You've probably typed a "thanks" at the end of a conversation with an AI, then felt a slight sense of absurdity. Or conversely, you find it ridiculous to say hello to software. Both reactions are reasonable, and the question is more interesting than it seems: it touches on what we become when we interact with these tools.
Let's start by setting aside the awkward question: no, an AI doesn't feel anything when you thank it. It doesn't take offence when you're curt, nor does it rejoice when you're kind. That's not the point. The point is what it does, and to whom.
Does it change anything technically?
A little, in a counter-intuitive way. A model produces the most likely continuation of the text it receives, as we explained in our article on LLMs. The register of your message therefore steers the register of its response.
A carefully worded request, in full sentences, tends to yield a more developed and structured response. Not because the model rewards your politeness, but because in the texts it was trained on, a polished exchange begets another. Courtesy isn't a magic word; it's a register signal.
That said, the effect is modest. What really matters is clarity and context, as we detailed in our article on phrasing requests. A "please" has never rescued a vague instruction.
An added word means one more token, and therefore one more computation. At an individual scale, it's perfectly negligible, and making anyone feel guilty over a "thanks" would be ridiculous. At the scale of hundreds of millions of users, the accumulation becomes measurable, which ties into the energy consumption questions we've covered. It's a real argument, but it doesn't weigh heavily against what follows.
The real question is about us
Here's the heart of the matter. We spend more and more time phrasing requests to conversational interfaces. These exchanges resemble conversations, unfold in natural language, and trigger the same social reflexes as our human interactions.
Two positions clash, and both hold up.
The mental hygiene position. Habits transfer. Someone who spends hours barking curt orders at a machine that obeys without flinching is, a little, training themselves into a certain kind of relationship. It's not the machine you're protecting by staying courteous; it's the groove you're carving. You don't behave badly without it affecting you.
The lucidity position. Treating a tool like a person fosters a confusion with its own dangers. We saw in our article on AI companions that attachment to these systems can become problematic. Keeping a clear distance, including in tone, would be a way of reminding yourself what you're addressing.
A third way
There may be a fairer position than either of these. Politeness toward a machine is neither a category error nor a moral duty: it's a matter of personal comfort, like tidying your desk or greeting a self-checkout machine.
Some find it natural to speak politely, regardless of who or what they're addressing, because it's their way of being and they don't want to change it depending on the recipient. Others find it more honest to be direct with a tool, and that says nothing about their relationship with humans. Both attitudes are defensible, and it would be absurd to turn it into a matter of principle.
What does deserve attention, however, is the moment the distinction blurs: when you start expecting something from the machine in exchange for your courtesy, or feeling guilt for not having shown it. At that point, it's no longer a habit; it's a confusion.
What to take away
Say thanks if you like; don't if it feels foolish. Neither option is a fault, and no one should lecture you on it.
What matters more is knowing at all times that you're talking to a system that produces probabilities, whatever the apparent warmth of its responses. That lucidity is useful. Politeness, for its part, concerns only you. And there's something rather reassuring in the fact that so many people ask the question: it means the reflex of consideration persists, even when no one demands it.