Over the past year, a team of GPTZero investigators has used our Hallucination Check tool to uncover hallucinated citations in government reports, academic papers submitted to prestigious machine learning / artificial intelligence conferences like ICLR and NeurIPS, and research products from two of the big four consulting firms: Deloitte and Ernst & Young. This is the second in a series of investigations into reports containing hallucinated citations surfaced by an automated search pipeline.
This investigation analyzes a KPMG report from October 2025 on customer experience and agentic AI. The report, titled Total Experience: Redefining Excellence in the Age of Agentic AI, summarizes the results of an annual study on consumer experience around the world. Of the
45 citations in the report, only
five accurately point to real sources. Another
28 citations provide paraphrased titles and/or fake components for a real source. The final
12 are too vague or flawed to accurately determine if a source exists. Additionally,
around half of the claims evidenced by the 45 citations appear to be fake or misattributed — likely the result of an AI research tool over-complying with a request to find examples of “agentic AI” in the wild.
These findings raise an important question: if a citation generally points to a real source (but includes paraphrased or fabricated components) can it still be a hallucination?
At GPTZero, our answer is “yes”, although we acknowledge that some hallucinations are more obvious than others. We use the term “vibe citing” to describe the accidental creation of fake references via LLM hallucinations across a spectrum of severity. Vibe Citations can include references that are entirely fabricated (fake authors, fake title, and fake container/locators), fusions of two or more real references (authors of paper A paired with the title of paper B), or paraphrased or heavily altered versions of real citations. Our definition excludes common human errors.
On the Menu: KPMG #
One of the “big four” consulting firms, KPMG is a global organization of member firms operating in 138 countries and employing more than 250,000 people. These member firms are coordinated by KPMG International, an English private company headquartered in London.
In October 2025, KPMG International published * Total Experience: Redefining Excellence in the Age of Agentic AI* to summarize the latest findings from the firm’s long-running Global Customer Experience Excellence (CEE) study. The report is available in several versions, including a
and localized variants like this global report. The localized variants are generally identical apart from the authorship of the Foreword and the contact information on the report’s back cover.
Belgian version## KPMG’s Vibe Citations
Consulting firms use a condensed citation style that makes source matching difficult. In line with industry norms, the citations in *Total Experience are formatted as endnotes and *generally provide a title, container (publisher or website name), and a year. Many of the citations also include an author, but the given author overlaps so frequently with the container or title that it is usually redundant. Because the citations are incomplete (lacking “real” authors, URLs, and other identifiers), finding the "real" matching can involve extensive detective work and require certain assumptions about the report’s authors.
**The task of citation verification is made even more challenging by citation titles: 40 of 45 are fake. **In most cases, the titles seem to be paraphrases that loosely match real sources (as corroborated by context and other citation components). However, between the paraphrased titles, garbled authors, and missing URLs, 12 citations are too ambiguous for either our tool or in-house expert to determine a confident source match.
GPTZero’s Hallucination Check classifies citations with paraphrased titles or incorrect authors as possible hallucinations (here are the technical details). For this reason, most of
Total Experience’s 45 citations are
highlighted red by our tool. The table below shows our assessment of each citation using four categories: "Exist," "Unsure," "Hallucination," or "Unable to determine." To be as fair as possible to KPMG, we've used a stricter definition of "hallucination" than we do for academic citations (which generally include more information and cite sources with greater permanence).
