A Crisis of Confidence
The central promise of generative AI in the workplace is efficiency. But a disconnect is growing between this promise and the reality of its day-to-day use. While adoption is high, with some studies showing more than 60% of workers now use AI, trust has
not kept pace. The headline figure—that a significant number of employees feel compelled to double-check AI's work—points to a deeper issue. A Slingshot report found that while managers deploy AI for research and analysis, nearly two-thirds of employees use it primarily to double-check their own work. This isn't a sign of defiance, but of caution. Employees are discovering that AI can be confidently wrong, producing plausible-sounding but inaccurate information, a phenomenon often called 'hallucinations'. This verification behaviour is widespread, with one survey finding 77% of workers review a colleague's work more carefully if they know AI was involved.
The Productivity Paradox
This constant need for verification creates a productivity paradox. A tool meant to save time is costing time in other ways. One survey found that workers using AI reported saving about six hours a week, but they also spent about four hours a week correcting its output. Another study from Adaptavist calls this the 'verification tax', noting that 42% of workers spend more time checking AI output than the time it saves. The issue is compounded by what some are calling 'workslop'—AI-generated content that looks polished but is substantively weak, requiring significant human intervention to fix. This not only drags down project timelines but also erodes trust between colleagues when low-quality, AI-assisted work is passed along. One study showed 45% of workers have had to fix or redo a coworker's output because they believed it relied too heavily on AI.
Why The Distrust is Justified
The skepticism is not unfounded. Employees across various sectors report being forced to correct AI-generated mistakes. A major KPMG study found that 56% of employees admitted to making errors in their work due to AI. The reasons for this are varied. AI models are only as good as the data they are trained on, which can be outdated, biased, or incomplete. They lack real-world context and the nuanced judgment that comes from experience. This is why many employees are wary of its use in critical functions. A survey by SHL found that 58% of workers do not want AI used to evaluate their performance, and a similar number don't trust it for hiring decisions. The fear is that AI will automate judgment without understanding context, leading to flawed and unfair outcomes.
Building a Culture of Critical Use
The solution isn't to abandon AI, but to get smarter about how it's used. The current verification trend should be seen not as a failure, but as a necessary phase in workforce education. Companies bear a significant responsibility here. Despite rapid adoption by employees, many organisations have failed to provide adequate guidance. A recent study by idealis found that while 62% of workers use AI, only 40% report their company has clear usage guidelines. This leaves employees to figure it out on their own, often using personal or unauthorised tools. Effective training needs to go beyond basic prompting skills. It must teach employees how to critically evaluate AI output, identify potential biases, and understand the tool's limitations. Framing AI as a co-pilot, rather than an autopilot, is key. It's a powerful assistant for drafting and brainstorming, but the human user must remain the ultimate editor and arbiter of quality.
