Insights · Brand and culture

AI slop at work: what causes it and how to stop it

AI slop, or workslop, is AI-written work that looks finished and isn’t. Look past the tool and write down what good looks like.

A piece of work lands on your desk. It is tidy and longer than it needed to be. It could have been written for any company in your market, and one of the figures in it is wrong. You send it back. The second draft reads much like the first.

Managers are asking what to do about this: how to respond when someone hands in AI-written work and won’t redraft it, and whether quality standards are still worth holding once the tools are in everyone’s hands. The usual answers are a better tool or a course in writing prompts.

The name for it is AI slop. Two widely read explanations of the term agree on where it comes from. Jesus Iniesta’s guide to AI slop puts it down to publishing output that nobody has properly reviewed, whatever the tool. Leon Furze, who writes about generative AI, polled 188 people and reports that just over half blamed the author, while 8% saw slop as an unavoidable by-product of the technology. His definition carries a line worth keeping: “Slop is a judgement, not automatically a property of the text.” Both are writing mainly about what gets published online. This piece is about the version that arrives inside your company, from a colleague.

We think the cause sits earlier. In a growing company the standard for good work can stay unwritten. It lives in the founder’s edits and a few senior people’s instincts. Nobody can meet a standard they have never seen, and that is as true of a language model as it is of a new hire.

The cost lands on whoever receives the work

In 2025, researchers at BetterUp Labs and the Stanford Social Media Lab gave the problem a name: workslop, AI-generated work that looks finished and lacks substance. In their survey of 1,150 full-time desk workers in the United States, 40% said they had received it in the previous month. By their own estimate, each incident took nearly two hours to deal with. About half of those who received it said they came to see the sender as less creative, capable and reliable. BetterUp is a commercial coaching company and these are survey answers, so treat them as an indication.

Our read: the time saved goes to the sender and the bill goes to the reader. Sooner or later that reader is a client.

The checking is uneven as well. A 2025 study by the University of Melbourne and KPMG surveyed more than 48,000 people in 47 countries. Among employees who use AI at work, 66% said they had at some point relied on its output without evaluating it and 56% said they had made mistakes in their work because of AI. We read those figures alongside a simple point: a check needs something to check against. Where the standard is unwritten, there is nothing to hold the draft up to.

Without a standard, AI gives you the average

Left without direction, a language model tends towards the likely version of whatever you ask for. An experiment published in Science Advances in 2024 asked 293 people to write short stories, some with ideas supplied by a language model. The assisted stories were rated as better written. They were also more similar to each other than the stories people wrote alone.

A short story is a long way from a client proposal. Still, the pull towards the middle is one we recognise in business writing. One request we hear from leaders is that whatever AI does, it must protect the company’s voice and the personal touch clients notice. That instinct is right. It can only be acted on once the voice exists somewhere a person or a tool can read it.

One pattern we have met: the founder is the standard. Important work passes through their hands and gets corrected there, and the corrections are never recorded. If AI then raises the number of drafts, the queue at that desk only grows.

AI multiplies what you bring. Bring a clear standard and it carries that standard into more work than your senior people could review alone. Bring nothing and it fills the gap with everyone else’s writing. So standards matter more after a rollout than before it. When anyone can produce something that looks finished in a minute, the standard is what separates finished-looking from good.

The way your company works is an asset. It deserves to be codified on purpose, with the care you would give a product or a name. That is the work of Codify Your Signature, working sessions we run in which a team names what makes it distinctive and turns it into standards everyone can carry. The team leaves with its Signature Playbook.

A one-page written standard stops AI slop at the source

Write one for each type of work that matters, such as the proposal or the reply to a new enquiry. Each needs four parts.

  • What good looks like, with a real example. Choose a piece you were proud to send and mark up why it worked. An annotated example teaches more than adjectives such as “clear” and “warm”.
  • What we never do. The phrases you would not use with a client and the claims you won’t make without a source.
  • What must be true before it leaves. Every figure traced to where it came from. Nothing in it that could have been sent to a different client unchanged.
  • Who signs off, and on what. Name the person and the question they answer. “Is this accurate?” and “Does this sound like us?” are separate checks and can belong to different people.

This has a cost. A first version takes a senior person’s time for each type of work, and it will be wrong in places until the team has argued with it. A team small enough that everyone still sees the founder’s edits may not need it yet. You need it from the moment new people stop seeing them.

Give the same page to the tool and the new hire

With AI, the page does two jobs. It goes in before the draft, so the tool starts from your example and your “never” list. Then the sender reads the draft against it before anyone else sees the work.

It also changes the conversation with the person who won’t redraft. You are no longer arguing about whether they used AI. You are pointing at the line of the standard the work missed.

New hires and juniors should produce the work by hand against the standard a few times before they pass it to AI. Someone who has never made good work your way cannot tell when a draft falls short of it. Let them use AI once they can.

Start with the last thing you sent back

This week, find the last piece of work you returned to someone. Write down in plain sentences what was wrong with it and what you would have accepted in its place. Then ask whether the person who wrote it could have known that before they started.

If they couldn’t, you have the first lines of a standard. Give them to the person and to the tool, and read the next draft against them.


Sources

  1. BetterUp Labs and Stanford Social Media Lab, Workslop: the hidden cost of AI-generated busywork, 2025. An online survey of 1,150 full-time desk workers in the United States, conducted in September 2025.
  2. BetterUp, The hidden cost of AI “workslop”, 29 September 2025. BetterUp’s write-up of the same research, including how recipients came to see the sender.
  3. University of Melbourne and KPMG, Trust, Attitudes and Use of Artificial Intelligence: A Global Study 2025. More than 48,000 people in 47 countries, surveyed between November 2024 and mid-January 2025. Workplace figures are from the full report, page 75.
  4. Anil R. Doshi and Oliver P. Hauser, Generative AI enhances individual creativity but reduces the collective diversity of novel content, Science Advances, 12 July 2024. An online experiment with 293 writers and 600 evaluators.
  5. Jesus Iniesta, AI Slop Is Eating the Internet: The 2026 Guide, 9 February 2026. A guide to what AI slop is and how to avoid publishing it.
  6. Leon Furze, What is AI slop?, 30 August 2026. A definition built from reader polls, the first with 188 respondents.