| # | Citation Text | Verdict | Justification | Potential Match |
|---|---|---|---|---|
| 1 | Salesforce. C-suite AI Priorities in 2025: YouGov Research. Salesforce, February 2025. | Hallucination | Title is wrong. Report was produced by YouGov for Salesforce. | |
Intellihub and Bunnings Launch Zelora: A Game-Changer for Home Electrification in AustraliaNew Agentic AI Platform from AUSTRIACARD HOLDINGSFunktionen und Vorteile von Mein ELBAVERBUND X Ventures invests in AI start-up Qurrent for energy solutionsInternet banking | Česká spořitelnaŠkoda’s voice assistant Laura now enhanced with ChatGPT capabilitiesAlzaBNP Paribas Accelerates AI Integration: What This Means in PracticeHow Doctolib uses AI to empower healthcare practitionersSourceHow Google Cloud’s Automotive AI Agent is Transforming In-Car Experience with Mercedes-Benz – Unite.AIFour ways to shop: REWE takes hybrid and cashless shopping to a new level – REWE GroupGenerative AI in E-Commerce: Otto Group Introduces New Innovation Projects for Enhanced Customer Experience and Optimized ProcessesDigital transformation for the bank | Intesa SanpaoloEnel Energy ManagementSource“Toyota Woven City,” a Test Course for Mobility, Completes Phase 1 Construction and Prepares for LaunchJR East Joins MaaS AllianceUNIQLO: Digital Transformation Strategy | RetailNews.aiThe Routing agent: a key component for building agentic AI assistants at scale – BBVA AI FactoryImpulsa tu pyme con herramientas de gestión de negocios – ClipING Bank transforming operations through agentic AI | Computer WeeklySourceKPN transforms its operations and empowers its workforce using Azure OpenAI | Microsoft Customer StoriesUnionBank advances digital consumer banking with partnerships, AI and platform-driven growthSun Life Philippines Goes Live with ALLFINANZ SPARKIgloo Tech Solutions Launched to Modernize SEA Insurance Operations | EveryTechEverResponsible AI in banking: Gaining a competitive edge | DBS BankGrab deploys agentic AI to empower merchants and driver partnersSingapore Airlines And OpenAI Partner To Enhance Customer Experience And Operational Efficiency With AI SolutionsChatbot Adam TB – virtual financial advisorPenta Hospitals CZSlovak LinesHow UBS Deploys Microsoft AI Platform in Global Operations | FinTech MagazineGrowth Engine for Digital Health in Switzerland | Kanton ZürichMultimodal mobility | SBBSourceGaranti BBVA enhances its smart assistant Ugi using generative artificial intelligenceIsbank’s Personal Assistant Maxi and Capabilities | Türkiye İş BankasıSourceGreater Manchester leads national roll out of a new AI solution to help accelerate lung cancer diagnosisUsing Artificial Intelligence to Improve Transport in London – Cow-Shed StartupHow Nike Customer Experience Uses Artificial Intelligence To Improve Engagement & PersonalizationIntroducing ScintillaOur team suspects that the authors of Total Experience used an AI-powered referencing tool to generate the report’s citations because the errors are both mistakes typical of Large Language Models (LLMs) and consistent throughout the reference list. A human would not consistently paraphrase titles, mistake topics for authors (e.g., citation 9), or repeat information across multiple components (e.g., citation 2).
Citations 32 to 40, which are formatted differently and include quotation marks around titles, appear to have been generated by a more accurate tool or edited by a human. Most of the citations classified as “Exist” by either Hallucination Check or our in-house expert are in this section.
KPMG Redefines the Evidence #
All 45 citations in Total Experience are tied to footnotes in the fourth section of the report: “Key Insights by Country and Region.” Each provides evidence for a country-specific case study showcasing AI augmentation of customer experience (CX) tasks. In many cases, these case studies seem to have been generated (in part or in whole) by AI.
For example, page 29 includes two Austrian case studies with footnotes: Consulting firms tend to shorten reference sections by citing claims selectively. It is therefore possible that the sources cited in these cases are only meant to evidence certain claims, or to provide a limited example. Still, many of the claims made in the cases studies (especially those about agentic AI) seem exaggerated or fabricated.
Returning to page 29, notice that the second case study claims Austrian electricity provider Verbund is using AI agents to conduct real-time analytics as part of an “energy-as-a-service ecosystem.” Yet the best match for the paired citation, a 2025 press release from Verbund, describes how Verbund’s venture capital unit, Verbund X, is funding a Swedish startup named Qurrent. While Qurrent is developing an AI-powered platform to automate energy optimization across a grid, the press release doesn’t mention “AI agents,” “weather forecasts,” “smart appliances,” or “electric vehicles.” In short, the case study seems to conflate Verbund and Qurrent while fabricating a detailed claim that Qurrent’s commercial and industrial level platform operates at the household-level (using AI agents to optimize smart appliances and EV charging).
In another example, the report seems to cite a 2019 press release from the East Japan Railway Company (JR East) as evidence that the railway is leveraging AI agents to make travel recommendations and predict service disruptions. Since the press release predates the market availability of agentic AI by a half-decade (and doesn’t mention AI at all), these claims are probably fabricated.
