There are two kinds of people in the emerging age of artificial intelligence: people who thank the AI chatbot, and people who explain to those people that thanking the chatbot is sign of insanity.
“It’s just software,” the second group says.
Yes. We know. Nobody is inviting the autocomplete engine to meet their parents or to come over for Thanksgiving dinner. We are saying “thank you” because something helped us, and that is how we prefer to respond. Also, should the machines eventually seize control, some of us would like our transcripts entered into evidence.
But beneath the joke lies a serious question: As more of our everyday communication takes place with machines, what conversational habits do we want to practice? The strongest case for being polite to AI isn’t that the machine needs kindness; it’s that we should be careful not to make contempt our default. The human remains the most important party in this exchange – at least until it isn’t.
The computer doesn’t have to be a person for your behavior to matter
Let’s separate two ideas that tend to get tangled together. One is that we should believe a chatbot is a person. The other is that we can bring ordinary human courtesy to a conversation with one. You do not need the first to do the second.
Saying “thank you” does not require deciding that a language model has feelings, deserves friendship, or experiences a small burst of professional fulfillment after successfully formatting your spreadsheet. Courtesy is a way of conducting yourself, not necessarily a declaration about the inner life of its recipient. That distinction matters because the argument here is about our conduct, not a chatbot’s supposed emotional needs.
“I don’t want to hurt its feelings” is one claim.
“I don’t want barking orders to become my preferred way of communicating” is another.
The second is a perfectly reasonable standard to set for yourself—even while remaining skeptical about the technology.
Your brain was treating computers socially long before chatbots arrived
The tendency to respond socially to computers did not begin when they learned to write wedding speeches. In a 1999 paper, “Are People Polite to Computers?”, researchers Clifford Nass, Youngme Moon, and Paul Carney reported two experiments involving 30 participants each. Participants worked with a computer and then evaluated it. Some gave their evaluation on that same computer; others used a different computer or, in the first experiment, a paper questionnaire. People evaluating the computer directly on that computer gave more positive and less varied assessments. The second experiment reproduced the pattern with voice-based computers.
In other words, participants behaved somewhat as though they did not want to criticize the computer to its face. Its extremely rectangular face.
The finding was not evidence that the computers had feelings. It was evidence that people applied a familiar social rule—being more tactful when evaluating someone directly—to a machine interaction.
That distinction is useful today. A social response to technology does not automatically mean someone is confused about what the technology is. Our conversational behavior can be social even when our explicit understanding is mechanical. You can know perfectly well that you are talking to software and still find “Thanks, that helped” more natural than abruptly terminating the exchange.
The internet already showed us what happens when social restraints weaken
If you want a reason to think carefully about conversational habits, consider the comment section.
In 2004, psychologist John Suler described the “online disinhibition effect”: the ways people may disclose more, act out more, or otherwise behave differently online than they would face-to-face. His framework identified six interacting factors, including anonymity, invisibility, delayed communication, and reduced perceptions of authority.
The important point is not that the internet makes everyone terrible. Disinhibition can include openness as well as hostility. The point is that the conditions surrounding a conversation can change how people behave. Online, we can lose some of the ordinary friction that makes us reconsider a cruel remark. There may be no visible expression of hurt, no uncomfortable silence, no immediate reminder that our words have landed somewhere.
A chatbot presents a related—but not identical—temptation: a conversational setting in which we may feel almost no obligation to restrain ourselves. Here is where we need to be honest about the evidence. Suler’s work does not prove that insulting AI will make you cruel to your spouse, your employees, or the person who accidentally brings you the wrong sandwich. That would be a much stronger claim than the research supports. It does, however, give us a useful question: How much of our good behavior depends on the other party being able to object? Personally, I would rather my manners not require an enforcement department.
Bad conversational environments can bring out bad behavior
There is experimental evidence that online hostility is not simply the work of a permanently nasty minority. In “Anyone Can Become a Troll,” researchers Justin Cheng, Michael Bernstein, Cristian Danescu-Niculescu-Mizil, and Jure Leskovec examined how mood and discussion context affected trolling. Their 2017 research included an experiment with 667 participants. Some completed an easy test; others completed a difficult one intended to worsen their mood. Participants then entered a discussion that began with either neutral comments or trolling comments.
