How Much Time Does AI Actually Save You?

Two hours a day. That’s the time savings workers reported in the Adecco Group’s Global Workforce of the Future survey, and it’s the kind of number that ends up in keynote slides and tool marketing within weeks. A quarter of your working day, handed back by AI.

Researchers at the Federal Reserve Bank of St. Louis went looking for the same thing with a more careful method, asking workers about actual hours saved in their most recent work week. The answer came out to about 2.2 hours per week for generative AI users, roughly 5% of their working time. Per day, that’s a bit over 25 minutes.

So the popular claim and the measured result differ by roughly a factor of five. Before deciding AI productivity is a mirage, it’s worth understanding where the gap comes from, because both numbers are real answers to different questions.

Why the numbers disagree

Survey design does a lot of quiet work here. Ask “how much time does AI save you?” and people estimate their best days, the dramatic wins, the report that took one hour instead of four. Ask “how many hours did AI save you last week?” and you get the boring average, which includes the days they barely touched it.

Frequency is the other half. The Fed survey’s headline figure averages over everyone who used AI at all in the previous week, and intensity varies enormously. Heavy daily users genuinely do save multiples of what occasional users save. Gensler’s 2026 Global Workplace Survey found that about 30% of office workers now qualify as what it calls AI power users, and that group looks different on almost every measure, spending more of the week learning and collaborating rather than grinding alone.

In other words: two hours a day is what enthusiastic, frequent users feel on a good week. Twenty-five minutes a day is what average use actually measures. Neither is a lie. One is a ceiling, the other is a median.

The part that’s true in the hype

Twenty-five minutes a day is still a lot. Over a working year it’s roughly two full weeks. No other workplace tool of the past decade can credibly claim that for the average user, and the figure rises with skill and frequency, which means it’s partly under your control.

The honest framing is that AI time savings behave like compound interest, not like a lottery win. Small, recurring, and only impressive if you keep collecting them.

Where the saved time actually goes

There’s a second problem the vendor surveys skip entirely: saved time doesn’t store itself. Twenty-five recovered minutes scattered across a day default to email, Slack, and a slightly longer scroll. Unless you decide otherwise, AI converts drudgery into diffuse slack rather than into anything you could point to in a performance review.

The fix is unglamorous. Track which tasks AI now does faster for you, then deliberately reassign the recovered block to something visible: a skill you’re building, a project nobody has time for, deeper review of the AI’s own output. The people who turn AI use into career progress aren’t the ones saving the most minutes. They’re the ones who can name what the minutes bought.