KPMG AI report withdrawn after citations found flawed and fabricated

KPMG’s latest AI research has become a cautionary tale for consultants, auditors and marketers inside the firm: a paper meant to showcase agentic AI was withdrawn after its references were found to be riddled with errors, fabrications and misattribution. The episode puts a spotlight on a basic workplace risk in the AI era: speed can impress clients and colleagues, but weak verification can erase trust in a single release.

The report, “Total Experience: Redefining excellence in the age of agentic AI,” was published in October 2025 as part of KPMG International’s long-running Global Customer Experience Excellence study. That benchmark is now in its 16th year and draws on more than 80,000 consumers across 16 markets worldwide, which made the citation failure more damaging because the paper carried the weight of a flagship KPMG research product.

GPTZero said it checked all 45 citations in the report and found only five that accurately pointed to real, uncorrupted sources. It flagged 40 citation titles as fake, said 28 citations contained paraphrased or fabricated components, and said 12 were too vague or flawed to verify. City A.M. reported that about half the factual claims backed by the citations appeared false or misattributed, and the report was removed from KPMG’s homepage after scrutiny. KPMG said it was reviewing the circumstances surrounding publication.

The examples show how easily AI-assisted research can go off the rails when no one checks the underlying material. One cited source was a 2019 East Japan Railway press release, used to support claims about AI-agent usage even though the term “agentic AI” did not enter wider public discussion until 2024. Other references confused article subjects with authors, including a TfL-related blog post credited to Transport for London and a RetailNews article about UNIQLO attributed to UNIQLO itself. GPTZero called the pattern “vibe citing,” its term for AI-generated or AI-assisted references that are fabricated, fused from multiple sources or heavily altered.

The credibility hit is sharper because KPMG is also selling the upside of agentic AI. On its own site, the firm says the technology can automate complex processes and improve human-AI collaboration, and one KPMG page says it could lift workforce efficiency by up to 30 percent and cut operational costs by 25 percent by 2027. In the same month that KPMG’s report said 55% of CEOs ranked AI as their top investment priority, KPMG’s 2025 CEO Outlook put the figure at 71%, underscoring how a single unchecked statistic can weaken the firm’s own narrative.

For KPMG employees, the practical lesson is straightforward. If generative AI is being used to draft client work, internal research or thought leadership, it needs visible human ownership, source-by-source verification and clear accountability before anything goes out under the firm’s name. In professional services, the product is trust, and a flawed citation trail can turn an AI story into a governance problem fast.