A project update lands in your inbox. The formatting is clean. The tone is confident. Everything is “progressing well.”

You read it twice and still cannot tell whether the project will finish on time.

So you reply. What is finished? What is blocked? Do you need something from me?

The person who wrote the update may have saved time. You are now spending yours completing it.

That is the risk behind AI workslop: a faster draft can move work from one person to another while looking like a productivity gain. AI can make unfinished thinking look finished, turning incomplete notes into confident prose before the facts, decisions or next steps are settled. The extra work then lands with a colleague, client or manager.

The time saved belongs to the whole job

Researchers at BetterUp Labs and Stanford’s Social Media Lab use “workslop” to describe AI-generated material that looks substantial but leaves others to make it useful. In their September 2025 survey of 1,150 full-time U.S. desk workers, 40% reported receiving it in the previous month. Respondents reported spending roughly two hours resolving an incident, on average. These are self-reported experiences, not a measurement of what AI costs every business. BetterUp’s research overview

There is credible evidence that AI helps with writing, too. In a randomized experiment involving 453 college-educated professionals, access to ChatGPT reduced average completion time on professional writing tasks by 40% and improved assessed quality by 18%. Those results describe the tasks studied, rather than the full cost of using AI across a company. Noy and Zhang, Science, 2023

These studies ask different questions. Neither gives you a shortcut to judging your own workflow.

My view is that we should count the effort needed to make work usable, including checking, clarification and corrections. A document can be quick to produce and expensive to receive. The sender’s time saving is only part of the calculation.

A professional tone can conceal the decision

Consider a fictional agency preparing to launch a client’s website. The pages are built, but checkout testing is blocked until the client grants payment-provider access. The launch date remains a target until those tests pass.

A vague update might say:

“We’re making excellent progress and coordinating the final requirements for a smooth launch.”

Nothing in that sentence helps the client unblock the project. Worse, its reassurance makes the missing action easy to overlook.

A useful update would say:

“All five pages are built. Checkout testing is blocked because we still need access to your payment provider. Please grant access through our agreed secure process by Tuesday at 2 p.m. to keep next week’s launch target. Alex will run the tests once access is available. We’ll confirm the launch date after checkout passes.”

The second version gives the client something to do and explains the consequence of waiting. It also preserves a distinction that matters commercially: a target date is not a confirmed commitment.

Both examples are invented. The difference between them is the information the recipient needs to act, rather than the quality of the prose.

Decide what done means before asking AI to draft

“Make this sound professional” is a weak brief because it says little about the purpose of the work. Before opening an AI tool, decide what the recipient needs to do next.

For a client update, that might mean understanding a delay and providing an approval. For a supplier comparison, it might mean choosing between quotes with different delivery dates and exclusions. For meeting notes, it might mean knowing who agreed to do what.

That purpose gives you a useful completion standard:

  • The decision or action is clear. The reader knows what is being asked of them, or that the message is for information only.
  • The facts needed for that action are present and checked. Prices, deadlines, scope and conditions match the underlying material.
  • The limits are visible. Missing information, assumptions and unconfirmed commitments have not been smoothed away.

Give AI those requirements alongside your notes. Ask it to flag missing information instead of filling gaps. Use an approved tool for work information and check what data belongs in it. Leave passwords and unnecessary personal details out.

Then check the result against the source. Asking the same tool to review its answer may help, but it does not independently establish that the answer is correct.

Leave room for honest unfinished work

This standard should not force people to pretend every question is settled before they can share anything.

“We have two supplier quotes, but neither confirms delivery” can be a useful update. So can “No owner was assigned in the meeting.” Both make the remaining work visible.

The important distinction is whether the recipient can see what still needs doing. If you need someone else’s judgment, ask for it directly. If you are sharing an early draft, say which parts need review. A document that looks final but requires someone to reconstruct its assumptions is an avoidable burden.

If you review other people’s work, this also changes how to respond to poor AI-assisted drafts. “Make it better” invites another polished version. “Which decision does this support, and what information is still missing?” points to the actual problem.

Give the next person less work

You do not need a new reporting process to apply this. Start with the next email, proposal or handoff you are about to send. Check whether the person receiving it has enough information to take the intended next step. Resolve the gaps you can, and name the ones you cannot.

When evaluating AI in your team, pay attention to what comes back: clarification questions, corrections and decisions that still cannot be made. Those are clues that the apparent time saving may have been passed downstream.

The useful promise of AI is that it can help us get more work finished. Keep that as the standard, and a polished draft becomes a means to an end rather than evidence that the job is done.