Your AI Saves 11 Hours a Week. You Spend 6 of Them Fixing Its Mistakes.

87% of workers now use AI for their jobs. They report saving 11 hours a week. Sounds like a revolution — until you learn that 6.4 of those hours go right back into fixing what the AI produced. The net gain isn't 11 hours. It's 4.6. And 69% of workers admit they ship AI output to colleagues and customers without checking it first.

Welcome to the era of botsitting.

The Problem: AI Created a Second Job Nobody Hired You For

Glean surveyed 6,000 knowledge workers across industries. The numbers are consistent and damning.

Workers spend an average of 6.4 hours per week reviewing, correcting, and reworking AI-generated output. That's not a rounding error — that's more than a full workday consumed by quality assurance on work you didn't do.

Think about what that means practically. You ask an AI to draft a client proposal. It produces something that looks right — proper structure, confident tone, plausible numbers. But the numbers are slightly off. The client name is misspelled in one section. The timeline references a project that ended last quarter.

So you fix it. Line by line. For 45 minutes. Time you could have spent writing the proposal yourself.

The researchers gave this a name: botsitting. Like babysitting, but for software that was supposed to make your life easier.

Frustrated office worker reviewing AI-generated documents on laptop

Here's the part that should terrify any team lead: 69% of workers report shipping AI output without verifying it. Not because they're lazy — because they don't have time to check everything. The botsitting burden is so high that the rational response is to skip the QA step entirely.

That's not a productivity gain. That's a quality time bomb.

The Solution: Treat AI Output Like a First Draft, Not a Final Product

The fix isn't to stop using AI. It's to stop trusting AI output by default.

Botsitting time comes from three root causes:

No output validation layer. Workers manually eyeball AI output instead of having structured checks — spell-checkers, fact-checkers, number-verification tools — run automatically.
Wrong task delegation. AI is great at first drafts, summaries, and pattern matching. It's terrible at precision tasks — exact numbers, specific names, current data. Most workers don't know which is which.
No feedback loop. When AI produces a wrong answer and a human fixes it, the fix disappears. The same mistake shows up next time. Workers are correcting the same categories of errors repeatedly.

The organizations getting real value from AI are the ones that have built human-in-the-loop workflows — not as a checkbox, but as a designed system. AI generates. A validation layer catches obvious errors. A human does final review. The corrections feed back into the system.

That cuts botsitting from 6 hours to roughly 1. The math works because you're automating the checking, not just the generating.

The Benchmarks: What the Data Actually Shows

87% of knowledge workers use AI tools for work (Glean, 2026)
11 hours/week average self-reported time savings from AI
6.4 hours/week spent fixing, reviewing, and reworking AI output
69% of workers ship AI output to colleagues/customers without checking
Net time savings: ~4.6 hours/week — less than half the headline number
Caveat: Self-reported time savings are notoriously inflated. The actual net gain may be even lower once you account for the learning curve, tool-switching overhead, and prompt iteration time that workers don't clock as "AI QA."

The 11-hour number gets cited in every AI vendor pitch deck. The 6.4-hour number doesn't. That gap tells you everything about how the industry measures success.

The Impact: Your "AI-Powered" Team Is Running at 40% Efficiency

Let's do the math for a 100-person team.

If each person saves 11 hours/week on paper, that's 1,100 hours of reclaimed productivity. At an average fully-loaded cost of $50/hour, that's $55,000/week in theoretical value.

But 6.4 hours per person goes to botsitting. That's 640 hours — $32,000/week — spent fixing AI mistakes. The real savings drop to $23,000/week. And that's before you factor in the 69% shipping unchecked output, which creates downstream rework, client-facing errors, and trust erosion that's impossible to quantify but very real.

The hidden cost isn't the AI subscription. It's the quality debt accumulating in every document, email, and report your team produces.

The Bottom Line

The AI industry measures success by output generated. Workers measure success by work completed. Those are not the same thing.

Until companies start measuring net productive time — output minus QA, minus corrections, minus rework — the productivity numbers will keep lying. And the botsitting hours will keep climbing.

The smartest teams I've talked to have one rule: if the AI output takes longer to fix than to write from scratch, you're using the AI wrong. Most teams haven't done that math yet. They should.

Sources: Glean Workforce AI Survey 2026 — 6,000 workers surveyed