Botsitting Explained: The Hidden Hours Spent Checking AI

Last updated: November 6, 2026

You asked an AI tool for a first draft, a summary or a spreadsheet formula, and it came back in seconds. Then you spent the next half hour checking names, fixing a wrong number, rewriting the tone and asking again. The tool saved you time. Or did it?

That half hour now has a name: botsitting. The word spread this week after a recent survey of 2,052 US workers was syndicated to local news sites across the country. This page explains what botsitting means, where the word came from, and what seven studies actually measured, with every number linked to its source.

Key Statistics

• Workers spend 6.4 hours a week botsitting, 37% of the time they spend with AI, according to Glean's survey of 6,000 digital workers in the US, UK and Australia. Glean June 2026
• 82.5% of US workers who use AI weekly spend some time each week overseeing, checking or fixing it, and 15.8% spend 6 hours or more. Founder Reports October 2026
• 60% of workers who have trained an entry-level employee say getting good work from AI takes at least as much effort. Founder Reports
• Nearly 40% of the time AI saves is lost to rework, such as correcting errors, rewriting content and verifying outputs. Workday January 2026
• The average enterprise AI user spends about 4.5 hours a week revising or redoing AI output, and only 2% say they generally don't need to revise it. Zapier
• 40% of US desk workers received "workslop," AI content that looks good but lacks substance, in the past month, and each case took about 2 hours to resolve. BetterUp Labs and Stanford Social Media Lab September 2025
• 61.8% of workers who manage AI say their employer doesn't fully account for that time or they're unsure. Founder Reports

What Is Botsitting?

Botsitting is the work of making AI output usable: giving the tool context, watching what it does, checking the result and fixing what it gets wrong. Glean, which introduced the term in its June 2026 Work AI Index, calls it "the unrecognized work required to make AI usable." The word blends "bot" and "babysitting," and the joke lands because the work feels like minding a capable but unreliable helper.

The key word in Glean's definition is unrecognized. Botsitting rarely shows up in a job description, a timesheet or a productivity report. When a company says AI saves each employee 11 hours a week, the hours spent cleaning up after it usually sit outside that number.

Founder Reports, a research publisher, uses a slightly wider definition in its October 2026 report: "the work of training AI, overseeing it, checking its output, and fixing what it gets wrong." Both definitions describe the same thing from the worker's side of the screen.

Where Did the Word Botsitting Come From?

The earliest dated use we could verify is Glean's Work AI Index, released on June 10, 2026. Glean introduced two terms together: botsitting, and a cruder one, "botshitting," for shipping AI work that the person has not verified, doesn't fully understand or can't stand behind. Rebecca Hinds, who heads Glean's Work AI Institute, presented the findings.

The word then moved through podcasts and newsletters, including an episode of The AI Daily Brief on June 26, 2026. It reached a much wider audience in early November, when an article on the Founder Reports survey was syndicated by Stacker to local news sites from November 3, 2026.

Wiktionary has had a "botsitting" page since June 15, 2026, but its current definition describes looking after software bots for their owners, a different meaning from the workplace sense. No major dictionary lists the word yet.

How Much Time Do People Spend Botsitting?

Most estimates land between about 4 and 6.5 hours a week for people who use AI regularly. Glean measured 6.4 hours a week across its 6,000 workers, and Zapier found about 4.5 hours a week spent revising or redoing AI output. At the top of that range, it's close to a full working day each week spent minding the tool rather than doing the task.

Glean breaks its figure into three parts: feeding AI context takes 2.3 hours a week, supervising its output 2.2 hours, and debugging its mistakes 1.7 hours, according to Glean's report summary. The same summary says 36% of AI sessions need substantial rework or a full restart.

The Founder Reports survey asked the question differently and got a similar picture. Of 2,052 US workers who use AI at least weekly, 82.5% spend some time on oversight each week and 15.8% spend 6 hours or more. Managers carry more of it: 40.7% of managers spend 3 or more hours a week on AI oversight, compared with 25.0% of individual contributors.

The Botsitting Studies Side by Side

Each study measured the problem in its own way, so the numbers aren't directly comparable. Here is what each one counted, who it asked and when.

Four of the five come from companies that sell AI or workplace software: Glean sells an enterprise AI assistant, Workday sells HR and finance software with AI features, Zapier sells automation tools and BetterUp sells workplace coaching. That doesn't make their numbers wrong, but each one suggests a fix that its maker sells, so it's worth naming the source when you quote them.

