PwC Middle East reports found with AI made up citations, researchers say

In short: Researchers say some PwC Middle East research-style reports included AI errors like made up citations and hard to verify claims.

What happened

Researchers at GPTZero investigated four “thought leadership” reports published by PwC Middle East over the past two years. These reports covered topics such as artificial intelligence, government services, and predictions about electric and self-driving vehicles.

GPTZero said the reports show signs of “AI hallucinations.” That is when an AI writing tool confidently makes up information, like a student inventing sources for a school essay. The Financial Times said it reviewed the findings.

According to GPTZero, many footnotes and citations in the reports were unreliable. Some links pointed to pages that did not support the claim, and some links did not work at all. GPTZero also said one academic paper on air quality in Riyadh appeared to be entirely fabricated, since there was no record of the study in the journal or from the named authors.

The investigation also flagged unusual sourcing. One report cited a teenage blogger on Medium as a source for a JPMorgan example. Another citation included a tracking tag in the web address that read “utm_source=chatgpt.com,” which can be a clue that the link was gathered with ChatGPT.

PwC Middle East told the Financial Times it takes accuracy seriously and is “updating a limited number of supporting citations” in the reports. It said it has quality control processes, but did not explain how the errors got into the published work.

Why it matters

These reports are meant to build trust and attract paying clients. When citations are wrong or missing, it becomes harder for readers to check what is true, especially as consulting firms also sell advice on how to use AI responsibly.

Source: Financial Times