When a piece of AI-written work goes wrong in your business, whose name is on it? If you hesitated, we just found the gap. Most owners I work with are early in their AI journey, and the first thing they celebrate is speed. A proposal that took an hour now takes ten minutes. A job posting, a customer email, a report summary, all done before the coffee cools. I’m glad they’re trying it. But speed at the front of the process can hide a cost at the back, and that cost lands on whoever has to fix the work.
A Harvard Business Review article (opens in new tab) from researchers at BetterUp Labs and Stanford gave the problem a name: workslop. It is AI-generated content that looks polished but lacks the substance to do its job, so the thinking gets pushed onto the coworker who receives it. In their research, 41% of workers had run into it, and each instance took close to two hours to sort out. That is most of a morning gone because one person skipped a read-through and the next person inherited the problem.
Accountability is not a poster on a wall or even written standards of performance in a job description, without the employer consistently calling out areas where the employees are not following the guidelines.
— Gary Henson, Founder of Gary Henson Group
The numbers from inside the work look similar. A January 2026 survey of 1,100 U.S. AI users at companies with 250 or more employees, conducted by Centiment for the software company Zapier (so weigh it with that in mind), found workers spend an average of 4.5 hours a week revising, correcting, or redoing AI output. Only 2% said they generally don’t need to revise it. The same survey found that untrained workers were six times more likely than trained ones to say AI makes them less productive. Training matters as much as the tool does.
When I read figures like these, the problem I see is expectations, the same one I find when I walk into a business to work on culture. Most owners misread this. Employees want to be held accountable, and they want to understand exactly what the expectation of them is. If the expectation is unclear and the work comes back wrong, the employee gets reprimanded, and you end up with a negative environment. If the expectations are made clear, both written and oral, a quick conversation about what’s next makes a real difference.
Now apply that to AI. If nobody has said who checks AI-assisted work before it leaves the team, your people are guessing. Some will check carefully. Some will paste and send. The person on the receiving end, who is your internal customer, pays for the difference in time, and sometimes the external customer pays too. In the Zapier survey, 25% of respondents reported a customer complaint after AI-assisted work went out, and finance and accounting teams reported the most trouble, with 85% describing at least one negative consequence.
of workers had received AI-generated “workslop,” and each instance took close to two hours to sort out. Harvard Business Review (BetterUp Labs and Stanford) →
Here is where I would start. It takes one meeting and one page of paper.
First, put one name on every piece of AI-assisted work. The person who sends it owns it, the same as if they had typed every word. The tool doesn’t sign anything. Second, write down what gets verified before work leaves the team: names, dates, figures, and anything a customer will read. Third, decide what is off limits to a first draft from AI, such as pricing, legal language, and anything involving a sensitive employee matter. Fourth, bring AI into your monthly meeting as a standing item and ask the three questions I use everywhere: what’s working, what’s not working, and what’s missing?
Reviewing AI output is a skill, and it has to be taught. Put a strong result next to a weak one for the same task, so the standard is something your people can see rather than something they guess at.
Then do the part most owners skip, which is follow-through. Follow-through means one-on-one conversations, held consistently, where you point out what was missed and what comes next. Handled early and without drama, those conversations lift the quality of the work and protect the relationship. Handled late, they feel like a reprimand.
I would tell any client that “using technology to be more efficient in the literal sense (like having AI answer your phones or do the work for you) may not be valuable.” The reason is simple. If the hours AI saves at the front are spent on rework at the back, you have moved the work around without reducing it. The owners who come out ahead will treat AI the way they treat any new hire: train it into the role, set the standard, and check the output until trust is earned. If you haven’t trained your people yet, start with Before You Cut a Job for AI, Train the Person in It.
This is the kind of structure at the center of GH Group’s Inside-Out Method™: clear expectations first, then the tools, so your team gets the speed without handing the cost to the next person in line.
