Chasing the Hallucinations: KPMG's AI-Powered Attempt at "Redefining Excellence"

2026-06-12 · Source: Welcome to the Artificial Intelligence Incident Database · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, extended

Summary

A GPTZero investigation revealed widespread hallucinated citations and claims in KPMG's October 2025 report, "Total Experience: Redefining Excellence in the Age of Agentic AI," which summarizes an annual global consumer experience study. Out of 45 citations, only five accurately pointed to real sources. Another 28 citations contained paraphrased titles or fake components for real sources, while 12 were too vague to verify. Approximately half of the claims supported by these citations also appeared fake or misattributed, likely due to an AI research tool over-complying with a request for "agentic AI" examples. GPTZero defines these as "vibe citations," accidental fake references via LLM hallucinations, ranging from entirely fabricated to heavily altered versions of real sources. These errors are consistent with LLM generation, suggesting a lack of human verification.

Key takeaway

For research scientists or consultants relying on external reports, you must critically evaluate all citations, especially those from high-traffic sources. The prevalence of "vibe citations" and AI-generated claims, as seen in KPMG's report, means unverified information can quickly propagate. Implement rigorous human fact-checking and utilize AI hallucination detection tools to prevent incorporating fabricated data into your own work, safeguarding your credibility and the accuracy of your analyses.

Key insights

AI-generated "vibe citations" and claims are prevalent in professional reports, risking widespread misinformation.

Principles

• LLM-generated citations often contain consistent, subtle fabrications.
• Unverified AI-generated content can lead to significant factual errors.
• "Vibe citing" encompasses a spectrum from fully fake to heavily altered references.

Method

GPTZero's Hallucination Check tool identifies AI-generated citations by flagging unverified or flawed references, including paraphrased titles and incorrect authors.

In practice

• Use AI detection tools to verify citations in reports, especially those from LLMs.
• Implement human review for all AI-generated research content and references.

Topics

• AI Hallucinations
• Citation Verification
• Agentic AI
• Consulting Reports
• GPTZero
• Factual Accuracy

Best for: CTO, Executive, AI Product Manager, AI Scientist, Research Scientist, Consultant

Related on AIssential

• KPMG Allegedly Published AI Report Filled With Hallucinations — Unverified AI-generated content, particularly citations, can lead to widespread misinformation and reputational damage.
• Where Fake Citations Are Made: Tracing Field-Level Hallucination to Specific Neurons in LLMs — LLM citation hallucination is field-specific, with distinct neural mechanisms for different bibliographic components.
• Hallucinated citations are polluting the scientific literature. What can be done? — AI-generated hallucinated citations are rapidly increasing, polluting academic literature with fabricated and inaccurate references.
• They're Finally Solving AI Hallucinations! — ChatGPT's hallucination rate has significantly decreased, improving reliability for specific, unambiguous information requests.
• Why AI Confidently Makes Things Up — AI hallucination stems from pattern generation, not misperception of a factual database.
• Citation Discipline in Spec-Driven Development: A Cross-Model Empirical Study of Output Determinism and Automated Hallucination Detection in LLM-Generated Code — Mandatory inline code citations in LLM-generated code trade output determinism for automated hallucination detection.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Welcome to the Artificial Intelligence Incident Database.