KPMG's AI Report Had 40 of 45 Fabricated Citations — GPTZero Calls It 'Vibe Citing'

In October 2025, KPMG published a report titled Total Experience: Redefining Excellence in the Age of Agentic AI. The document circulated widely in consulting circles, was cited by multiple industry publications, and helped establish KPMG’s positioning in the emerging agentic AI market. By early June 2026, AI detection firm GPTZero had completed a forensic review and found the report largely fictional: of 45 citations, only 5 pointed to real sources.

The other 40 ranged from entirely invented titles to mangled fusions of multiple real papers. GPTZero’s Hallucination Check tool flagged 89% of citations as flawed and assessed the document itself as “likely AI-generated.”

What Is “Vibe Citing”

GPTZero coined a new term for the phenomenon: vibe citing. The analogy is to vibe coding, the practice of letting a language model generate code without understanding it. Vibe citing is what happens when an AI generates references that look credible — correct author name format, plausible journal titles, proper citation style — but don’t correspond to anything real.

The damage mechanism differs from vibe coding in an important way. With vibe coding, the failure shows up immediately: the program doesn’t run. With vibe citing, the errors circulate for weeks or months before anyone checks. By then, secondary sources have quoted the fabricated data, AI systems have ingested it, and the false claims have entered the broader information supply chain.

The Organizations Named

KPMG’s report fabricated AI deployment case studies for several well-known institutions:

UBS was described as integrating AI agents across investment advisory, risk management, and compliance monitoring through a platform co-developed with Microsoft. A UBS spokesperson told the Financial Times the assertions were factually incorrect.

NHS Greater Manchester and Transport for London were both listed as AI adoption success stories. Both organizations denied the characterizations.

The Emirates example was particularly egregious: the report claimed the airline’s chatbot Sara could modify passenger bookings. Sara is actually a physical robot introduced in 2023 that has no reservation management capabilities.

JR East was cited via a 2019 press release as evidence of agentic AI deployment. In 2019, commercial agentic AI technology did not exist.

Austrian energy company Verbund had its identity merged with a startup it funded, producing a fictional case study about AI-managed household appliances.

KPMG’s Response

After the Financial Times validated GPTZero’s findings, KPMG removed the report and issued a statement saying the company was “reviewing the circumstances surrounding its publication,” reiterating its policy requiring human oversight to validate content.

The statement raised an obvious question: if that oversight policy existed when the report was published, why weren’t the citations checked? The policy was on paper. Whether it was applied is a different question.

The pattern is not isolated to KPMG. Deloitte recently refunded the Australian government after AI-generated content appeared in a taxpayer-funded report. The Big Four appear to be running similar AI workflows with similar risks of exposure.

The Harm Has Already Spread

Before KPMG pulled the report, its fabricated statistics had already been quoted in multiple industry publications and at least one major Czech newspaper. Those secondary sources were subsequently ingested by other AI systems as reference material, propagating the false claims further.

GPTZero CEO Edward Tian has described this as a systemic but underappreciated risk in enterprise AI adoption: the risk is less about models making mistakes and more about model outputs being distributed at scale without adequate human review. A Big Four consulting firm’s report carries enough institutional credibility that readers skip the citation check entirely.

KPMG is among the consultancies that sell AI governance frameworks to enterprise clients.

If this was useful, subscribe to the newsletter for weekly AI PM insights and GenAI case studies.

Related: Claude Fable 5 Released: Anthropic’s Opening Move Against GPT-5