GPTZero says it found 31 citations that it classifies as hallucinated across three papers linked to Accenture, but the consultancy's documented response addresses one report only. Accenture attributed errors in that agriculture report to human error and said it found no evidence that AI caused them.
Three Papers, but Only One Was Co-Published by Accenture
GPTZero's investigation examined a 2025 agriculture policy paper co-produced by Accenture's African arm and IE University, plus two 2026 research papers posted on arXiv by Accenture researchers.
It counted 15 flagged citations among 89 in the policy paper, seven among 102 in ContextEcho, and nine among 54 in Authenticity Debt.
GPTZero acknowledges that the two arXiv papers were not official Accenture publications. It says they were written by Accenture researchers, using company affiliations, and describes its findings across all three as evidence of 'hallucinated' citations.
The consultancy's response reported by the Financial Times, however, concerns the agriculture paper; the available coverage does not establish that Accenture investigated or addressed the other two papers in the same way.
The label needs context. GPTZero's stated definition covers citations that do not match any real source or combine details from different sources. Its examples include apparently nonexistent sources, but also real publications with incorrect titles, dates, authors or other bibliographic details.
The 31 should therefore be described as citations GPTZero flagged under its definition, not as 31 proven nonexistent studies.
The Report's Errors Do Not Establish Their Cause
The Financial Times examined examples in the agriculture paper, reporting factual errors and citation problems, including claims it said were unsupported by their cited sources. One example said predictive AI had improved detection of cocoa swollen shoot disease in Colombia and Ecuador to 99.1% accuracy.
GPTZero said the cited study concerned African cocoa cultivation, not Latin America.
Particularly in the context of (appropriate) crackdowns on LLM-generated papers (e.g., on arXiv) and more and more tales of fabricated citations, two reminders: (1) LLMs do not "cite their sources" (2) Using an LLM to generate, complete, or fix LaTeX code can result in fabricated citations
— Dr. Casey Fiesler (@cfiesler.bsky.social) 2026-05-17T14:59:41.407Z
Accenture told the Financial Times that its review found 'human error' and 'no evidence' of AI hallucinations. It said the report would be corrected and republished, but declined to explain how it reached that conclusion. IE University said its authors were reviewing the sources and data, and would correct confirmed errors and strengthen safeguards.
Almost surely. We are already seeing this on the ACM CCS PC: minor bib errors get flagged as potential hallucinations when the cause is entirely unclear. Pretending that non-LLM based citation generators like Google Scholar are infallible is absurd when we all know that data errors happen
— Toby Murray (@tobycmurray.bsky.social) 2026-05-17T00:44:04.660Z
GPTZero says it manually checked the citations it flagged, but also acknowledges that its detection tools do not definitively prove that a language model generated the papers or citations. Citation discrepancies and evidence of AI authorship are separate claims; the available reporting does not settle the latter.
