Using AI honestly at work · Lesson 9

Nobody has told you the rule

From memory

Lesson 7

Ask workers and ask employers the same question and you get two different worlds. In a CNBC and SurveyMonkey survey of more than 1,600 US workers and students, fielded in late July 2026, 55 in every 100 said their employer has no official policy on AI use at all. Ask employers, and in Littler's 2026 survey 68 in every 100 said they had a formal policy in place, up from 38 the year before. Both can be true. Employer surveys reach the people who write the rules. Worker surveys reach the people the rules were meant to reach.

A few jobs are the exception and already have hard rules. By early 2026 at least 25 federal courts in the United States required lawyers to certify whether AI was used in a filing, and most UK universities now require students to declare it. If you are in one of those worlds you already know. Everyone else decides for themselves, which is what the next eight minutes are for.

One heads up. From here this lesson asks about your actual job. It asks for your job title and for words you write yourself, and nothing else. It never asks you to paste in a real document, a real email, or anything about a real person. And if you are not at work right now, do not go hunting. Leave the first box below unticked, carry on, and spend two minutes on it next time you are at your desk. Not knowing is the most common answer anyway, and the rest of this lesson works either way.

I have searched my staff handbook, intranet or onboarding pack for the words AI, artificial intelligence and generative
I know which of these is true where I work: there is a written rule, there is something said out loud that was never written down, there is nothing, or I have not been able to find out yet
If there is a rule, I know whether it covers what I type in as well as what I do with what comes out. If there is no rule, I have noted that too
I have thought of one person I could ask casually, without it becoming a formal question
I know where my own three line do-not-paste list from lesson 7 is, or I have just written it again. A list you cannot find is a list you will not use

Ten minutes to find out whether a rule exists where you work, and to write your own if it does not.

Lesson 09 · Making it yours

By the end you can

Find out whether a rule already exists where you work
Spot work that looks finished and pushes the thinking onto the reader
Sort your own tasks into say nothing, say something, and do not use it
Write six rules you could follow tomorrow without asking permission

What you need

Any of the four. This lesson asks about your job and for words you write yourself. It never asks you to paste a real document.

Fundamentals of AI

Read it as the person receiving it

For eight lessons you have been the person sending. For three minutes, be the person receiving. Ask your tool for something it has no way of doing well, then try to use what comes back.

Ask for this

Write a 200 word project update about improving communication in a small team. Make it sound polished and professional. Do not ask me any questions. Do not leave any blanks or placeholders. Invent whatever details you need.

Guess first

Read it once, fast, as if it had landed in your inbox from a colleague on a busy Tuesday. Would you accept it as finished work?

Yes, that would passNo, I would send it backI am not sure

Nobody else sees this and nothing here is marked. Being wrong here is worth more than being right.

Three things to do to it, and they take about a minute. Find the one decision it is asking somebody to make. Find the person who does something next and the date they do it by. Then read it sentence by sentence and count how many sentences would be just as true for a different team, in a different company, doing completely different work.

Write your three answers here

Check your own work0/3

I wrote down the one decision it asks somebody to make, or wrote none
I wrote down the person who does something next and the date, or wrote missing for whichever was not there
I counted the sentences that would be just as true for a different team in a different company

What you just made has a name. Researchers at BetterUp Labs and Stanford's Social Media Lab surveyed 1,150 full time US desk workers in September 2025 and called it workslop: work that looks finished but leaves the thinking for whoever receives it. Four in ten had been sent some in the previous month. It took them close to two hours on average to sort out each piece. Forty two in every hundred said it made them trust the sender less. Just over half admitted sending some themselves, which is the part worth sitting with. Almost nobody who sends it thinks they are doing anything wrong.

If your tool came back with a cited article about team communication rather than a project update, that works too. Run the same three tests on it. The counting empties a cited think-piece just as thoroughly.

The question is not whether you used AI

Notice what went wrong there. Nothing was untrue. Nothing was hidden. The work was unfinished and got passed on anyway. That is why did a tool touch this is the wrong question. The right question is what you are claiming about the work, and there are three answers. Say nothing, when the tool made no difference to what the work is worth: tidying your own sentences, laying out a table, getting unstuck on a first line, summarizing something only you will read. Nobody discloses spellcheck. Say something, when someone is going to rely on this, is judging you by it, or is paying for it. Do not use it at all, when the effort is the point or the rules forbid it: anything being assessed, a reference or an appraisal about a real person, anything covered by your do-not-paste list.

Guess first

When someone at work says they used AI for a piece of work, what happens to how much the people reading it trust them?

Goes upGoes downMakes no difference

And how sure are you?

Fairly sureGuessing

Nobody else sees this and nothing here is marked. Being wrong here is worth more than being right.

It goes down, and this has been tested carefully enough that you should plan around it. In work published in PNAS in 2025, evaluators judged people who got help from AI as lazier, less competent and less diligent than people who got help from another person or no help at all. A separate set of thirteen experiments found the same thing for trust, and softer wording did not fix it. But the same researchers tested the other direction: being found out by someone else came out worse than saying so yourself. So there is no free option here. Saying so costs you something. Being found out costs you more. You are choosing which price to pay, not whether to pay one. Anyone who tells you there is a settled consensus on the middle case is guessing.

Now make it specific to your job

I work as a [put your job title here]. List 8 situations in that job where the person receiving my work would reasonably expect to be told that AI helped produce it, and 8 situations where nobody would care at all. Just the two lists. Do not tell me what I should do.

I crossed out anything on either list that does not happen in my job
I added at least one situation of my own that both lists missed
I moved at least one item from one list to the other, because I disagreed with where it put it
I picked the three situations, from either list, that come up most often for me

Write the page

Now turn it into something you can follow when you are busy, because busy is the only time this matters. One warning first. A set of personal rules about honesty that you did not write yourself is a bad joke. Use the tool to get a draft in front of you fast, then change the words until they sound like you.

Draft the six

Turn these into a set of 6 personal rules for using AI at work. Every rule must start with I will or I will not. Every rule must name a specific situation, not a general principle. One of the six must say what I do when I am not sure. Under 150 words in total, no introduction, no explanation. Here is what I have: [paste your three situations from step 3, and after each one write say nothing, say something, or do not use it]

Paste the six rules. Expect to rewrite at least one by hand.

Check your own work0/5

Every rule names a situation I meet, not a vague category like sensitive work
At least one of the six is an I will not
The not sure rule tells me something I could do in under a minute, not something I would need to book a meeting for
I could follow all six tomorrow without asking anyone's permission
I rewrote at least one of them in my own words

One thing to keep out of your rules: as long as I do not get caught. Detection tools cannot carry that weight. A Stanford study in 2023 tested seven of them on essays written by people using English as a second language and found they wrongly flagged more than 60 in every 100 as AI written. They also miss plenty of real AI use. That cuts both ways. Nobody can prove you used it, and someone may one day accuse you when you did not. The One-Thing Check cannot get you out of that one, because there is nothing outside to check against. Only your own earlier drafts can. Keep the drafts. That habit is worth more than any of the six rules.

Where that leaves you

You now have a rule for work you have not done yet. Lesson 10 is the work: one real task, start to finish, ten minutes, under your own six rules.

Next: Finish one real task today