PwC is the most recent Big Four company to be exposed for disseminating poor-quality AI content.

      The largest consulting firms worldwide are providing costly advice to companies on how to implement AI safely, yet their own reports continue to contain significant issues. PwC has recently come under scrutiny, as four of its "thought leadership" reports for the Middle East were found to be flawed, according to the Financial Times. These reports, focusing on AI and electric vehicles, included fake citations, incorrect attributions, and fabricated sources. The research organization GPTZero identified these inaccuracies as AI hallucinations, which the FT later confirmed.

      One obvious indicator was a footnote that concluded with ‘chatgpt.com’. The flaws were glaring. For instance, one report claimed a PwC survey revealed that 70% of chief executives in the Middle East anticipate generative AI will transform their businesses. However, the footnote pointed to a news article that did not mention the survey at all. Another footnote had a URL that included the tag “utm_source=chatgpt.com.” Paul Esau from GPTZero commented to the FT that the disorganized citation of sources illustrates the problems associated with AI-generated research, even though the firm promotes itself as a responsible AI advisor, particularly on avoiding these kinds of mistakes.

      PwC is not the only firm facing these issues; it adds to an unwanted trend. Earlier work from GPTZero led KPMG to retract an October report that falsely claimed AI was utilized at UBS, the UK’s NHS, and Transport for London. Additionally, EY had to withdraw a study the previous month due to incorrect citations, and Deloitte has twice introduced AI errors into government reports, including a $1.6 million healthcare initiative in Canada that referred to non-existent studies.

      The established pattern is consistent: the Big Four have released hundreds of AI thought-leadership articles to attract clients, while simultaneously encouraging their employees to leverage AI for increased efficiency. This has resulted in a collection of poorly produced content branded with reputable logos.

      Coincidentally, during the same week that PwC’s US CEO Paul Griggs was addressing Business Insider about common pitfalls in AI adoption, he identified one significant mistake: attaching AI to flawed processes. “All AI is going to do for you is highlight how problematic your disorganized process truly is,” Griggs stated, seemingly validating his own firm’s errors.

      What’s particularly troubling is the implication for the quality of the output. These firms demand high fees for their expertise and thoroughness, yet it seems no one verified the legitimacy of the sources cited. This reflects a similar vulnerability to those seen in AI-generated code that is released with security vulnerabilities: the results may appear correct at first glance, leading individuals not to scrutinize them further.

      Furthermore, this contradicts their pitch to clients. Consultancies are among the most vocal advocates asserting that AI will revolutionize their operations. Ironically, their own reports demonstrate how AI can subtly undermine quality and how the excitement surrounding it often surpasses safe practices. The identified inaccuracies serve as a red flag, and a detector consistently uncovers them, yet these firms persist in publishing.