How an AI Report on AI Became a Cautionary Tale: KPMG's Report Pulled Over Fabricated Citations

KPMG's report was part of its annual Global Customer Experience Excellence study, designed to showcase how leading organizations are "delivering on the Total Experience promise" through AI that is personal, intuitive, and anticipatory . The final document was organized into case studies that named specific companies and public bodies, detailing their supposed agentic AI transformations.

The problems started when GPTZero, an AI-detection research firm, analyzed the report's 45 citations. Its findings were damning :

• Only 5 citations (roughly 11%) were accurate and verifiable, correctly pointing to the intended source.
• 28 citations contained completely made-up content or paraphrased titles that did not match any real source.
• 12 citations were too vague or flawed for GPTZero to determine whether a source even existed.
• Around half of the factual claims backed by those 45 citations appeared to be false or misattributed.

The most glaring example involved a claim that East Japan Railway used agentic AI for customer support. The citation pointed to a 2019 press release—years before agentic AI was a recognized term . Another citation in the report claimed that 55% of CEOs rank AI as their top investment priority, contradicting KPMG's own publicly available CEO survey data .

The Organizations That Pushed Back

When the Financial Times independently verified claims in the KPMG report, several prominent organizations denied the depictions of their AI usage or confirmed they were inaccurate. According to reporting, the organizations that disputed the report's claims included :

• UBS – The Swiss bank was portrayed as having deployed AI agents across investment advisory, risk management, and compliance monitoring. UBS flatly denied this, calling the claims "factually incorrect" .
• UK National Health Service (NHS) – The report linked NHS organizations to significant agentic AI deployments, which were disputed when verified externally .
• Swiss Federal Railways (SBB) – The report claimed SBB had become a "holistic mobility orchestrator" powered by AI agents. SBB confirmed the description was "not accurate" .
• Transport for London (TfL) – TfL was among the public transit organizations named in the disputed case studies .

KPMG removed the report while investigating the allegations .

GPTZero's Verdict: AI Hallucinations, Not Just Sloppy Editing

GPTZero did not frame the errors as mere editorial mistakes. The firm attributed the widespread false citations and fabricated case studies to AI hallucinations—output from generative AI models that appears plausible but is factually incorrect or entirely invented .

GPTZero also introduced a term to describe what may have happened behind the scenes: "vibe citing." The concept suggests that AI tools, when prompted to support a narrative, generate citations that "feel right" rather than correspond to real sources . In KPMG's case, this meant references that sounded academically rigorous or journalistically credible but which, upon inspection, led nowhere.

The firm's investigation concluded that the report's composition was consistent with heavy AI assistance that had not been subjected to adequate human verification. The combination of hallucinated footnotes, misattributed statistics, and non-existent case studies painted a picture of a research process where AI output was published with minimal oversight .

The Bigger Pattern: EY Canada's Retraction One Month Earlier

The KPMG incident did not happen in a vacuum. In May 2026, just weeks before KPMG's report came under scrutiny, EY Canada withdrew a cybersecurity study titled "Points of Attack: Uncovering Cyber Threats and Fraud in Loyalty Systems" after GPTZero flagged extensive AI hallucination problems .

The EY report was found to have fabricated 16 of its 27 references—roughly 60%—including a citation to a non-existent McKinsey & Company document called the "Loyalty Economics Report" . GPTZero also estimated that 72% of the report's content was AI-generated .

EY Canada removed the report from its website and stated it was reviewing the circumstances that led to its publication . Like the KPMG case, the retraction raised serious questions about how one of the world's largest professional services firms could publish marketing collateral containing demonstrably false information without catching it during review.

Both incidents exposed a structural vulnerability in the Big Four: as firms race to publish thought leadership on AI topics, they are increasingly relying on the very tools they're writing about—sometimes with embarrassing and reputationally damaging consequences .

What This Means for Professional Services and AI-Assisted Research

The back-to-back retractions at EY and KPMG in 2026 are more than isolated PR problems. They represent a warning for any knowledge-intensive industry where credibility is currency.

Professional services firms have spent years advising clients to "embrace AI responsibly" . When those same firms are caught publishing AI-hallucinated research—about AI, no less—the contradiction undermines their authority as trusted advisors. The KPMG report was not an internal draft or a low-stakes blog post. It was a flagship global study intended to demonstrate the firm's expertise in customer experience and emerging technology .

The episode highlights a growing asymmetry: generative AI can produce polished, citation-dense research reports in minutes, but verifying each claim and tracing every footnote to its original source still demands hours of skilled human labor. GPTZero's methodology in both cases—manually checking every citation against its claimed source—showed that the verification gap is where catastrophic errors slip through .

For organizations publishing research in 2026 and beyond, the lesson is unambiguous: AI can accelerate drafting, but it cannot replace verification. The reputational cost of a retracted report, especially one caught citing sources that don't exist, far outweighs the time saved by skipping the fact-checking step.