
A study of 6,000 knowledge workers puts botsitting at 6.4 hours a week, almost a full working day.
This supervision time reduces the gains from AI and weighs on motivation, with a higher risk of employees leaving.
In Luxembourg, no study yet measures the phenomenon, but the rapid adoption of AI and the frequent lack of internal policy create fertile ground.
AI was supposed to free up employees' time. In practice, part of that time is taken back by a task nobody has really written into a job description: checking, correcting and redirecting what AI tools produce. Anglo-Saxon researchers have given it a name, botsitting, which means "babysitting the robot".
The first figures, drawn from American, British and Australian studies, suggest that the phenomenon is far from marginal. This invisible work raises questions about how companies, including in Luxembourg, really measure the benefits of AI.
The term was popularized by Glean, an American provider of AI tools for businesses, following a survey of 6,000 knowledge workers in Australia, the United States and the United Kingdom. According to the summary published in September 2026 by the Australian association AHRI, respondents spend 6.4 hours a week supervising and reviewing the work of AI. This botsitting accounts for 37% of the total time spent with these tools, more than the 36% spent using them to produce work.

Employees estimate they save around 11 hours a week thanks to AI, but botsitting absorbs a large share of it: the net gain falls to 4.6 hours, as detailed in TechTarget's analysis of the Glean survey. The same report links frequent botsitting to a 73% higher risk of leaving.
These figures should nonetheless be taken with caution: they come from a provider that itself sells AI solutions, and they rely on employees' own statements rather than field measurements.
For Rebecca Hinds, head of the Work AI Institute at Glean and co-author of a Harvard Business Review article on the subject, the main driver of botsitting is simple: AI is given a complex task without the context it needs. Connecting all of one's documents to a tool is not always enough, she adds, since it may draw only a very limited understanding from them.
"It's monotonous, it's tedious and it doesn't give them meaning or purpose."
The consequences can also be seen in morale. Rebecca Hinds told AHRI about work that, once it becomes a daily routine, pulls employees away from higher-value tasks: "It's monotonous, it's tedious and it doesn't give them meaning or purpose." She also notes that reinvesting the time saved in more tasks of the same kind often contributes to burnout.
To limit botsitting, she recommends well-documented processes and a culture of experimentation, so that everyone knows what to do when an AI session fails. Botsitting has a cousin, known as "workslop": AI-generated content that looks polished but lacks substance. In a Stanford and BetterUp study reported by Fast Company, 40% of the 1,150 employees surveyed said they had received some from colleagues in the previous month, and about half consider those who send it less reliable.

To our knowledge, no study measures botsitting in Luxembourg. The finding is broader: Sidonie Paris, senior economist at the Chamber of Commerce, said in Paperjam in late 2025 that there was not yet a general study on the real adoption of AI across the entire Luxembourg economy, apart from sectoral surveys, notably in finance.
The conditions for the phenomenon to emerge nonetheless seem to be in place. According to STATEC, nine out of ten workers in Luxembourg are estimated to be exposed to AI, with 55% seeing their job augmented by the technology. On the business side, the 2026 survey by FEDIL, Luxembourg AI Factory and Luxinnovation of 136 professionals, relayed by Paperjam, shows that nearly six in ten companies have deployed generative AI and that 88% cite productivity gains as the main driver, while fewer than one in two has a formal AI policy.
The question of real gains is at the heart of the local debate. "We always say that productivity is stagnating in Luxembourg," says Carlo Thelen to the media, Director General of the Chamber of Commerce. He adds that AI can boost productivity if properly applied, but that the gain is not automatic, including in banking, where it requires heavy investment and training.
Botsitting is a reminder that an AI tool does not eliminate work: it shifts part of it to checking, correcting and supplying context. For employees, the challenge is to have this burden recognized in their objectives rather than bearing it in silence. For HR departments, it means no longer assessing AI solely by hours saved, but by the time genuinely freed up for higher-value tasks.
For Luxembourg companies, where internal rules evolve more slowly than usage, counting supervision hours, training teams and planning what to do when AI fails would be useful first steps. It remains to be seen, backed by local data, whether botsitting has already begun in the Grand Duchy.
What is botsitting?
Botsitting refers to the time spent supervising, correcting and providing context for an AI's work to make it usable. The term comes from Glean's research and a Harvard Business Review article.
How much time do employees spend on it?
A study of 6,000 knowledge workers estimates it at 6.4 hours a week. This supervision reduces the net gain from AI to 4.6 hours a week.
Does botsitting exist in Luxembourg?
No local study measures it to date. According to STATEC, nine out of ten workers are exposed to AI, and FEDIL notes rapid deployment of generative AI.
Cover image: Glean