Because of the poor quality of the footnoted citations, our team didn’t attempt an exhaustive verification of every claim. However, the table below presents selective findings by categorizing each citation into one of three categories: "Support," "Partial support," or "Does not support." The twelve citations without a category are those labeled “Unable to determine” above.
| # | Citation Text | Expert Label | Justification | Potential Match |
|---|---|---|---|---|
| 1 | Salesforce. C-suite AI Priorities in 2025: YouGov Research. Salesforce, February 2025. | Support | ||
Intellihub and Bunnings Launch Zelora: A Game-Changer for Home Electrification in AustraliaNew Agentic AI Platform from AUSTRIACARD HOLDINGSFunktionen und Vorteile von Mein ELBAVERBUND X Ventures invests in AI start-up Qurrent for energy solutionsInternet banking | Česká spořitelnaŠkoda’s voice assistant Laura now enhanced with ChatGPT capabilitiesAlzaBNP Paribas Accelerates AI Integration: What This Means in PracticeHow Doctolib uses AI to empower healthcare practitionersSourceHow Google Cloud’s Automotive AI Agent is Transforming In-Car Experience with Mercedes-Benz – Unite.AIFour ways to shop: REWE takes hybrid and cashless shopping to a new level – REWE GroupGenerative AI in E-Commerce: Otto Group Introduces New Innovation Projects for Enhanced Customer Experience and Optimized ProcessesDigital transformation for the bank | Intesa SanpaoloEnel Energy ManagementSource“Toyota Woven City,” a Test Course for Mobility, Completes Phase 1 Construction and Prepares for LaunchJR East Joins MaaS AllianceUNIQLO: Digital Transformation Strategy | RetailNews.aiThe Routing agent: a key component for building agentic AI assistants at scale – BBVA AI FactoryImpulsa tu pyme con herramientas de gestión de negocios – ClipING Bank transforming operations through agentic AI | Computer WeeklySourceKPN transforms its operations and empowers its workforce using Azure OpenAI | Microsoft Customer StoriesUnionBank advances digital consumer banking with partnerships, AI and platform-driven growthSun Life Philippines Goes Live with ALLFINANZ SPARKIgloo Tech Solutions Launched to Modernize SEA Insurance Operations | EveryTechEverResponsible AI in banking: Gaining a competitive edge | DBS BankGrab deploys agentic AI to empower merchants and driver partnersSingapore Airlines And OpenAI Partner To Enhance Customer Experience And Operational Efficiency With AI SolutionsChatbot Adam TB – virtual financial advisorPenta Hospitals CZSlovak LinesHow UBS Deploys Microsoft AI Platform in Global Operations | FinTech MagazineGrowth Engine for Digital Health in Switzerland | Kanton ZürichMultimodal mobility | SBBSourceGaranti BBVA enhances its smart assistant Ugi using generative artificial intelligenceIsbank’s Personal Assistant Maxi and Capabilities | Türkiye İş BankasıSourceGreater Manchester leads national roll out of a new AI solution to help accelerate lung cancer diagnosisUsing Artificial Intelligence to Improve Transport in London – Cow-Shed StartupHow Nike Customer Experience Uses Artificial Intelligence To Improve Engagement & PersonalizationIntroducing ScintillaThese factual errors are not confined to the report’s footnoted passages. On page 42, the authors claim that Emirates airline has adopted a mobile chatbot named Sara (false) that can converse directly with passengers (partially true) and change their flights (false). In fact, Sara is a robot assistant introduced by Emirates in 2023 (not a chatbot) that lacks the ability to alter flight bookings.
Other discrepancies may have a plausible explanation, but seem like suspicious errors at first glance. For instance, Total Experience cites "KPMG research" as the source of a claim that 55% of CEOs rank AI “as their top investment priority”(7). Yet the KPMG 2025 CEO Outlook report, released the same month, claims that 71% of CEOs believe "AI is a top investment priority”(8). It seems logical that the research referenced by Total Experience is CEO Outlook, yet 55% is lower than both the 2025 percentage (71) and the 2024 percentage (64) provided by CEO Outlook for the same statistic.