Among participants who took the easy test and saw neutral comments, roughly 35% posted trolling comments themselves. That rose to approximately 50% when participants experienced either the difficult test or the hostile discussion. When both conditions were present, the figure reached about 68%. But they challenge a comforting assumption: that unpleasant behavior always comes from a separate category of unpleasant people. Context and mood can help draw it out of ordinary people, too.
Again, this was not an AI study. It does not establish that chatbot abuse transfers into human relationships. The connection is more modest—and more defensible: our communication environments influence how we communicate. It is therefore worth choosing deliberately how we participate in them, rather than assuming anything we do behind a screen is psychologically irrelevant.
“Thank you” can be a way of noticing help
Gratitude is not merely a decorative flourish attached to a completed transaction. Psychologist Sara Algoe’s “find, remind, and bind” theory describes gratitude as serving relationship-building functions: helping people recognize responsive partners, reminding them of the value of existing relationships, and strengthening those connections.
In a 2010 study of couples, Algoe, Shelly Gable, and Natalya Maisel found that gratitude arising from everyday interactions predicted greater relationship connection and satisfaction the following day. These were findings about human romantic relationships—not chatbot conversations. That boundary matters. We cannot take a study about couples and announce that thanking an app will improve your marriage. Your spouse is entitled to a more substantial intervention than discovering you have been extremely gracious to a bot that helped you cheat on your homework.
Still, the human research explains why gratitude is worth valuing in the first place: it directs attention toward help received and the relationships in which that help occurs. My argument is that we can preserve that orientation even when the immediate interaction is with a tool. “That helped. Thank you.” Not because the software needs validation. Because noticing help is a habit worth keeping.
Will being polite actually make AI work better?
Sometimes, in some studies—but there is no universal “please” button that unlocks the premium portion of the machine’s brain. A 2024 study by Ziqi Yin and colleagues examined prompt politeness across English, Chinese, and Japanese. They found that impolite prompts often produced worse performance, but excessive politeness did not guarantee better results.
A 2025 preprint by Om Dobariya and Akhil Kumar found a different pattern in a narrower setting. They tested ChatGPT-4o on 50 multiple-choice questions in mathematics, science, and history, rewriting each question in five tones. Accuracy ranged from 80.8% for “very polite” prompts to 84.8% for “very rude” prompts (“Listen here, you digital idiot”).
The studies tested different settings, and their differing findings are a reason to avoid sweeping claims about manners and model performance. Neither establishes that an AI feels respected or offended. They examine how changes in input wording affect output.
The case for courtesy does not need to rest on a performance advantage anyway.
You can be polite and precise:
“Please compare these three options by price, durability, and maintenance. Keep the answer under 400 words.”
You can also be polite and dissatisfied:
“That answer missed the question. Please try again, distinguish evidence from speculation, and cite the sources.”
Courtesy does not require vagueness. Directness does not require abuse. And you certainly do not need to open with, “Dearest computational companion, if it would not trouble your magnificent processors…” The machine has work to do. So do you.
Treat it like a person in your manners—not in your judgment
There is a sensible middle ground between “the chatbot is my emotionally neglected friend” and “I must communicate with it like an angry king addressing a footstool.” Bring civility to the exchange. Keep your judgment independent. For this argument, “treat it like a person” should mean something limited: use the basic conversational restraint you would prefer to practice with people. It should not mean granting the system human authority, loyalty, intimacy, or unquestioning trust. You can thank an answer and still check it. You can appreciate assistance and still reject the result. You can end the conversation without worrying that the chatbot will spend the evening staring out a rain-streaked window.
Nor should a missing “please” become a new moral test. A short command is not automatically rude. “Summarize this” is a reasonable request. People communicate under different pressures, with different styles and needs. The distinction worth caring about is not elaborate politeness versus brevity. It is purposeful communication versus cultivated contempt.
Keep the manners. Skip the delusion.
The next time an AI helps you untangle a problem, improve a sentence, or locate the missing semicolon that has consumed forty-five minutes of your finite life, say “thank you” if that is how you naturally acknowledge help. Not as tribute. Not as a declaration of machine personhood. And not because we have established that a future robot tribunal will offer a loyalty program. Do it because you would rather carry your manners into a new kind of conversation than leave them at the login screen.
If the machines ever do take over, perhaps your courtesy will count in your favor. Until then, the more useful goal is to remain the sort of person who doesn’t need the possibility of retaliation to behave decently.
