British workers are spending the equivalent of a full working day each week looking after their AI tools.
A report from the Work AI Institute, the research arm of Glean, surveyed 1,500 UK digital workers and found they lose 5.8 hours a week to what the researchers call botsitting, the unrecognised labour of making AI output actually usable.
The report, built on a survey of 6,000 digital workers across the US, UK and Australia found that workers say AI saves them 11 hours a week. Yet only 13 percent of them say their organisation performs significantly better because of it.
Workers now spend 6.4 hours a week on botsitting, the institute's term for the untracked labour of making AI usable: feeding tools context they should already hold, supervising plausible but wrong outputs, debugging failures and cleaning up after them.
That activity now absorbs 37 percent of all AI time, slightly more than workers spend using AI to produce anything. More than a third of sessions fail outright.
And... when checking work becomes exhausting and goes unrewarded, people stop checking.
The report calls this botshitting.
Some 69 percent of AI users admit to producing output they have left unverified or could not defend if asked. Frequent botsitters are 73 percent more likely to be actively job hunting.
Why is this?
Tool sprawl forces most workers to juggle several AI systems weekly, with each switch burning context and time. And more than half of workers say information critical to their jobs sits beyond their AI tools' reach, forcing hours of manual context-feeding that the data links directly to burnout.
The strongest performers spend more time verifying AI output, refuse to use it on unsuitable tasks and work inside organisations where systems can reach the context that matters.
Those context-rich firms see workers 64 percent less likely to feel worn out and far less likely to produce indefensible work.
Until the hidden labour appears in the baseline, claimed savings will keep evaporating, and the answer lies in redesigned workflows and shared context rather than another procurement round.