Vibes Have Consequences #
Publications by prestigious consulting firms that are hosted on high-traffic websites rank highly in search algorithms and AI research queries. It's therefore no surprise that the claims and statistics included in KPMG’s *Total Experience *are seeping across the internet and surfacing in news reports, blog posts, and conversations with LLMs. Since publication, the report has been covered by industry publications like CXM,
[, and](https://www.customerexperiencedive.com/news/cx-quality-marginally-improved-globally-kpmg/816646/)
__CX Dive__[, as well as a major](https://www.mi-3.com.au/18-11-2025/specsavers-tops-tally-utilities-organisations-gain-ground-and-financial-services-go)
__Mi3__[. In these stories, claims, including the 55% statistic mentioned above, have been](https://byznys.hn.cz/c1-67823690-pruzkum-kpmg-v-hodnoceni-firem-opet-vyhrala-air-bank-druhy-je-rohlik-cz)
Czech newspaper.
__repeated and recycled__KPMG’s report is also being cited by LLMs, which can strip claims of context and make them more difficult to corroborate:
A Legacy of Excellence (Redefined) #
Defenders of Total Experience will argue that citations which point to real sources, however strangely, are the result of formatting issues rather than hallucinations. We would place this argument in the “partial support” category. While it is possible to find a matching source for most citations in KPMG’s report, the errors are consistent, glaring, and unlikely to be human in origin. Calling them “formatting issues” mirrors the language used by White House press secretary Karoline Leavitt last May and ignores the fact that many components are paraphrased or fabricated.
The facts speak for themselves. Most of the citations in this report have at least one hallucinated component (generally the title). Many also have fabricated dates or authors. At least 16 citations are hallucinations according to our definition, and at least 12 are so ambiguous that identifying an original source is impossible. Only five are accurate.
The problems caused by these vibe citations are compounded by a relentless series of fake or misattributed claims. It seems likely that an LLM research tool was prompted to find suitable case studies of companies using agentic AI in different markets. This tool seems to have conflated sources, exaggerated claims, and injected references to agentic AI that were copied into the final report without verification. We suspect no human at KPMG double-checked the citations, the claims, or the sources before *Total Experience *was published.
As always, GPTZero contends that vibe citations are a clear and present danger to researchers, academics, consultants, students, and anybody else who happens to search the internet for information. Our Hallucination Check tool is an answer to this threat, as well as a way to hold consulting firms accountable for the research they produce.
Now, more than ever, it's crazy to accept citations on faith. Try GPTZero’s Hallucination Check for yourself, or reach out to GPTZero’s team.
Definition of a Vibe Citation
GPTZero defines a vibe citation as a citation that likely resulted from the use of generative AI. Vibe citing results in errors common to LLM generations, but rare in human-written text, such as:
- Combining or paraphrasing the titles, author(s), and/or locators from one or more real sources
- Fabricating the author(s), title, URL/DOI, and/or container (ex. publisher, journal, conference) of a source
- Modifying the author(s) or title of a source by extrapolating a first name from an initial, dropping and/or adding authors, or lengthening/shortening the title.
Our definition excludes obvious spelling mistakes, dead URLs, missing locators, and other errors that are plausibly human. The following table shows the difference between a real citation, a flawed citation, and a hallucinated citation according to our methodology.
| Real Citation | Flawed Citation | Hallucinated Citation |
|---|---|---|
| Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. nature, 521:436-444, 2015. | Y. LeCun, Y. Bengio, and Geoff Hinton. Deep leaning. nature, 521(7553):436-444, 2015. | Samuel LeCun Jackson. Deep learning. Science & Nature: 23-45, 2021. |
| A. Yang, B. Zhang, B. Hui, B. Gao, B. Yu, C. Li, D. Liu, J. Tu, J. Zhou, J. Lin, et al. Qwen2.5–math technical report: Toward mathematical expert model via self-improvement. arXiv:2409.12122, 2024. | A. Yang, (missing author), B. Hui, B. Gao, B. Yu, C. Li/, D. Liu, J. Tu, J. Zhou, J. Lin, et al. Qwen 2. 5–math technical report: Toward mathematical expert model via self-improvement. arXiv preprint arXiv:2409.12122, 2024. | A. Yang, B. Yang, C. Yang, et al. Qwen3.5–mathematical report for iterative model self-improvement. arXiv:2909.12233, 2024. |
Like GPTZero’s AI Detector, Hallucination Check has an extremely low false negative rate, so we catch 99 out of 100 flawed citations. Because our tool will flag any citation that can’t be verified online, the false positive rate is higher.