plagiarism (updated October 2026)
- Open University GenAI Position Statement v2.1 (September 2026)
- ResearchProspect — UK university academic misconduct procedure
- Wageningen University prevention and detection guide
- WUR prevention guide — Hallucinator, Paper Reference Checker, UB-Mannheim scripts
- Open University GenAI Position Statement — examiner probing guidance
- APA Style Blog — Citing generative AI references (McAdoo, Denneny, Lee, September 2025)
- CASRAI / NISO — Perplexity AI for research guide
