Marcel HeadlamSay Please to AI. It's Not for the AI.
At the front of the message box, before everything you need to say, sit six letters that spell please. After them comes the text to your landlord that you want AI to word for you, and after that the cursor, blinking. You look back at the first word and wonder who it was for.
The software won't notice, because AI feels nothing. Nobody is hurt when you're short with it, and nobody is pleased when you're kind.
AI reads text in small pieces called tokens, so every "please" gives it a little more to read. One costs next to nothing, but millions add up to something.
People who leave the word out aren't being rude. They're treating a tool as a tool.
Immanuel Kant gave lectures in the 1770s, two hundred years before anyone typed to a machine, and he made a point about animals that fits this message box. His students wrote down what he said, "he who is cruel to animals becomes hard also in his dealings with men". The source is Kant's Lectures on Ethics, in Louis Infield's translation, as quoted by Linzey.
What Kant means. Kant didn't think we owe animals anything directly. His worry was what cruelty does to the person being cruel, because someone who gets used to being harsh to animals starts being harsh to people too.
Why it matters here. A habit doesn't stay where you put it, so the way you speak to AI could leak into the way you speak to people.
Kant meant living creatures, and stretching his idea to a chatbot is my own reading. But I think it holds, because his argument was never about the one on the receiving end.
How you speak, dozens of times a day, is practice, and practice travels. Imagine you type to AI in a way you'd be embarrassed to use with a colleague. Some of that tone will turn up in your next message to a person.
The software won't have minded, but you will have rehearsed.
None of this means saying please on every line, since nobody does that with a colleague either. Write the first line the way you'd say it to a person, and after that, be as brief as you like. "Shorter" and "try again" are fine.
Then watch for the days you stop bothering. The tone that slips when you're tired is the same tone that slips with people, and the chat window is a cheap place to catch it.
Once a month, at the end of a long chat, ask the AI to read your side back to you.
Look back at my messages in this chat, not yours.
If I had sent these to a colleague, how would my tone have come across?
Quote the two or three lines that would have come across worst,
and don't soften it. I'll decide what to change.
The AI can quote the lines, but it can't tell you what kind of person you want to sound like. And that was always the point of looking.
Whether manners change the answers
On that question, the research is mixed. A 2024 study tested prompts in English, Chinese and Japanese, where a prompt is the message you type to the AI. The study found that rude prompts tended to make the AI do worse, but being extremely polite did not reliably help (Yin and colleagues, arXiv 2402.14531).
In 2025 a short paper tested 50 multiple-choice questions on one AI, ChatGPT 4o. It found the rudest prompts scored a little higher, getting 84.8% against 80.8% for the most polite (Dobariya and Kumar, arXiv 2510.04950).
Both studies are small, and results differ between AIs, languages and tasks. So I won't claim that manners get you better answers. I also haven't found any research on whether being short with AI makes people shorter with each other.
Mine is an argument about habit, and nobody has measured it.
You send thousands of messages, and you might say typing a command is more like pressing a button than kicking a dog. So cutting out the extra words is only sensible.
You'd be right. Brief and rude are different things, and brief is all this asks of you.
Nobody will ever hear that first word. Who are you practising to be?