There’s an irony here I can’t unsee: the very consultancies selling companies on confident AI adoption keep tripping over their own AI use. This time it’s PwC’s turn.
What GPTZero found
AI-detection firm GPTZero examined four ‘thought leadership’ reports PwC Middle East published over the past two years — the kind of studies meant to drum up consulting work. The findings were verified by the Financial Times.
For a 2025 report titled ‘Transforming Governance’, GPTZero put the probability that it was entirely AI-generated at 84 percent — rising to 100 percent once you exclude the reference section. And that’s exactly where the trouble was: footnotes pointed to web pages that didn’t actually contain the claimed evidence. Other links led nowhere. And one cited academic paper on air quality in Riyadh appears to have been invented by the AI outright.
Not a one-off among the Big Four
The pattern isn’t new. Earlier GPTZero investigations had already pushed EY and KPMG to retract reports. EY pulled a study on loyalty rewards programs in May after it turned out to contain hallucinated footnotes. Add Deloitte, and all four big consulting firms have now been caught. First KPMG, then EY — now PwC.
My take
It’s easy to laugh at the consultants. But the real lesson lands on all of us who work with AI: the model invents sources so convincingly that they only fall apart when you click through. That’s exactly where the danger sits. A hallucinated citation looks like a real one — clean formatting, plausible title, a serious-looking URL. It’s just that there’s nothing at the end of that URL.
So I’ve turned it into a hard rule: every source an AI hands me, I click before I pass it on. Every time. Not because AI is useless — it’s fantastic for a first draft. But because the last step, the checking, can’t be delegated. Skip it, and sooner or later your name is on the cover of a fabricated study. And that costs a lot more than three minutes of fact-checking.
Sources: