Survey Exposes 6.4-Hour Weekly 'Botsitting' Burden

A survey of over 1,000 full-time workers across the United States, the United Kingdom and Australia has put a name to one of AI’s most persistent drags: “botsitting.” Workers report devoting an average of 6.4 hours a week—almost a full workday—to the invisible labor of making AI tools usable. The data, published in Harvard Business Review’s August 10 newsletter and drawn from research by Rebecca Hinds of Glean and Paul Leonardi of the University of California, Santa Barbara, shows that while most respondents say AI makes them individually more productive, only 13% believe their company’s overall performance has improved.

The botsitting tasks break down into three main activities. Feeding the AI sufficient context eats up 2.3 hours a week; identifying and fixing errors, biases or unexpected behaviors takes another 2.2 hours; and reviewing outputs to confirm they are fit for purpose accounts for 1.7 hours. In total, 37% of the time employees spend with AI is consumed by this upkeep—slightly more than the 36% they spend actually using the technology to produce work.

The newsletter also highlights separate pieces on indirect influence and CEO-board dynamics, but the botsitting research stands out for quantifying a cost that often goes unmeasured. The authors warn that the burden is “not just time” but also the mental wear of constantly checking, correcting and stabilizing a system before its output can be trusted, creating a hidden layer of cognitive load that erodes the very productivity gains AI promises.

Why AI’s Productivity Promise Often Falters

The disconnect between individual productivity and firm-level performance signals that organizations are underestimating the wraparound effort AI demands. The 6.4-hour weekly time sink suggests that many deployments are being treated as plug-and-play, when in reality they require continuous human oversight to stay aligned with business needs. This oversight—checking for errors, adding context, reworking outputs—is rarely tracked or rewarded, yet it quietly inflates the total cost of AI ownership.

Advertisement

The researchers point to three common miscalculations by leaders that amplify botsitting. First, an overestimation of an AI system’s out-of-the-box readiness leads to insufficient upfront training and integration time. Second, the assumption that once deployed, AI tools will self-improve, which ignores the need for ongoing human feedback loops. Third, a failure to formally recognize botsitting work, leaving it invisible to performance metrics and therefore unmanaged. These blind spots help explain why individual users feel more productive—they are shouldering the undocumented cleanup—while the organization fails to capture net gains.

The findings align with the newsletter’s parallel discussion on indirect influence: the ability to mobilize people without formal authority. Leaders hoping to embed AI effectively may find that persuasion, timing and back-channeling are more powerful than mandates, especially when adoption depends on employees voluntarily taking on the hidden labor of botsitting.

Three Levers to Curb Botsitting and Boost AI ROI

  • Factor ‘botsitting time’ into AI ROI models. The 6.4-hour weekly average—equivalent to roughly 16% of a full-time schedule—should be treated as a recurring operational cost, not a one-off training expense. Budget for it in staffing plans and vendor contracts.
  • Design feedback loops into the AI system from day one. Build tools that allow users to flag errors quickly and that feed corrections back into the model, reducing the time workers spend repeatedly fixing the same issues. This shifts some botsitting from manual labor to a structured process.
  • Recognize and reward botsitting work. Make the invisible visible by tracking the effort employees put into AI maintenance, and include it in performance reviews. Small acknowledgments can prevent burnout and signal that the organization values the work that makes AI truly productive.