“AI will save teachers time.” Somebody still has to check what the AI produces, and that time can disappear from a workload estimate.
There’s a word for it: botsitting.
What the survey found
Founder Reports surveyed 2,052 employed adults in the US in October 2026. All used AI at work at least once a week.
Of the 1,856 who’d trained an entry-level colleague, 60% said getting good work from AI took at least as much effort as training a new employee. About one in three said it took more. The other 40% said it took less.
Across respondents, 31.6% reported spending three or more hours a week on AI oversight. Just over 8% reported more than ten hours.
Among managers and above, 40.7% reported spending at least three hours a week overseeing AI and correcting errors. Among individual contributors, the figure was 25%.
These are estimates from a US survey run by a business publication. They tell you what those workers reported. You’ll need local evidence to say how much time checking costs a school.
Other research on checking time
Glean’s Work AI Index puts the average at 6.4 hours a week for the work its respondents do to make AI useful. Glean’s measure covers more than correcting a finished draft. Don’t lift that figure into a school business case as an estimate of teachers’ checking time.
What this looks like in a school
Picture a head of department who uses Copilot to draft a scheme of work. Producing the draft might take ten minutes. Checking it against the specification and correcting it for the pupils who’ll use it might take an hour.
In that example the task took seventy minutes. Whether that’s a saving depends on how long the department would usually spend producing work of the same quality.
Now imagine the head of department also checks AI-drafted resources from three colleagues. The people producing the drafts could each report a saving, whilst the checking lands in somebody else’s timetable.
If a trust says AI will save each teacher two hours a week, does that figure include the checking? Who’s doing it?
What to do with this
Count the checking time. In a pilot, ask staff to log how long the draft takes and how long they spend getting it ready to use. Then compare the whole task with their usual approach.
Watch where the work lands. Give the colleague who checks the output somewhere to record that time, even if they didn’t make the original request.
Treat checking as a skill. A trainee teacher or apprentice can take a plausible answer, compare it with the source material and explain which parts they’d change. An experienced colleague can then check their reasoning.
For anyone working through the AI Leadership units, pick one task where AI is supposed to save time. Once the person responsible for the final version has finished checking it, does the saving still hold?