We all know the type: the programmer who prides himself on efficiency and keeps his prompt history as tidy as a sock drawer. "Explain the tax rule." "Rewrite this, shorter." "No. Again." Not one please, not one thank you; on principle, he will not waste keystrokes on a text predictor. Then one evening his phone buzzes, the chatbot confirms the dentist appointment it has just booked for him, and out it comes without a flicker of irony: "thanks so much!!" The blunt operator and the grateful one are the same person, and he is my favorite small comedy of the moment.
That small contradiction is now a mass phenomenon. In a late-2024 survey by Future, the publisher behind TechRadar, 67 percent of American users and 71 percent of British ones said they are routinely polite to AI. Of the polite Americans, 82 percent said they do it because it is simply the nice thing to do, machine or no machine. The remaining 18 percent are hedging against a robot uprising, which I find both absurd and, on reflection, no sillier than touching wood.
For a while there was a respectable practical excuse for the courtesy. A 2024 cross-lingual study from Waseda University found rude prompts often produced worse answers, and Microsoft's own public WorkLab guidance said a polite tone "sets a tone for the response." Say please, get a better paragraph. Tidy, self-interested, easy to defend at a dinner party.
That excuse has now collapsed from both ends. In October 2025, two researchers at Penn State ran fifty questions through ChatGPT in five tones, from very polite to abusive, each run ten times, twenty-five hundred prompts, in a single, not-yet-peer-reviewed study. The rude versions scored 84.8 percent correct; the very polite ones, 80.8. The reason was not that the machine enjoys being bullied. It was that "would you be so kind as to consider the following" is clutter, and "solve this" is not. Clarity won, not cruelty. Then, in December, a study across GPT, Gemini, and Llama found the effect had all but vanished on the newest models: tone moved the answer by under two percentage points, and on Gemini not at all. The machine has stopped listening to our manners entirely.
So here is where we stand in 2026. Politeness to a chatbot earns you nothing on the page. It does cost something. Sam Altman, asked last April what all those pleases were doing to OpenAI's electricity bill, replied that it was "tens of millions of dollars well spent, you never know," an off-hand estimate OpenAI has never audited. A single reply burns roughly ten times the power of a web search, by Epoch AI's estimate, with a billion of them a day. Every thank-you is a few more tokens. The habit is now pure expense with no return.
Which is exactly the situation in which manners become interesting.
I have spent thirty years arguing that form is a technology for kindness, not a substitute for it, and that the point of knowing the rules is so you can stop thinking about them and attend to the person across the table. The chatbot is the first companion in history who is not across the table, who has no feelings to spare and cannot be wounded by a curt "no, again." Every reason I have ever given for good manners fails here. That is the whole point. When courtesy can do nothing for you and nothing for the other party, what remains is simply the sort of person you have decided to be when nothing at all is keeping score.
The pragmatists are right about the keyboard. If you want the cleanest answer, drop the please, write "explain photosynthesis," and spare the planet the tokens. I will not scold anyone who treats a search box like a search box. That is not rudeness, it is grammar. The mistake is thinking the keyboard is the whole question. Ben Wood of CCS Insight put the worry plainly: if it becomes acceptable to bark at the assistant, "that behaviour will start to leech into interpersonal interactions, and that's a slippery slope." Manners are a muscle. You do not get to keep a warm one for people and a cold one for machines. You get one, and it hardens toward whatever you practice most.
So here is my ruling, and it is not the one you would expect from me. Be as blunt as you like with the machine while you are working. But if a thank-you slips out at the end, the way it does for our efficiency purist, do not correct yourself. That reflex is not naivety. It is the last honest witness to how you treat the things that cannot answer back. Keep it.
