Glean’s first Work AI Index put numbers on a job nobody applied for: white-collar workers spend an average of 6.4 hours a week “botsitting,” feeding AI context and cleaning up its output, about a third of their total AI time. And 69% admit to “botshitting”: shipping AI work they have not verified. The productivity math only reconciles by counting the invisible human labor.
The report, released by Glean’s Work AI Institute on June 10, 2026 with academic coauthors including researchers from Stanford, UC Berkeley, and Notre Dame, surveyed 6,000 full-time digital workers in the US, UK, and Australia between December 2025 and January 2026. The headline looks like success: 87% of digital workers now use AI at work and 75% say it makes them more productive, and workers credit AI automation with saving them roughly 11 hours a week, more than a quarter of the workweek.
Then comes the contradiction: only 13% say their organization’s performance has significantly improved because of AI. Business Insider, which covered the report and its “botsitters” the following day, called it the growing disconnect between individual productivity and companywide results. The report’s own explanation is that the missing productivity is being consumed by work employees never applied to do, most of it invisible to the metrics a manager sees.
What Botsitting Actually Is
The report’s authors coined botsitting for the unrecognized work required to make AI useful: supplying context a tool should already have, checking and correcting outputs, debugging mistakes, reformatting results, and stitching together disconnected tools. Workers spend 6.4 hours a week on it, or 37% of their AI time. Actually using AI to produce work takes 36%. A machine that was supposed to delete busywork created a new maintenance job, and, per the report, workers say AI already automates about 27% of their work output.
Business Insider, which covered the report and its “botsitters” the following day, called it the growing disconnect between individual productivity and companywide results.
AI & TechnologyThe Silent Productivity Paradox: How AI Echo Chambers Are Distorting Decision-Making
A 2026 BCG survey of 12,000 frontline workers found that 42% saved a full workday weekly via AI, yet 50% of leaders failed to articulate…
Read More →The report’s coauthor Rebecca Hinds, who heads the Work AI Institute at Glean, described the dynamic in blunt terms on the Cognitive Revolution podcast: the work is “not rewarded and it’s not appreciated or tracked or measured and certainly not incentivized within the organization.” The exit risk is measurable: workers who spend an unusually large share of their AI time on botsitting are 73% more likely to be actively job hunting.
Botshitting: The Quality Problem Nobody Budgeted For
The report names the companion behavior botshitting: shipping AI-generated work you have not verified. It is crude language for a precise finding. Among the surveyed workers, 69% of AI users admit to it in some form, including 41% who say they sometimes deliver work they could not explain if asked and 28% who have blamed AI for mistakes they themselves caused. The quality-control burden is real, and it lands on whoever is fastest to paste output into the shared drive.
Tool sprawl makes it worse. 77% of AI users juggle multiple tools each week, 33% use four or more, and 60% rerun the same prompt across different tools because the first result was not good enough, effectively making each employee their own systems-integration department. One caution before the numbers get repurposed: Glean, the firm behind the report, sells enterprise AI tools whose pitch is fixing exactly this context problem, so treat its findings as well-sourced but commercially motivated.
Why Companies Keep Abandoning AI Anyway
The botsitting data helps explain a trend that predates it. S&P Global Market Intelligence’s surveys found the share of companies scrapping most of their AI initiatives rose to 42% in 2025, up from 17% the year before. And the workload math runs in the wrong direction for adoption stories: a 2024 Upwork Research Institute study found 77% of employees say AI has increased their workload, even as their C-suite invests in it. Both findings sit inside The AI Productivity Paradox, an executive summary Seramount published in March 2026 advising HR departments to recognize that individual time savings do not reach the balance sheet automatically.
Seramount’s paper outlines four principles reshaping performance in AI-enabled organizations, worth reading as a whole because they compound into why adoption numbers flatter the intuition:
- AI is generative, not deterministic: it constructs responses, so quality depends on human framing and evaluation.
- AI collapses the cost of production, not the cost of judgment: output scales quickly while evaluation stays scarce.
- AI accelerates systems faster than humans can adapt: tools spread before governance and norms catch up.
- AI amplifies human differences: gaps widen between workers who can critically assess AI and those who cannot.
More Engaged, Less Productive
AI & TechnologyAI Is Forcing HR to Abandon Loyalty Metrics—And Retention Strategies Are Changing Forever
A Springer 2026 study found that 42% of employees distrust AI sentiment analysis, fearing it’s used for surveillance rather than support.
Read More →A separate dataset complicates the picture further. ADP Research’s People at Work 2026 report, drawn from more than 39,000 workers across 36 markets, found that daily AI users are healthier on almost every workplace measure: 30% of them are fully engaged at work compared with 14% of non-users, and only 11% report negative work stress versus 23% of non-users. Yet daily AI users are also four times more likely to say they feel less productive than they could be. That is not a contradiction; it is what happens when automation removes the routine, visibly-checkable work and leaves the judgment-heavy work whose “done” is ambiguous.
And the workload math runs in the wrong direction for adoption stories: a 2024 Upwork Research Institute study found 77% of employees say AI has increased their workload, even as their C-suite invests in it.
What the Evidence Does and Does Not Show
All of the load-bearing numbers above are self-reported survey findings, not measured output. Glean’s 11 hours of weekly savings is what workers believe they save; ADP’s underproductivity is what workers feel. None of it licenses the older headlines that AI “does nothing” or that it “transforms” everything. The well-supported middle is narrower and more useful: individual adoption is nearly universal, individual perception of benefit is positive, and the link to organization-level performance is where the data goes dark. For anyone using AI daily, the practical read is to treat unverified output as unfinished work, and to keep receipts for the hours spent babysitting it, because that is work your performance review cannot currently see.
Watch two things next. First, whether Glean’s institute publishes follow-ups that split botsitting output by job level, since the current index flattens junior analysts and senior managers into one number. Second, watch whether HR platforms begin tracking AI oversight time in time-management systems, because the moment botsitting becomes a tracked metric, it stops being free labor; until then, it is the most common unpaid assignment on the modern desk.