Is Botsitting the Same as Workslop?

No, but they are two sides of the same problem. Workslop is the low-quality AI output itself, defined by BetterUp Labs and the Stanford Social Media Lab as "AI-generated content that looks good, but lacks substance." Botsitting is the time spent checking and fixing output, whether it's your own or a colleague's. Workslop that one person sends creates botsitting for the next.

The workslop research, first published in Harvard Business Review in September 2025, put a price on it. BetterUp reports that 40% of US desk workers received workslop in the previous month, that each case took an average of 2 hours to sort out, and that this costs about $186 per employee per month, or $9 million a year for a company of 10,000 people. Those are BetterUp's own estimates from a survey of 1,150 full-time desk workers.

Does Botsitting Cancel Out the Time AI Saves?

Not entirely, but it takes a large share. Workday found that nearly 40% of AI time savings are lost to rework, and that only 14% of employees consistently get clear, positive net results from AI. Glean's workers said AI saves them about 11 hours a week, while also reporting 6.4 hours of botsitting. The savings are real; they are just smaller than the headline number.

Some of the strongest evidence comes from software developers, who can time their work precisely. In a 2025 study by METR, a nonprofit research group, 16 experienced open-source developers took 19% longer to finish tasks when allowed to use AI tools, even though they expected AI to speed them up by 24%, and afterwards still believed it had sped them up by 20% METR July 2025. METR now notes that these early results no longer reflect current tools, so treat it as a lesson about how easy it is to misjudge AI time savings, not as today's speed.

Stack Overflow's 2025 survey of developers points the same way. The most common frustration, cited by 66%, was "AI solutions that are almost right, but not quite," and 45.2% said debugging AI-generated code takes more time Stack Overflow Developer Survey 2025. "Almost right" is the exact output that needs botsitting.

Why Do People Skip the Checking?

Because checking is slow and nobody counts it. Glean found that 69% of AI users admit to botshitting, including 41% who sometimes deliver work they couldn't explain if asked and 28% who have blamed AI for mistakes they caused. The KPMG and University of Melbourne global study of more than 48,000 people in 47 countries reports that many "rely on AI output without evaluating accuracy (66%)" and are "making mistakes in their work due to AI (56%)" KPMG 2025.

There is a split here. Workday found that 77% of employees review AI work at least as carefully as work done by people. So some workers check everything and carry the hours, while others pass unchecked work along, and the checking lands on whoever receives it.

What Botsitting Means If You Write With AI

For writers, bloggers and editors, botsitting is mostly fact-checking. AI drafts fail in predictable places: invented quotes, wrong dates, statistics with no source and links that go nowhere. Those take longer to catch than typos, because they read smoothly. It's the same reason our AI writing statistics page links every figure to the study it came from.

A few habits keep the time down:

• Ask for sources, then open them. A source the AI names is a lead, not a fact, until you've read it yourself.
• Check names, numbers and dates first. They are the errors readers notice and the ones most likely to be wrong.
• Budget the time. If a draft takes 10 minutes to generate and 40 to verify, plan for 50, not 10.
• Say what you checked. A short "last updated" line and linked sources tell readers the work was verified by a person.

The same gap between promise and practice shows up outside writing: AI agents that book or buy things for people still complete very few orders, as we found in our look at AI shopping agents.

What to Watch Next

• A dictionary entry. Words that spread this fast often get added to dictionaries' watch lists. Several word-of-the-year announcements are still to come for 2026.
• Companies counting the time. 61.8% of workers who manage AI say their employer doesn't fully account for it, according to Founder Reports. Expect more firms to start measuring AI oversight in performance reviews and workload plans.
• Better tools, fewer hours. AI makers are pitching agents that check their own work. The test is whether the weekly hours in surveys like Glean's start to fall in 2027.

Methodology

We gathered every study on this page from its publisher's own page and read each figure there on November 6, 2026, recording the sample, fieldwork dates and exact wording. Where a publisher sells AI or workplace products, we say so. The side-by-side table compares studies that measured different things in different ways, so read it as a range, not a single answer. We did not include figures we could only find second hand.

How to Cite

You're welcome to use the numbers and tables on this page. Please credit BestWebsiteTemplate.com and link to this page, and cite the original study for each figure, which is linked next to it.