Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Two consulting-report audits show why an AI-generated citation must be checked against the source, not trusted because it looks plausible. GPTZero reported that only five of 45 citations in a KPMG report accurately pointed to real sources; the other 40 were not all wholly invented references. The cases point to failures of evidence verification, not proof that AI cannot assist with research.
What did the KPMG citation audit find?
KPMG International’s October 2025 report, Total Experience: Redefining Excellence in the Age of Agentic AI, summarized its Global Customer Experience Excellence study. In an investigation published on 12 June 2026, GPTZero examined the report’s 45 citations and classified five as accurately pointing to real sources. It said 28 had paraphrased titles and/or fake components attached to real sources, while 12 were too vague or flawed for a confident source match. GPTZero also said roughly half of the claims evidenced by the citations appeared fake or misattributed. These are GPTZero’s audit findings, not an independent replication. Read GPTZero’s KPMG investigation.
The shorthand “five real sources out of 45” can mislead. GPTZero said 40 titles were fake, but its breakdown does not establish that 40 citations pointed to entirely invented sources: many were altered or paraphrased matches, and 12 were ambiguous. Its classification could count paraphrased titles or incorrect authors as hallucinations. The number describes GPTZero’s assessment of this report’s citations; it is not a general failure rate for AI research.
What did GPTZero report about EY Canada?
GPTZero’s 14 May 2026 investigation examined EY Canada’s late-2025, 44-page report Points of Attack: Uncovering Cyber Threats and Fraud in Loyalty Systems. The investigators described broken or fake URLs, mismatched titles, misattributions, and contradictions. One example they identified involved the report’s claims about global loyalty points: it cited a $200 billion value and separately treated $200 billion as unredeemed points, while also stating that 30–50% go unused. GPTZero argued those figures were internally inconsistent. They are examples of claims scrutinized by the investigators, not validated estimates of the loyalty-points market or unused balances. Read GPTZero’s EY investigation.
#1 Best Overall
The two audits are not directly comparable scorecards. GPTZero gave a specific citation-count breakdown for KPMG; its EY investigation highlighted particular citation defects and conflicting claims. The categories and evidence presented differ, so the findings do not support ranking the firms or estimating how often AI citations fail across consulting firms generally.
What does “vibe citing” mean?
GPTZero uses “vibe citing” for references that appear credible but were accidentally created through language-model hallucinations. It defines the term as “the accidental creation of fake references via LLM hallucinations across a spectrum of severity.” The phrase covers a range: a citation may have a fabricated URL, an altered title, an incorrect author, or a real source that does not support the attached claim. The label is GPTZero’s terminology, not a formal research standard. GPTZero explains its usage in the KPMG investigation.
How can you check whether an AI citation is real?
A citation should pass two separate checks: the source exists, and it actually supports the statement it accompanies. Rob Smith, CEO of Lionfish Tech Advisors, frames the core test as: “Does the cited source exist, and does it say what we claim it says?” He calls the sequence “The fix is order of operations.” The following checklist turns that advice into a practical review; it is a workflow, not a tested intervention or formal standard. Smith’s commentary in The AI Journal.
- Open the source. Follow the supplied URL rather than relying on the citation text or a search-result snippet. If the link is broken, search for the exact title and publisher; do not treat a plausible replacement as a verified match until you confirm it.
- Match the bibliographic details. Check the title, author, publisher, publication date, and locator, such as a page number or section. Record discrepancies, including paraphrased titles or an author attached to the wrong work.
- Find the supporting passage. Locate the text, table, or data that backs the claim. A real source is not sufficient if it does not say what the report attributes to it.
- Preserve the scope of numbers. Verify the population, geography, time period, and definition behind each figure. A number can be copied accurately yet mislead if the report changes what it measures or presents a limited estimate as a broad fact.
- Check descriptions of named organizations. Ask whether the organization would recognize the description, especially where the characterization is consequential. For high-impact claims, seek confirmation from the named organization when practical.
- Assign an accountable reviewer. Identify who opened and checked each source. A citation checker can flag suspicious references for review, but a person still needs to confirm that the source exists and supports the claim.
What do these cases establish—and what do they not?
The reported defects make a strong case for verifying citations before AI-assisted research is used in a consequential report. They do not prove that AI cannot help with research, establish a portfolio-wide error rate, or show what controls the firms later adopted. The investigations are evidence of what GPTZero said it found; they are not independent proof of every underlying citation or factual claim. The available findings also do not establish whether either report is currently available from its publisher.
GPTZero says it used its Hallucination Check tool during both investigations. A tool that flags references can assist triage, but it cannot substitute for opening the source and checking the claim against it. The essential safeguard is an accountable verification step between generating a research draft and publishing or relying on it.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
