PwC’s AI Report Failed the Reading Test

PwC published four reports between 2024 and 2026 that were intended to attract consulting work in the Middle East and contained fabricated claims, mismatched citations and confused drafting. According to reporting summarized by Futurism, researchers at the AI-detection company GPTZero found passages that appeared to have been generated by artificial intelligence and not adequately checked before publication.

One 2025 report described a government framework called “Citizen Pulse” as though it were in use around the world. GPTZero found that the framework did not exist. Other citations either contradicted the claims they supposedly supported or led to sources too weak for the institutional weight PwC placed on them. A teenage Medium blogger was cited in connection with a JPMorgan AI initiative. A familiar traffic-accident statistic appeared three times within two pages, each time with a different source.

The obvious joke is that AI hallucinated in a report about AI. The useful finding is less charming. AI did not approve the copy, apply PwC’s branding or release the document to prospective clients. People and procedures did that. The machine produced uncertain text; the quality-control system converted it into corporate authority.

Consultancies sell judgment under conditions where clients lack time, staff or specialist knowledge. Automating research can reduce production costs, but it also creates a temptation to preserve the price and appearance of expert work while removing part of the labor that made the work dependable. Verification then slides downhill. A junior employee, client or reader must open the links and discover whether the polished sentence has a factual floor.

Letterhead is not a fact-checking tool. If PwC regards these reports as evidence of AI-enabled efficiency, the firm should make that efficiency testable: identify a named reviewer, retain a source ledger, require direct checks of every citation and record accountable sign-off before publication. High-risk claims should receive subject-matter review rather than a final pass for tone and formatting.

Those controls cost time. That is the point. PwC can either include verification in the service it sells or admit that its clients are being asked to perform it after delivery. The next meaningful response is not another statement about responsible AI. It is a review trail showing who read the next report, what they checked and whose name sits beside the release decision.