Can You Trust the AI Footnotes?

Introduction

AI can be an excellent research assistant, but it is a poor authority. One of the most important habits in modern critical thinking is learning to use AI for organisation, summarisation and question generation without automatically trusting the citations it provides. The problem is not that AI always invents sources. The problem is that it can mix genuine references, distorted references and entirely fabricated references into a response that sounds equally confident throughout. As a result, readers may mistake polished presentation for verified evidence.

Within the broader goal of doing your own research without cherry-picking, AI creates a new challenge: it can make unsupported claims feel researched. The solution is not to avoid AI altogether. It is to treat every citation, quotation and source attribution as a lead to investigate rather than proof that the investigation has already been done.

Can You Trust the AI Footnotes?

The safest answer is: trust citations only after you have verified them yourself.

Large language models generate text by predicting plausible sequences of words. They are designed to produce answers that sound coherent, not to guarantee that every source reference exists or supports the claim attached to it. Researchers and academic integrity specialists have repeatedly documented fabricated references, incorrect author names, wrong publication dates and citations that point to real papers but misrepresent what those papers actually concluded. [Taylor & Francis Online+2PMC]tandfonline.comTaylor & Francis OnlineHallucinated citations produced by generative artificial…by DB Resnik · 2026 · Cited by 2 — In this article, we…

This problem is significant enough that universities, publishers and research organisations increasingly advise users to independently verify every AI-generated reference before relying on it. Cornell University’s guidance on AI and academic integrity explicitly recommends verifying the accuracy of all citations and references generated with AI tools. [Center for Teaching Innovation]teaching.cornell.eduRequest that students provide verification of references…Read more…

The key lesson is simple: a citation is not evidence until you have checked the original source.

Why Fluent AI Answers Feel More Settled Than They Are

Human beings often use confidence as a shortcut for credibility. AI systems exploit this tendency unintentionally because they usually present information in a smooth, authoritative style.

A traditional search engine shows uncertainty openly. You see competing links, conflicting headlines and different viewpoints. An AI assistant often compresses that messy landscape into a single narrative. The resulting answer may feel more settled and complete than the underlying evidence actually is.

Several warning signs deserve attention:

The danger is not merely factual error. A polished answer can discourage further checking by creating the impression that the checking has already been done.

How Hallucinated References Enter Research

A hallucinated citation is a reference that appears legitimate but is inaccurate or entirely invented.

These errors typically enter research in several ways:

Complete fabrication. The AI invents a paper title, author list, journal or publication year that never existed. Researchers studying citation hallucinations have identified this as a common failure mode. [arXiv]arxiv.orgCompound Deception in Elite Peer Review: A Failure Mode Taxonomy of 100 Fabricated Citations at NeurIPS 2025February 5, 2026…Published: February 5, 2026

Partial corruption. A real paper exists, but details such as authors, publication dates or journal names are altered. [arXiv]arxiv.orgCompound Deception in Elite Peer Review: A Failure Mode Taxonomy of 100 Fabricated Citations at NeurIPS 2025February 5, 2026…Published: February 5, 2026

Real source, wrong claim. The reference exists, but the AI incorrectly describes its findings or conclusions. Studies examining AI-generated references have found that citation errors often involve misrepresenting the content of genuine sources. [Wikipedia]WikipediaHallucination (artificial intelligenceHallucination (artificial intelligence

Invented URLs and documents. Some systems generate links or reports that appear credible but cannot be located because they never existed. Recent research auditing AI-generated citations found measurable rates of non-existent URLs and fabricated references. [arXiv]arxiv.orgDetecting and Correcting Reference Hallucinations in Commercial LLMs and Deep Research AgentsApril 3, 2026…Published: April 3, 2026

Because the fabricated references often resemble real scholarly work, they can evade casual inspection. A reader who checks only the title may not notice that the article, DOI or publication record is missing.

Real-World Failures Show Why Verification Matters

The citation problem is no longer a laboratory curiosity. It has appeared in legal filings, government reports, consulting documents and academic publications.

In 2026, a United States federal judge sanctioned and disqualified multiple lawyers after court filings contained AI-generated legal citations that did not exist. The court emphasised that responsibility for verification remained with the attorneys, regardless of which tool produced the errors. [Reuters]reuters.comJudge rules both sides in lawsuit misused AI, disqualifies lawyersDistrict Judge in Mississippi, Sharion Aycock, has disqualified all attorneys involved in a contract dispute case after discovering both…

The same year, analyses of scientific literature identified a sharp rise in non-existent references associated with AI-assisted writing. One large-scale study estimated that nearly 147,000 hallucinated citations entered scholarly literature during 2025. Researchers warned that these errors were appearing not only in preprints but also in peer-reviewed publications. [arXiv+2Nature]arxiv.orgOpen source on arxiv.org.

