Good Transformer
Telling your team to use AI is not training them
Untrained AI use produces work that looks finished and is not, and whoever receives it spends about two hours putting it right.
What does it cost to tell your team to use AI without training them? Around two hours of somebody else's time, every time a piece of unfinished work gets passed along. In a survey of American desk workers, two in five said they had been handed work like that within a single month.
Those figures come from researchers at Stanford's Social Media Lab, who gave the problem a name: workslop. It means output that looks polished but is flawed enough that the person who receives it has to work out what is wrong and redo it.
This has become everyone's problem rather than a technical team's because AI drafting now sits inside the tools your staff already open every morning. Output reaches colleagues by default, whether or not anyone was taught to check it.
Most firms are set up to produce exactly this. They buy the tools, tell everyone to use them, and count that as adoption. The training either never happens, or it covers which menus to click and nothing else. The result is a cost that moves quietly onto somebody else's desk.
What untrained AI use produces
The Guardian reported one version of this in April. A copywriter at a large Miami cybersecurity firm spoke to them under a pseudonym. He described what happened after his chief executive made several colleagues redundant, then told the rest to use AI chatbots to raise output.
First drafts came easily. Rewriting and reconciling each other's chatbot output then took the team longer than the whole job had taken before the mandate. He summarised it this way: "Quality decreased significantly, time to produce a piece of content increased significantly and, most importantly, morale decreased."
When staff reported the drop in output, the executives blamed the staff.
Be careful what you take from that case. Morale was going to fall after a redundancy round whatever came next, and the chatbots cannot be blamed for all of it. What the mandate added was a second problem: work that arrived looking finished and was not, on top of a team already short-handed.
Writing in Harvard Business Review, the researchers reported a survey of 1,150 full-time desk workers carried out in September 2025. Of those, 40% said they had received workslop in the previous month, and sorting out a single instance took about two hours on average.
The direction matters more than the exact figure. The time was not saved. It was transferred. It moved from the sender to the receiver, and nobody measures it there.
The gap between who encourages AI and who uses it
Jeff Hancock, the Stanford researcher who co-authored the study, named the cause in the Guardian's reporting: "People are being told to use AI, often without direction or support."
British evidence points at who is doing the telling. The Chartered Management Institute polled more than 1,000 UK managers for its June 2026 report, Artificial Intelligence; Real Leadership. Sixty-four per cent of senior leaders say they encourage their people to experiment with AI. Asked separately, only 13% of managers strongly agreed that those leaders use the tools themselves.
Just 12% of managers said they feel very confident managing a team that uses AI.
Put those together and you get a specific situation. Leaders who do not use the tools are telling managers who do not feel able to supervise the output to press on. In that setting, unchecked work travels furthest before anyone stops it, because nobody feels qualified to send it back.
None of this argues against using AI. In the hands of someone who can spot a draft that only looks right, the same tools save the receiving colleague time instead of costing it. Firms are buying tools and licences. Almost nobody is buying that judgement, and it has to come first.
Put judgement before tools
The usual sequence is tools, then mandate, then training once the results disappoint. Reversing it works better, and it starts smaller than most leaders expect.
Build the leader's judgement first. Someone senior has to be able to look at AI-assisted work and say what is wrong with it. That means judging the work, not the technology: whether the claims are true, and whether it would survive a client reading it closely. A leader who cannot do that cannot set a standard or coach anyone towards it, and will keep receiving polished work they have no way to assess. Building that judgement on a leader's own live work is what our one-to-one AI lessons for leaders exist to do.
Then train the team on their real work. Generic tool training produces staff who know the menus and cannot judge the output, which is why most AI training fails. Useful team training starts from the actual tasks a team does, works out which ones AI should touch, and rebuilds those tasks with the checks built in. Rebuilding one process takes a few days. A whole department changing the way it works needs someone alongside the team for months, and it is worth being honest with yourself about which of the two you are facing.
Then set the standard for what may be passed on. Training changes what people are capable of. A written standard is what stops unfinished work being sent anyway. Without one, the judgement you have just paid to build stays in a single person's head.
The handover standard
This is one question set a team agrees before anyone sends AI-assisted work to a colleague or a client. It takes a minute, and it is what prevents the two hours downstream.
Before you pass on any work AI helped you produce, you must be able to answer yes to all five.
| # | The check | Why it is there |
|---|---|---|
| 1 | Can you state the main point in your own words, without rereading the document? | If you cannot, you skimmed it rather than read it |
| 2 | Have you checked every name, figure, date and quotation against a source you opened yourself? | This is where AI-assisted work fails most often and most expensively |
| 3 | Have you cut anything you could not defend if a client challenged it? | Plausible filler survives only because nobody asks |
| 4 | Would you put your name on this if AI had nothing to do with it? | The standard does not change because the drafting was faster |
| 5 | Have you cut the length the model added, so it is no longer than the reader needs? | Padding is the most reliable sign nobody edited the draft |
If the answer to any of them is no, the work is not ready to send. That is the whole rule.
Three things make it stick. Attach it to a checkpoint you already have: the file review sign-off, the job sheet or the template you use for client drafts. A standalone list gets circulated once and then forgotten. Say plainly that using AI is fine, so that nobody conceals it, because concealed use cannot be checked. And have the leadership team run the five questions on their own work first, in public, before asking anyone else to.
What to do this month
Ask one question at your next team meeting: when work arrives from a colleague, how often do you have to work out what is wrong with it before you can use it? The answers will tell you whether you have this problem, and they cost nothing to collect.
If the answers are uncomfortable, the response is not more tools, and not a ban. Build the judgement at the top, train the team on their own tasks, and agree a standard for what leaves a desk.
If you want help working out which of those your firm actually needs, book a conversation with us.
Sources and further reading
- The Guardian, "Bosses say AI boosts productivity, workers say they're drowning in 'workslop'", 14 April 2026, by Ramin Skibba. Source for the Miami copywriter's account and for Jeff Hancock's quotation.
- Stanford Social Media Lab researchers, "AI-Generated 'Workslop' Is Destroying Productivity", Harvard Business Review, 22 September 2025. Survey of 1,150 full-time US desk workers. Source for 40% receiving workslop in a month and about two hours to resolve each instance. Not peer reviewed, and US-based.
- Chartered Management Institute, "Artificial Intelligence; Real Leadership: The Management Imperative in AI Adoption", 9 June 2026. Polling of more than 1,000 UK managers. Source for the 64%, 13% and 12% figures.
Related: the AI skills gap starts at the top.