UK workers lose 5.8 hours weekly botsitting, even as 90% now use AI

Almost all UK digital workers are now dealing with AI, but the productivity story isn’t landing the way boards hoped. In “The Work AI Index: UK 2026,” the Work AI Institute, a research arm of Glean Technologies, says workers waste an average of 5.8 hours each week “botsitting.” That is time spent overseeing outputs, reloading context the system should have remembered, restarting failed sessions, and reworking material until it is usable.

This matters because workers are not refusing AI. The report says 90 percent of the 1,500 people surveyed are now required to use AI in their roles, and 80 percent use multiple AI tools every week. On average, those workers report that AI automation saves them roughly 12 hours a week, or just under a third of their working week. But only 18 percent agree AI has significantly improved their organization’s performance. The missing link is that the “time saved” is not turning into more productive work, it is getting absorbed by the unglamorous human labour required to keep the tools running.

So what exactly is botsitting? The report describes it as the integration work employees have to do because the systems often do not reliably produce what they need the first time. For every hour a UK staffer spends getting output from AI tools, the report claims they spend roughly another hour making it usable. On the surface, that sounds like “AI needs QA,” but botsitting goes further than typical review. Employees frequently have to reload context into tools and then oversee the output by checking whether answers are wrong, incomplete, or missing important context.

When problems show up, the loop can get expensive. Workers may have to re-prompt, add more context, swap models, and re-prompt again until something acceptable arrives. And if employees are not diligent enough to spot errors, the consequences can spread. The report notes that mistakes can land on colleagues who were not involved with the work, but now have to fix something they did not break. This is where the productivity dividend gets quietly converted into coordination overhead, which is harder to measure than hours spent writing.

The report also points to failure rates as a major driver of the lost time. It says more than a third of AI sessions fail outright, at 36 percent, requiring a full restart or substantial reworking. A lot of that “redo” labor is described as grunt work: reloading context into different tools, catching hallucinations, and verifying outputs that may look fine at first glance. Then there is a behavioral component. Workers can get burned out by the constant need to supervise the system, and the report warns they may cut corners over time, becoming less diligent in checking outputs, verifying sources, or judging whether recommendations make sense.

The report’s findings get even more consequential when you zoom out to what this implies about deeper workplace adoption. The UK, it says, has moved fast on AI uptake, even leading the US on key adoption metrics. But the Work AI Institute’s framing is that the country is going beyond “content generation” into activities that shape working life. It warns AI is now being used in higher-stakes, tightly regulated areas like HR decisions. More than half of UK workers say they are comfortable with AI playing a role in performance evaluation, and nearly 40 percent say it is already used in reviews.

Comfort and actual usage still diverge by decision type, and regulation is part of the reason. The report says British workers are more comfortable than Americans with AI in hiring, promotion, compensation, and even termination decisions. Yet local organizations are less likely to use AI in termination decisions because employment law makes dismissal harder to defend than in the US. Put differently: even if an AI system can generate outputs quickly, UK governance and legal defensibility increase the cost of getting it wrong, which in turn raises the stakes for how much human verification is required.

The board-level takeaway is blunt. The report concludes that Britain has built a stronger institutional foundation for workplace AI than almost any other country, which could be an advantage. But it also says the value of the investment will come from operational discipline, and measuring whether the work produced is better, not just faster. Without that discipline, the hours workers “save” get lost again in botsitting, effectively turning “AI adoption” into “AI overhead.” As Dr Rebecca Hinds, head of the Work AI Institute at Glean, put it: “Adoption alone doesn't equal transformation.” If employees are spending the productivity dividend on botsitting, companies have not eliminated work, they have created a new layer of overhead.