TL;DR
- Take on the cognitive load: edit and shorten AI drafts.
- Be clear about AI’s role when it affects how people should read the work.
- Check facts and sources before sharing them.
- Ask before using someone else’s work with an AI tool.
- Use AI when it helps; write it yourself when it doesn’t.
Did AI save time for the team, or only for me?
Generative AI makes it easy to produce messages, documents and presentations. But saving time for the person writing them does not always save time for the team. Sometimes using AI saves the writer ten minutes and costs everyone else twenty minutes trying to make sense of the result.
One phrase from a recent conversation with my colleagues stayed with me:
“Cognitive load shouldn’t shift from the writer to the reader.”
I have nothing against people using ChatGPT or Claude. What gets to me is opening a polished document, reading the headings and takeaways, and still having to work out what the author is actually saying. I’ve had that happen more than once.
When AI saves time for one person
In 2025, BetterUp and Stanford Social Media Lab published research on “workslop”: AI-generated work that looks finished but creates more work for the person receiving it. In their survey, 53% of employees said at least some of the AI-generated work they send could probably fit that description.
That number didn’t surprise me. It is so easy now to produce another page, another set of options, another summary or table. Writing 1,500 words instead of 300 takes very little extra effort. We haven’t suddenly gained more time to read it all; somebody still has to work out what matters and what needs a response.
We have AI policies. Do we need AI etiquette as well?
Our policies cover confidential information, personal data, security, intellectual property and customer communication. I started wondering what guidance we have for the everyday cases.
I mean the smaller things: a huge document that could have been five paragraphs, a statistic nobody checked, or a polished message that leaves you guessing what its author thinks.
While reading about this, I came across AITiquette, a community project about thoughtful AI use at work. I used AI to help with some of that research, which felt fitting. It got me thinking about the principles I want to follow, and I wrote down five.
1. Take on the cognitive load yourself
If an AI assistant gives me 1,500 words, my colleague does not have to receive all 1,500. I still have to work out what’s worth sending and what they actually need.
I suggested to colleagues that we do more than proofread an AI draft: we should cut it down too. Some felt that was too much to ask for every little message. I can see their point. I still think that if AI saves me twenty minutes, I can spend a couple of those minutes making the message less of a chore for the next person.
Generating another page costs me almost nothing, but someone else still has to read it.
2. Be clear about where the work came from
I do not think every Slack message needs an “AI-assisted” label. If I tagged every sentence an AI tool helped me rewrite, people would stop paying attention to the label. There are cases where the process matters, though. Did you analyse the data or ask a tool to summarise it? Did you read the original document? Did you check the recommendation? Are these your conclusions, or options a model came up with?
If I have only asked a model for options, I should say so. If I checked the source and read the original, I can say that too. My colleagues do not need a label on every message; they need enough context to know what I did and how much confidence to place in the result.
3. Don’t outsource fact-checking to the reader
I’ve seen a confident “research shows” followed by a statistic and a link. Then I open the source and find it says something else. AI can help find research or navigate a long report, but I shouldn’t leave it to a colleague to check whether the claim holds up.
If I send someone a statistic, I should make sure the source exists, the number is right and it supports my point. A wrong claim can end up in a presentation or influence a decision before anyone catches it.
4. Be careful with other people’s work
The answer depends on the context and the tools your company allows. Is it okay to upload someone else’s draft, summarise a private meeting, or rewrite a colleague’s text without telling them? What about using an internal discussion to generate a presentation?
Policies cover some of this, especially security and confidentiality. When I’m unsure about the everyday cases, I ask myself, “Would I be comfortable telling the person exactly what I did with their work?”
If not, I should ask first.
5. Use AI when it actually helps
I use AI regularly. It helps me explore alternatives, challenge an argument, organise messy thoughts and handle repetitive work. But sometimes I reach for it just because it is there.
For example, I’ll ask a model to draft a short message, then delete the intro, the bullets and the conclusion until only four lines remain. I could probably have written those four lines myself.
I do not want to use AI just to say I used it. If I can’t point to what it improved, I probably didn’t need it.
A few habits may be enough
When I brought these ideas to my colleagues, they had their own suggestions, and not everyone agreed with mine. That was helpful. More than once, someone asked me not to turn this into another internal document.
I understand why. Nobody wants another long policy to read. I’d rather see people read what they send, cut what their colleagues do not need, check their sources and say when they used AI if it matters. That seems like a good place to start.
My colleagues and I are still figuring out what good AI-assisted work looks like. I have started asking myself:
Did AI save time for the team, or only for me?
If the answer is “just me,” I know I have another edit to make before I send it.
I wrote the first draft of this article myself, then used AI to help revise and shorten it. I reviewed the final version and decided what to keep.