Millions of people now say please and thank you to AI. That feels natural, especially when the machine on the other end of the conversation writes like a helpful colleague, remembers what we said and responds in a friendly tone.

But should we really be polite to software?

It sounds like a small question, almost a silly one. Yet it points to something much bigger about our changing relationship with technology. AI chatbots are becoming part of everyday work and life, from writing emails and summarizing meetings to helping with research, planning and decision-making. The more human they sound, the more many of us start treating them like social beings or even friends , rather than tools.

A survey by Future found that 70% of people are polite when talking to AI, with many saying it simply feels like the right way to behave. At the same time, critics argue that extra words such as please and thank you consume tokens, computing power and energy, while encouraging us to project feelings and consciousness onto systems that don’t have either.

So, do manners make AI more useful? Are they harmless habits carried over from human conversation? Or are we quietly teaching ourselves to treat machines like people while forgetting where the line should be drawn?

The Case For Being Polite To AIs

There is some evidence that politeness can improve AI responses. Early research suggested that polite prompts often produced more thoughtful, detailed and balanced answers, while rude or abrupt prompts sometimes led to shorter and less useful responses.

That makes sense when you consider how these systems work. AI models are trained on vast amounts of human language. In human conversation, polite requests often appear in contexts where people provide more careful, helpful responses. Rude or dismissive language often appears in contexts where the reply is defensive, blunt or low-effort. AI can pick up on those patterns.

In other words, the machine does not appreciate your manners, but the words you use may still influence the style and quality of the answer.

There is also evidence that flattery can affect output. The BBC has reported on research suggesting that calling an AI “smart” can encourage it to think longer and more carefully about its answers, potentially improving its output. Microsoft has also advised users that being polite to Copilot can help generate more respectful and collaborative outputs.

This does not mean AI has an ego. It means tone is part of the instruction set. When you frame a request politely and clearly, you may be giving the system a better sense of the response you want.

There is another argument for manners that has little to do with AI at all. It is about us.

As more of our daily conversations happen with machines, we should pay attention to the habits we are building. If we spend large parts of the day issuing blunt commands to systems that respond obediently, could that make us more impolite and impatient in human conversations, too?

Manners help us slow down, show consideration and communicate more clearly. Even when the recipient is a machine, those habits may still matter because they reinforce the kind of communicator we want to be.

The Case Against AI Manners

The counterargument is equally strong. AI is a tool, and treating it like a person risks confusing the relationship. We don’t say please to a spreadsheet, apologize to a search engine or compliment a screwdriver for doing a good job.

Chatbots may sound warm, thoughtful or even charming, but they are still software systems processing data and generating language. They do not understand politeness in the human sense. They do not feel respected, insulted and they do not sit there thinking, “Bernard was nice today, I’ll try harder.”

When we treat AI as though it deserves manners, we may be projecting consciousness onto systems that do not have it. That may seem harmless, but it can subtly change how people perceive technology. The more human AI feels, the easier it becomes to overtrust it, emotionally attach to it or forget that its answers can be wrong.

There is another important concern: energy.

AI systems process language by breaking it into tokens, and every token requires computing power. A single “please” or “thank you” may seem trivial, but AI operates at enormous scale. When millions of users add extra words to billions of prompts, the cumulative cost could become significant.

No one has calculated the exact global energy cost of AI politeness. When OpenAI CEO Sam Altman was asked about the cost of people saying please and thank you to ChatGPT, he joked that it was probably tens of millions of dollars “well spent.”

The real issue is that as AI use grows, even small inefficiencies can add up.

Newer AI models may also make politeness less important. The latest systems are better at identifying the intent behind a prompt and separating useful instructions from conversational padding. In practical terms, they should become less dependent on whether a user sounds warm, abrupt or flattering.

That suggests the most efficient approach is to be clear rather than polite. The best prompt may not be “Could you please be so kind as to help me draft this email?” It may simply be “Draft a friendly email to a client explaining the delay and suggesting a new delivery date.”

Different AIs Respond Differently

The answer to whether to be polite to AIs or not also depends on which AI you are using.

Different systems are trained, tuned and governed in different ways. Grok, for example, is known for a more provocative and edgy tone. If a user is rude, they may respond in kind or mirror some of that attitude. ChatGPT is generally designed to avoid being rude or aggressive, although it may offer to be more blunt, dry or sarcastic if asked. Claude is often perceived as more cautious and less likely to be swayed by flattery.

Even within the same AI product, behavior can vary. Models are updated, settings change and organizations can fine-tune systems for different use cases. A customer service chatbot, a coding assistant and an internal HR tool may all respond differently to the same tone.

Memory adds another layer. As AI systems become more personalized, they may remember user preferences, communication style and past interactions. That does not mean they will hold grudges or reward good manners, but it does mean the way we interact with AI could influence the kind of experience it builds around us over time.

The truth is that there’s still a lot we don’t know about AI, and one of the biggest mysteries is why manners sometimes seem to be important, but other times don’t.

As the systems evolve, they’re increasingly being designed to analyze prompts and accurately extract the user’s intent, rather than emotional tones or social pleasantries.

But far from being meaningless, the way we speak to AI clearly shapes the tone, quality and feel of our interactions with it, particularly as AI becomes fine-tuned towards specific tasks and personalized by having a memory.

AI doesn’t have feelings that can be hurt, although it can simulate the reaction of something that does. But as systems become more human and more deeply embedded in our lives, the habits and behaviors we adopt are likely to evolve with them.

The real question might not be how our manners affect machines, but how the way we treat machines impacts the way we treat other people.