There’s a particular kind of irony in a consultancy that sells advice on using artificial intelligence responsibly, then publishes its own reports that appear to have leaned on that same technology a little too eagerly. That’s the position PwC found itself in after the Financial Times reported that four of its Middle East “thought leadership” documents were riddled with problems.
The reports, covering artificial intelligence and electric vehicles, carried the tell-tale fingerprints of generative AI gone unchecked: fake footnotes and citations pointing to sources that don’t hold up under scrutiny. In other words, the kind of confidently fabricated detail that large language models are notorious for producing when nobody bothers to verify their output.
PwC is not alone here, and that’s the uncomfortable part. The world’s largest consulting firms have spent the past few years positioning themselves as the trusted guides for corporations nervous about adopting AI without getting burned. The pitch is compelling: hire us, and we’ll help you deploy these tools safely, avoid the pitfalls, and steer clear of embarrassing mistakes. The problem is that their own published material keeps stepping on exactly the rakes they warn clients about.
Why this keeps happening
Generative AI is extraordinarily good at producing text that looks authoritative. It will happily generate a footnote, complete with author, publication and year, that reads like a legitimate academic reference. The catch is that the reference may not exist at all. These are the so-called hallucinations, and they are stubbornly difficult to eliminate because the model isn’t retrieving facts — it’s predicting plausible-sounding sequences of words.
For a consumer glancing at a report, a citation is a signal of credibility. For a machine, it’s just another string of text to be produced convincingly. When a document meant to demonstrate expertise turns out to rest on invented sources, the damage isn’t just factual — it’s reputational.
The bigger lesson
The episode is a neat case study in why human review still matters. The tools themselves aren’t the villain; they’re powerful drafting assistants that can accelerate research and writing enormously. But they require oversight, fact-checking and a healthy dose of skepticism — precisely the disciplined workflow that firms like PwC advise their clients to build.
There’s a practical takeaway for anyone using these tools, whether you’re drafting a report, a research paper or a product review:
- Verify every citation. If a source is quoted, confirm it exists.
- Treat AI output as a first draft, not a finished product.
- Assume confidence is not accuracy. The more assured the text sounds, the more it deserves a second look.
The consultants preaching responsible AI adoption have just handed everyone a memorable reminder: the technology is only as reliable as the people checking its work.