Investigations have also found fabricated or distorted references in high-profile reports and books. In one widely discussed case, an AI-related consulting report was found to contain numerous inaccurate citations despite appearing professional and authoritative. [TechRadar]techradar.comThe report contained 45 citations, with only five found to be accurate; the rest were either fabricated, distorted, or misleading. GPTZer…

These incidents matter because citations do more than support arguments. They shape future research. Once a false reference enters circulation, other writers may repeat it without checking the original source.

Safe Uses of AI Before Checking Original Sources

The existence of citation errors does not make AI useless. It simply changes where trust should be placed.

AI is often valuable for:

• Generating research questions.
• Producing summaries that help orient a reader to a topic.
• Identifying competing viewpoints worth investigating.

In these roles, AI acts as an organiser rather than an authority.

A useful mental model is to treat AI as a research assistant who is energetic, fast and sometimes careless. Its suggestions may be valuable, but every important claim still requires independent verification.

A Practical Verification Routine

When AI provides a citation, a few checks can dramatically reduce the risk of being misled.

Find the original document. Do not rely on the citation text alone. Open the paper, report or article itself.

Confirm that it exists. Search library databases, publisher websites or recognised indexing services.

Check whether the source says what the AI claims it says. Many citation errors involve real sources being attached to inaccurate summaries. [Wikipedia]WikipediaHallucination (artificial intelligenceHallucination (artificial intelligence

Look for independent confirmation. Strong claims should usually appear in multiple credible sources.

Follow citations outward. If a source makes an important factual claim, examine where that source obtained its information.

Treat missing sources as a warning sign. If a paper, report or legal case cannot be located despite detailed citation information, assume the reference may be fabricated until proven otherwise.

This process is slower than accepting an AI answer at face value, but it is still far faster than conducting all research manually from scratch.

The Governance Challenge

The growing use of AI in research, journalism, policy and professional work raises governance questions that go beyond individual mistakes.

Institutions increasingly face decisions about how AI-generated citations should be handled, audited and disclosed. Universities are developing verification requirements. Publishers are exploring automated citation-checking tools. Researchers are studying ways to detect fabricated references before publication. [Center for Teaching Innovation+2arXiv]teaching.cornell.eduRequest that students provide verification of references…Read more…

The broader issue is accountability. An AI system cannot be held responsible for a false citation in the way a researcher, journalist, lawyer or policymaker can. For that reason, governance efforts generally focus on human responsibility for verification rather than assuming that citation errors will disappear through technical improvements alone.

In a world where AI can produce convincing evidence trails as easily as convincing prose, critical thinking increasingly means separating the appearance of sourcing from actual sourcing. The strongest defence remains the oldest research habit: go back to the original source and see what it really says.

Amazon book picks

Further Reading

Books and field guides related to Can You Trust the AI Footnotes?. Use these as the next step if you want deeper reading beyond the article.

Book

Calling Bullshit

By Carl T. Bergstrom, Jevin D. West

Encourages source checking and skepticism toward unsupported references.

See on Amazon

Book

Thinking, Fast and Slow

By Daniel Kahneman

Explains why confident-looking outputs can be persuasive.

See on Amazon

Book

The Craft of Research (Chicago Guides to Writing, Editing, an...

By Wayne C. Booth, Gregory G. Colomb et al.

First published 1995. Subjects: Technical writing, Methodology, Research, Recherche, Guides, manuels.

See on Amazon

Book

Verified

By Mike Caulfield, Sam Wineburg

First published 2023. Subjects: Computer literacy, Internet.

See on Amazon

Browse more on Amazon: Calling Bullshit Thinking, Fast and Slow The Craft of Research (Chicago to Writing, Editing, and Publishing) books

As an Amazon Associate I earn from qualifying purchases.

eBay marketplace picks

Marketplace Samples

Example marketplace items related to this page. Use the search link to explore similar finds on eBay.

Shop location

Using USA USA USAUKCanadaAustraliaIreland

Example eBay listing

Social Media Can Wait It's Time For Framed Wall Art Poster Canvas Print Picture

Search eBay.co.uk: social media poster

Browse similar on eBay.co.uk

Example eBay listing

Social Media Killed Romance Love Fr Framed Wall Art Poster Canvas Print Picture

Search eBay.co.uk: social media poster

Browse similar on eBay.co.uk

Example eBay listing

Mercedes F1 Social Media Poster Signed Lewis Hamilton George Russell Saudi 2023

Search eBay.co.uk: social media poster

Browse similar on eBay.co.uk

Example eBay listing

BUSINESS SIGN WORK OFFICE SOCIAL MEDIA DISPLAY EVENT PRINT POSTER FOAM BOARD

Search eBay.co.uk: social media poster

Browse similar on eBay.co.uk

Browse more on eBay.co.uk

Example items shown for inspiration; availability and pricing can change. Branchoria may earn a commission if you purchase through outbound eBay links.

Endnotes

1. Source: pmc.ncbi.nlm.nih.gov
   Link: https://pmc.ncbi.nlm.nih.gov/articles/PMC12658395/
   Source snippet

   by J Linardon · 2025 · Cited by 31 — One type of hallucination generated by LLMs that has received increasing attention among research...

2. Source: teaching.cornell.edu
   Link: https://teaching.cornell.edu/generative-artificial-intelligence/ai-academic-integrity
   Source snippet

   Request that students provide verification of references...Read more...

3. Source: techradar.com
   Title: Tech Radar5 signs that Chat GPT is hallucinating
   Link: https://www.techradar.com/ai-platforms-assistants/5-signs-that-chatgpt-is-hallucinating
   Source snippet

   These hallucinations stem from how AI is trained: by predicting text sequences without verifying facts. 1. **Strange specificity without...

4. Source: arxiv.org
   Link: https://arxiv.org/abs/2602.05930
   Source snippet

   Compound Deception in Elite Peer Review: A Failure Mode Taxonomy of 100 Fabricated Citations at NeurIPS 2025February 5, 2026...

   Published: February 5, 2026

5. Source: arxiv.org
   Link: https://arxiv.org/abs/2602.15871
   Source snippet

   CheckIfExist: Detecting Citation Hallucinations in the Era of...by D Abbonato · 2026 · Cited by 2 — Recent investigations have documente...

6. Source: Wikipedia
   Title: Hallucination (artificial intelligence)
   Link: https://en.wikipedia.org/wiki/Hallucination_%28artificial_intelligence%29

7. Source: arxiv.org
   Link: https://arxiv.org/abs/2604.03173
   Source snippet

   Detecting and Correcting Reference Hallucinations in Commercial LLMs and Deep Research AgentsApril 3, 2026...

   Published: April 3, 2026

8. Source: arxiv.org
   Link: https://arxiv.org/abs/2605.07723

9. Source: reuters.com
   Title: Judge rules both sides in lawsuit misused AI, disqualifies lawyers
   Link: https://www.reuters.com/legal/litigation/judge-rules-both-sides-lawsuit-misused-ai-disqualifies-lawyers-2026-06-09/
   Source snippet

   District Judge in Mississippi, Sharion Aycock, has disqualified all attorneys involved in a contract dispute case after discovering both...

10. Source: nature.com
    Link: https://www.nature.com/articles/d41586-026-01545-1
    Source snippet

    Hallucinated citations highest in social sciences preprints site14 May 2026 — More than 140,000 fake citations across four research repos...

    Published: May 2026

11. Source: nature.com
    Link: https://www.nature.com/articles/d41586-026-00969-z
    Source snippet

    Hallucinated citations are polluting the scientific literature....1 Apr 2026 — Tens of thousands of publications from 2025 might include...

12. Source: techradar.com
    Link: https://www.techradar.com/pro/a-major-kpmg-report-on-ai-was-found-to-be-chock-full-of-ai-hallucinations
    Source snippet

    The report contained 45 citations, with only five found to be accurate; the rest were either fabricated, distorted, or misleading. GPTZer...

13. Source: arxiv.org
    Link: https://arxiv.org/abs/2509.04664
    Source snippet

    [2509.04664] Why Language Models Hallucinateby AT Kalai · 2025 · Cited by 272 — We argue that language models hallucinate because the tra...

14. Source: Wikipedia
    Title: Generative AI
    Link: https://en.wikipedia.org/wiki/Generative_AI
    Source snippet

    Generative AIGenerative artificial intelligence (GenAI) is a subfield of artificial intelligence (AI) that uses generative models to g...

15. Source: research-and-innovation.cornell.edu
    Title: generative ai in academic research
    Link: https://www.research-and-innovation.cornell.edu/generative-ai-in-academic-research/
    Source snippet

    AI in Academic Research: Perspectives and...This report offers perspectives and practical guidelines to the Cornell community, specifica...

16. Source: tandfonline.com
    Link: https://www.tandfonline.com/doi/full/10.1080/08989621.2026.2645390
    Source snippet

    Taylor & Francis OnlineHallucinated citations produced by generative artificial...by DB Resnik · 2026 · Cited by 2 — In this article, we...

17. Source: library.hkust.edu.hk
    Link: https://library.hkust.edu.hk/news-events/news/five-ai-research-tools-referencing-genuine-sources
    Source snippet

    AI Research Tools That Referencing Genuine Sources31 Aug 2023 — In this post, we introduce five AI research tools that can generate answe...

18. Source: cabidigitallibrary.org
    Link: https://www.cabidigitallibrary.org/do/10.5555/blog-artificial-intelligence-hallucinations-best-practice/abs/
    Source snippet

    Nevertheless, verifying responses using other means...Read more...

19. Source: library.yale.edu
    Title: using ai research
    Link: https://library.yale.edu/help-and-research-support/research-support/using-ai-research
    Source snippet

    AI in Research | Yale LibraryWe can help you get started with AI, evaluate the outputs of AI systems, and provide tailored advice on effe...

20. Source: developers.google.com
    Link: https://developers.google.com/machine-learning/gan/generative
    Source snippet

    Machine LearningAug 25, 2025 — A generative model includes the distribution of the data itself, and tells you how likely a given exampl...

Additional References

1. Source: clinicallab.com
   Link: https://www.clinicallab.com/ai-generated-[fake-references
   Source snippet

   AI-Generated Fake References Increasing Across...6 days ago — Study finds AI-generated fake citations are becoming increasingly common i...

2. Source: merriam-webster.com
   Link: https://www.merriam-webster.com/dictionary/generative
   Source snippet

   GENERATIVE Definition & Meaning: having the power or function of generating, originating, producing, or reproducing; Elizabeth Gilbert di...

3. Source: pcgamer.com
   Link: https://www.pcgamer.com/software/ai/both-lawyers-in-case-use-hallucinating-ai-causing-judge-to-throw-up-hands-bar-them-for-2-years-fine-everybody-and-call-the-whole-thing-off-for-60-days/
   Source snippet

   Attorneys Kathleen M. Wilson and Kathryn Y. Williams used generative AI and did not verify the fictitious legal references it produced. T...

4. Source: researchgate.net
   Title: How do you view hallucinations of a Large Language Model
   Link: https://www.researchgate.net/post/How_do_you_view_hallucinations_of_a_Large_Language_Model
   Source snippet

   ?8 Sept 2024 — An inaccurate or incorrect response of a Large Language Model, sometimes referred to as "Hallucinations", is attributed to...

5. Source: insidehighered.com
   Link: https://www.insidehighered.com/opinion/columns/just-visiting/2026/05/06/what-happens-when-fake-citations-layer-bs-bs
   Source snippet

   n author does something like instruct a large language model to help...Read more...

6. Source: sourcely.net
   Title: A I Hallucinated Citations: How to Spot Fake Sources
   Link: https://www.sourcely.net/resources/ai-hallucinated-citations-spot-fake-sources-before-submit
   Source snippet

   AI Hallucinated Citations: How to Spot Fake Sources...May 15, 2026 — Learn quick checks and a step-by-step workflow to verify AI-generat...

   Published: May 15, 2026

7. Source: nursing.ufl.edu
   Link: https://nursing.ufl.edu/2026/03/17/the-illusion-of-evidence-why-fake-ai-citations-demand-caution-in-nursing/
   Source snippet

   Illusion of Evidence: Why Fake AI Citations Demand...17 Mar 2026 — If that foundation is false, the work's credibility is weakened from...

8. Source: enago.com
   Title: ai generated fake references scholarly integrity
   Link: https://www.enago.com/responsible-ai-movement/resources/ai-generated-fake-references-scholarly-integrity
   Source snippet

   AI-Generated References: 1 in 5 Are Fake20 Jan 2026 — Discover how AI tools like ChatGPT are polluting scholarly research with fake citat...

9. Source: medium.com
   Link: https://medium.com/%40nomannayeem/the-fabrication-problem-how-ai-models-generate-fake-citations-urls-and-references-55c052299936
   Source snippet

   heir citations, with some producing fake references more than...Read more...

10. Source: pvrticka.com
    Title: hallucinated citations and phantom references
    Link: https://pvrticka.com/2026/01/11/hallucinated-citations-and-phantom-references/
    Source snippet

    11 Jan 2026 — 'Hallucinated citations' and 'phantom references' are citations and references invented by Large Language Models (LLMs) tha...