The hidden cost of AI output nobody edits

Research from BetterUp Labs and Stanford Social Media Lab, published through Harvard Business Review in September 2025, surveyed 1,150 US desk workers. Forty percent had received what the authors call workslop in the prior month: AI output that looks like work but does not advance the task. Each instance cost the receiving colleague one hour and 56 minutes. The estimate came to around $186 US per employee per month, or roughly $9 million a year at 10,000 staff.

Those are US figures on a US sample, so treat the dollar amounts as a scale indicator rather than something to paste into an Australian budget. The behaviour is the transferable part, and every marketing team has it.

Where it shows up in a marketing function

The generic version of this finding is useless. The specific version is uncomfortable.

The campaign recap that summarises what happened without saying what it means, so the person who receives it has to reopen the data and do the analysis themselves.

The competitor brief that lists what five companies do, with no view on which of it matters, so the strategist reads all five sites anyway.

The creative brief that describes the audience in language that could apply to any audience, so the writer books a call to find out what the client actually said.

The research summary that reads well and contains one confident claim nobody can source, so somebody spends 40 minutes trying to verify it before quietly cutting it.

The first draft handed over with a note saying AI did a first pass, can you polish, which is not a draft. It is a research task with a formatting layer on top, and the person receiving it now has to do both.

Every one of these looks like output. Each one moves work downstream instead of finishing it.

Why it is invisible on the dashboard

Workslop does not appear as a cost because it appears as productivity. The sender's output goes up. The receiver's time goes down, but it goes down inside their existing hours, so it shows up as the team is stretched rather than as a line item.

Then it compounds. HBR followed the original research in June 2026 with the concept of knowledge decay: degradation of the information base a company makes decisions on. A summary of a summary of a summary, each step confident and each step slightly less anchored to the source, eventually becomes the thing everyone believes.

That is the version that actually costs money. Not the two hours. The strategy built on a claim that entered the company as a plausible sentence in an unedited summary.

The rule that fixes most of it

One rule, applied without exception: nothing goes to another person unless the sender has verified every claim in it and would sign their name to it.

That is the whole fix. It sounds obvious and almost nobody enforces it, because the failure mode is polite. Nobody wants to reject a colleague's document over a claim they cannot source.

Make it structural instead of personal:

1. 1.Named owner on every document, not a team. Prepared by, at the top, always.
2. 2.Any number in the document carries its source inline. No source, the number comes out. This alone kills most knowledge decay.
3. 3.No first-pass handoffs. Either the piece is finished to the sender's standard or it stays with the sender. If somebody genuinely needs help with a draft, they book fifteen minutes and do it together, which costs less than two hours of forensic reading.
4. 4.Receivers are permitted, explicitly and in writing, to send work back as not ready with no further explanation. Without that permission the cost stays hidden, because absorbing it is the socially easier option.

The honest trade-off

This slows the front of the process down. A person who was producing four campaign recaps a week will produce three, because the fourth was never finished.

That is the correct outcome and it will look like a productivity drop on whatever dashboard tracks output volume. Anyone measuring content operations by artefact count will read the fix as a regression. Change the measure before you change the rule, or the rule will not survive the quarter.

There is a second trade-off worth naming. Enforcing this hardest on junior staff, who lean on models most, can look like a tool ban aimed at the least senior people. It is not, and it should not be framed that way. The standard is the same for everyone and it is about the output leaving your desk, not the tools on it.

What to check this week

Pick the last ten internal documents that circulated in your team. For each one, ask the receiver two questions: did you have to redo any part of this, and how long did that take.

Do not ask the senders. They do not know. That is the entire mechanism.

If the answer is more than a couple of hours across ten documents, the number in the study is your number too, and you now know where a slice of your team's capacity has been going.

Finished work is the deliverable. Everything else is a task you handed to somebody else with a nicer font on it.