Should You Say You Used AI?

"Did you write this, or did the robot?"

A colleague says it half-joking, your report in hand, a moment after calling it good. You take a second, and the whole question lives in that second.

"Never admit it" is easier than the truth, and so is "always say". The truth depends on how the other person would feel if they found out later, and testing that takes about as long as your pause.

Do I have to tell them?

Ask yourself one thing, which is whether they would feel misled if they found out later.

Minding is different, because people mind plenty of things they've no right to mind.

Misled means they took something from the work that isn't true. Maybe they thought you spent a day on it, or that the words are the way you talk, or that the judgement in it is yours. Maybe they thought you could do this without help.

If none of that was part of the deal, say nothing. If any of it was, say so before they ask.

That pause is the sound of someone working out, too late, which kind of work it was.

When would they feel misled?

When the work is a test of you. That covers a job application, an exam, a portfolio piece or a writing sample.

They're asking what you can do, so an AI draft you didn't mention answers a question they didn't ask. Once they find out, they'll assume the rest was borrowed too.

When the people reading know your voice. Think of your newsletter, or a message to your team.

Readers who know you hear the gap, and to them it reads as "wasn't bothered". Ian Leslie's name for AI writing is a "double absence".

If you leaned on AI heavily for something personal, say so, or keep your name off it.

When the work is about someone else. That means references, eulogies, toasts and feedback on a colleague.

Those are gifts of attention, and people don't forgive finding out the attention was handed to a machine.

When a rule about AI already exists. It might be your employer's rule, or your university's, the journal's or the competition's.

The rule settles it. If you're unsure whether a rule exists, find out before you assume.

What if AI only tidied my draft?

Then the AI was an editor, and editors don't get their name on the work. Nobody declares a spellchecker.

If the AI fixed typos and suggested a clearer word, and the ideas and most sentences are still yours, then carry on.

The same goes when the result is the point, like the meeting summary or the spreadsheet formula. Nobody wanted to know you suffered over it. They just wanted it done.

A line at the bottom of every email saying "Written with the help of AI" is a nervous habit. That habit hands the reader the job of deciding whether the AI part matters, and that job was yours.

How do I say it without sounding guilty?

Say what the AI did, because saying "I used AI" tells them nothing and sounds like a confession.

"I had AI pull the figures together and I wrote the recommendation" tells them where you were, and most people hear skill in that.

Say it early and in passing, at the top of the report or in the first line of the reply. Say it before anyone asks, because owning up after the question feels different from owning up before it.

Be ready for someone to ask "Which bits?" Could you answer "the first draft and the summary table, but the argument and the numbers are mine"?

If you could, then you've shown your work. If not, the honest answer is "more than I'd like", and that's worth knowing about yourself.

When it's true, one line is worth keeping. "AI drafted, I decided."

Isn't this like declaring a calculator?

The best case for saying nothing is the calculator, and I take that case seriously.

Nobody mentions the calculator, the template, the colleague who read it over or three hours of Googling. Singling out AI makes it seem morally heavier than those other tools ever were. In a year or two, declaring AI may seem as odd as declaring a word processor.

People who argue this aren't hiding anything. They just refuse to treat a normal tool as a confession.

I think they're right about where we're heading, but wrong about now.

Most readers still assume a person wrote what carries a person's name. The gap between that belief and what really happened is exactly what the test measures. When the belief changes, so will the answer.

What if saying so costs me?

Some rooms punish honesty, and a workplace that looks down on AI use has a problem.

But it's your workplace, and you may choose to stay quiet there. That's a judgement, and a fair one, so keep it in that room.

Be honest, too, about what the reader assumed, because that's where people fool themselves most. If you catch yourself arguing with the test, you probably have your answer.

Think of the last thing you sent that AI touched. Could you say, in one sentence, which part was yours?