Why someone has to decide: workslop, trust and how we use AI

Daily Practice, No. 1  ·  For: anyone who uses, buys or reads AI-assisted work  ·  Reading time: 5 min

AI makes it easy to produce something that looks finished. A tidy report. A confident summary. A polished slide. The danger is not that it looks bad. It is that it looks good and says little.

Researchers at BetterUp Labs and Stanford Social Media Lab gave this a name: workslop. It is AI-generated work that appears polished but lacks the substance to move a task forward. They published it in Harvard Business Review in 2025.

What they found

In a survey of 1,150 full-time US employees, self-reported:

• 40% had received workslop in the past month.
• Of those, 53% felt annoyed and 42% saw the sender as less trustworthy.
• Dealing with each incident took nearly two hours: fixing it, asking questions or redoing it.
• 18% of AI users admitted sending low-effort or low-quality AI content.

Show and tell: real survey data. These are self-reported figures from US employees.

Notice what workslop does. It does not save the time. It moves the work to the next person, and costs trust on the way.

It is not just a workplace problem

A report prepared for a government has gone out with misquoted sources and references that did not exist, because nobody checked. The fix is never a better tool. It is a person who reads it.

A caution about statistics too: the widely quoted 95% of AI pilots fail comes from a preliminary MIT report (2025), based on interviews and a survey with self-reported outcomes. Treat it as a direction, not a law. Its useful finding is that results depended on how AI was fitted into real work, not on the model.

The five principles we follow

We are an AI-native company that is human first. AI helps us explore data faster. Humans decide what is true, what matters and what gets published.

• People stay accountable. A named person signs off before anything is published.
• Sources first. AI never replaces the primary source. Every figure shows where it came from.
• Your data stays yours. We use your data only for your project, never to train AI models, and delete it when you ask.
• We say where AI was used. In plain words, on request and in every client report.
• No harm. We do not invent quotes or people, imitate real individuals or push a political line. Stories about people are told with dignity and, where it matters, with consent.

A four-question check before you send AI-assisted work

• Did I read every line?
• Did I check every number against its source?
• Would I put my name on it?
• Does it say where AI helped?

If any answer is no, it is not ready.

Why we built it this way

Because trust is the product. Whether it is a chart in the Playground or a forecast for a board, the part you pay for is the judgment a person brings to it.

Have data? Let us talk