KPMG withdraws AI report after investigation uncovers widespread inaccuracies

A report published by KPMG on the growing role of artificial intelligence in customer experience has been withdrawn after an investigation revealed numerous inaccuracies, fabricated citations and questionable claims linked to AI-generated content.

The report, titled Total Experience: Redefining Excellence in the Age of Agentic AI, was released in October and examined how businesses were using AI technologies to enhance customer engagement and service delivery. KPMG, one of the world’s largest professional services firms, is part of the so-called Big Four alongside Deloitte, PricewaterhouseCoopers and Ernst & Young.

Concerns about the report emerged following an investigation by GPTZero, a company known for developing AI content detection tools. The findings were later verified by the Financial Times, which reviewed the report and confirmed several of the issues identified by GPTZero.

Investigation reveals fabricated citations and references

According to GPTZero, the report contained significant problems with its sourcing and references. Investigators examined 45 citations included in the document and found that only five were accurately linked to genuine sources.

The review found that 28 citations appeared to alter existing source titles or add information not present in the original material. A further 12 citations were described as too vague to verify. GPTZero referred to the practice of AI systems generating inaccurate or fabricated references as “vibe citing”.

The findings have raised concerns about the growing use of AI tools in professional research and publishing. While AI can help speed up information gathering and drafting, experts warn that automated systems can sometimes generate convincing but inaccurate information when reliable verification processes are not in place.

The issue is particularly significant because reports produced by major consulting and professional services firms are often used as trusted references by businesses, researchers and journalists. Errors contained in such publications can therefore spread widely as other organisations rely on the information for their own work.

Major companies dispute claims made in the report

Beyond the citation issues, investigators also questioned many of the report’s central claims. GPTZero estimated that about half of the examples and assertions in the document were either incorrect or inaccurately attributed.

The company suggested that the problems may have stemmed from an AI-powered research tool attempting to identify examples of agentic AI systems in real-world settings. According to GPTZero, some of the examples appeared to exaggerate or misrepresent the capabilities of existing technologies.

One example involved Emirates. The report stated that the airline had launched an AI-powered chatbot named Sara that could communicate with passengers and modify flight bookings on their behalf. However, Sara was introduced in 2023 as a digital assistant rather than an advanced AI chatbot. The assistant could not alter passenger bookings.

Another disputed claim involved the Swiss investment bank UBS. KPMG’s report stated that the bank had integrated agentic AI throughout areas including investment advice, risk management and compliance monitoring. UBS reportedly told the Financial Times that the information was “factually incorrect”.

Questions were also raised about claims concerning Swiss Federal Railways, commonly known as SBB. The report suggested that the railway operator had deployed AI agents capable of helping travellers plan, book and optimise journeys based on personal preferences, real-time travel conditions and environmental considerations. An SBB spokesperson disputed the statement, saying it was “not accurate.”

These examples have highlighted the risks associated with relying heavily on AI-generated research without sufficient human oversight and fact-checking. As businesses increasingly adopt AI tools, experts continue to stress the importance of verifying information before publication.

Concerns grow over the impact of AI-generated misinformation

The controversy has sparked broader debate about the reliability of AI-generated content in corporate research and thought-leadership reports. Industry observers note that documents produced by well-known firms often influence business decisions, academic research and media coverage.

GPTZero chief executive Edward Tian warned that inaccurate reports from highly respected organisations could have consequences beyond a single publication. He said such documents could “poison the well of information” by introducing false information that may later be repeated by researchers, journalists and AI systems trained on publicly available material.

The concern is that AI-generated inaccuracies can create a cycle in which false information is repeatedly cited and amplified. This process can make it increasingly difficult to distinguish between verified facts and AI-generated errors, particularly when the source is considered trustworthy.

In response to the findings, a KPMG spokesperson stated that the firm “takes the accuracy and integrity of its published content seriously.” Following the investigation, the company removed the report from circulation and began examining how it was produced.

KPMG said it is currently “reviewing the circumstances surrounding its publication.” The incident serves as a reminder that while AI technologies offer significant benefits for research and content creation, organisations remain responsible for ensuring the accuracy of information released under their name.