I have spent two hours reconstructing the context behind a polished update before I could use it. The document looked finished, but it left out who had made the decision, who owned the work, and the reasoning behind the proposal.
BetterUp Labs and Stanford’s Social Media Lab call this kind of incomplete AI output “workslop.” In their September 2025 survey of 1,150 full-time US desk workers, 40% said they had received it in the previous month. The researchers reported that recipients recalled losing about two hours per incident. That is a survey average, separate from my own experience.
I call the transferred work the slop tax. A sender can finish a draft quickly while leaving the recipient to recover the missing facts and reasoning. The survey estimates the cost of reported incidents; it does not put a price on every AI draft.
Incomplete work changes how colleagues see you
Recipients also reported judging the sender differently. In BetterUp’s summary, 42% saw the colleague as less trustworthy, 50% as less capable, and 32% did not want to work with that person again. Those are reported reactions. The study does not tell us whether those judgments lasted or affected anyone’s career.
The practical problem persists even after the sentences sound better. If an update still lacks a decision or an owner, the recipient has to ask for the information that should have been in it.
Measure time through to completion
Workday’s January 2026 research reported on 3,200 active AI users at large employers, surveyed in November 2025. Eighty-five percent said AI saved them one to seven hours a week, but Workday estimated that rework absorbed nearly 40% of the reported savings.
Applied to an illustrative five-hour saving, 40% would mean two hours of rework and three hours saved overall. Individual results can differ; Workday asked people to recall time spent across workplace tasks rather than timing a writing exercise.
METR measured completion time in a narrower setting. In METR’s 2025 controlled study, 16 experienced developers worked on 246 tasks in familiar repositories. With early-2025 tools, tasks randomly assigned to allow AI took 19% longer than those assigned to be done without it. The developers had expected a 24% speedup. The result applies to that setting and generation of tools; it cannot establish the effect on writing.
METR’s 2026 follow-up reported that developers were less willing to submit tasks that might have to be completed without AI. Because of that selection problem, the apparent speedups could not reliably replace the earlier estimate.
What to do before you send
Before sending an update, check whether the reader can find the decision and the person responsible for it. Include the supporting facts and any open questions they need to resolve. If that information is missing, add it before editing for length.
Compare the draft with its source material and check every name, number, quotation, link, and qualification. Then read it aloud. A low writing score cannot tell you whether a colleague has enough information to act.
Gallup’s May 2026 data found that 65% of US employees at organizations using AI reported a positive effect on productivity and efficiency. To judge the gain in your own work, count the time through review and correction. Include the work your recipient has to do after you send it.
Zero Slop adds that review to your existing AI assistant. Try it on a draft and inspect every change before you accept it.