What Is Workslop, and How Do You Avoid Producing It?

TL;DR: Workslop is AI-generated work that looks finished but lacks the substance to move a task forward, so the person who receives it ends up doing the real work^[1][2]. You avoid it by deciding what the work must achieve before you open a tool, giving the tool real material, checking every claim that matters, and refusing to send anything you couldn't explain and defend yourself.

I think we should look at AI like a smart intern who is doing something for the first time. Altough it's smart, but it needs guidance.

What is workslop?

Workslop is AI-made work content that passes for good work but can't advance the task it was made for. The researchers at BetterUp Labs and the Stanford Social Media Lab who surveyed workers about it define it as "AI-generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task"^[1].

In practice it might be a four-page project update that never says whether the project is on schedule. Or a slide deck with confident headings and no numbers. Or a literature summary that cites papers nobody checked, or a reply to a client that is polite and complete and answers a slightly different question from the one they asked.

The Harvard Business Review piece on the research describes it as "low-effort, passable looking work that ends up creating more work for their coworkers"^[2]. That last part is the useful test. A rough draft you label as a draft is honest. The trouble starts when unfinished output gets handed over as if it were done.

Why is workslop a problem if it saves time?

It saves the sender time and spends the receiver's. The cost moves to someone else, usually someone who didn't agree to pay it.

In the survey, recipients said each instance took an average of 1 hour 56 minutes to deal with^[1][3]. That's the time spent working out what the document is trying to say, checking figures against the original source, asking follow-up questions, and sometimes redoing the work from scratch.

A 2026 paper on the ethics of workslop splits responsibility in a helpful way. Managers shape the structural conditions, such as workloads, deadlines, metrics, AI policies and training. Workers control local choices about whether and how to use AI, "how carefully to check its output, and how much burden they pass on"^[4]. You can't fix your team's deadlines alone, but the second half of that list is yours.

How common is workslop?

The only solid number comes from one survey: 40% of 1,150 full-time U.S. desk workers said they had received workslop in the previous month^[1][5].

The same researchers estimated a productivity cost of about $186 per worker per month, and more than $9 million a year for an organisation of 10,000 people^[1][3]. Those figures get repeated in headlines, so treat them with care.

So the advice below is sensible practice, with thin evidence behind any single step. It's still worth doing, because each step is cheap and the downside of skipping it lands on a colleague.

How do you avoid producing workslop?

Keep your own judgment in charge at every step: define the outcome, feed the tool real material, check the claims, and edit for the person who will read it^[7][8][6].

Decide what done looks like before you start

Deel's guide for leaders opens with "Start by measuring outcomes"^[7], and the same applies to one person's task. Before you open any AI tool, write two sentences. Who reads this, and what should they be able to decide or do afterwards? "My manager needs to know whether we can ship on the 14th, and if not, what slips" gives you something to check the output against. "Write a project update" gives you nothing.

Give the tool real material

Generic input produces generic output. Paste in the meeting notes, the data, the client's own email, the section of the spec that matters. Providing relevant source material and context is one of the most consistent recommendations across the guidance^[7][8][6].

Ask for a draft or a structure

AI is useful for drafting, outlining and reorganising material you already have. It is risky when you ask it to supply facts, sources or authority it has no basis for^[7][8][6]. If you catch yourself asking it for "some statistics on this", stop and go find the statistics yourself.

Ask it to critique its own output

Korn Ferry suggests asking the AI "to critique its own output and identify what's missing or weak"^[6]. This is a good prompt to start your own review with. It doesn't replace that review, since the same tool that missed something may miss it twice.

Check every claim that matters

Proofed lists "fact-checking and verifying elements such as statistics and other claims" as a core step^[8]. Go through the draft and mark every number, name, date, quote and citation. Trace each one to a source you trust. If you can't find it in twenty minutes, cut it or say plainly that it's unverified.

Edit for the reader

This is the step where most workslop falls short. It has paragraphs where the reader needed a decision. Put the conclusion first. Cut paragraphs that restate the question. Name the trade-offs. End with a next step someone can act on.

How do you know it's ready to send?

It's ready when you could defend it in a meeting without the tool open. The person submitting the work should be able to explain its assumptions, confirm important claims against reliable sources, say what is still uncertain, and take responsibility for the final version^[6][9][4].

Before you hit send, ask:

• Could I explain every assumption in this if someone asked?
• Have I checked each claim that would cause harm if wrong?
• Does it say what is still uncertain?
• Does it open with a clear conclusion and end with a next step?
• Was AI the right tool for this at all?

That last question comes from Russell Reynolds, who argues that people need to learn "how to recognize when AI is the wrong tool for the task"^[9]. A two-line reply to a colleague, a judgment call about a client, or a decision that depends on context only you have may be faster and better written by hand.

Why does workslop happen even to careful people?

Usually because there was no time left to review. When the deadline is at five and the draft appears at ten to five, the checking step is the one that disappears. The ethics paper places workloads and deadlines on the manager's side of the ledger^[4], but you can protect review time in your own plan.

Treat review as its own job with its own time estimate. "Draft the update, 30 minutes" and "check figures against the dashboard and cut filler, 20 minutes" are two separate jobs, and the second should be scheduled before the deadline with room to spare. When TimeBud breaks a goal into jobs, you can add the review job as a dependency. We only plan into open calendar time and leave breathing room after events, so the check doesn't get squeezed into the last ten minutes.

What can teams and managers do about it?

Reward substance over volume or speed, set clear rules for AI use, train people in prompting and fact-checking, and use peer or editorial review for important work^[7][9].

If a team counts documents produced or tickets closed, it will get more documents and more closed tickets, some of them hollow. Russell Reynolds argues that leaders need to understand "how to provide feedback that refines outputs" and "how to blend AI capabilities with human insight"^[9]. A simple team norm helps: if you send something AI-assisted, you're expected to be able to answer questions about it the same day without rerunning the prompt.

The research on workslop is the BetterUp and Stanford team's published work^[1], with coverage in Harvard Business Review^[2] and The Conversation^[3] if you want to read the original framing.

Try TimeBud free