Before AI claims time saved, teachers reported 46-hour weeks

The classroom note

Any pitch that an AI tool will give teachers 'their time back' implies a starting number: time back from what baseline? The OECD's Teaching and Learning International Survey supplies one. Its 2018 round, covering lower-secondary teachers, produced a US-specific breakdown published by the National Center for Education Statistics in a First Look report: 'U.S. lower secondary teachers spent an average of 28 hours teaching and 46 hours working in total, per week,' meaning roughly 18 hours, well over a third of the reported working week, went to tasks other than direct instruction.

What the evidence says

TALIS is a survey, not an experiment: teachers' own self-reported estimates of how they spend time, gathered from a nationally representative sample, not a time-and-motion study with observed logs, and the report compares US figures to a TALIS average across other participating systems rather than a single fixed standard. It also situates the US within a range: teachers in Chile (28 hours) and Alberta-Canada (27 hours) taught comparably more, while Japan (56 hours) and Kazakhstan (49 hours) reported higher total working hours. A second source, the OECD's TALIS programme page, adds that the most recent round, TALIS 2024, expanded the sample to 'about 280 000 teachers in 17 000 schools across 55 education systems,' with initial results released in 2025, though this source does not itself report updated hours figures.

The implementation question

The mechanism a workload claim needs to specify is which slice of the roughly 18 non-teaching hours a tool addresses: marking, lesson preparation, administrative reporting and parent communication are different tasks with different automatable content, and TALIS's category of 'working hours' does not break down into a line item labelled 'time an AI tool could replace.' A claim that a tool saves teachers 'hours a week' should be read against this baseline and asked which of the 46 hours, measured how, it refers to.

What holds and what fails

What holds is that TALIS is the most methodologically documented, internationally comparable baseline available for teacher time use, with a clear sample and a stated survey design, which makes it a reasonable yardstick against which to test a saved-time claim. What fails is treating any single saved-time figure as validated against this baseline unless the claim's own source states its sample, measure and period as precisely as TALIS does; a rounder or larger number with no comparable methodology is not more credible for being bigger. This is an editorial standard, not one TALIS itself states.

• Does the vendor's saved-time claim specify which of a teacher's roughly 18 non-teaching hours it targets?
• Is the claim based on self-report, like TALIS, or on measured logs, and does it say which?
• Would a comparable TALIS-style survey in our own system show a similar 28-and-46-hour split today?

A saved-time claim is only as meaningful as the baseline it is measured against, and TALIS is one of the few baselines documented well enough to check a claim against.