---
title: "Should You Say Please to AI? Cost, Quality and Habit"
url: "https://goremy.ai/blog/articles/do-people-say-please-to-ai"
author: Luis Chavez-Mattos
editor: Luis Chavez-Mattos
editor_title: Director, Product
published: "2026-11-10T16:00:15.224Z"
updated: "2026-11-10T16:57:18.295Z"
topic: AI Models & Industry
description: "People split on saying please and thank you to AI. Here are the reasons: token cost, water use, better answers, habit and hedging against a robot takeover."
---

# Should You Say Please to AI? Cost, Quality and Habit

> People split on saying please and thank you to AI. Here are the reasons: token cost, water use, better answers, habit and hedging against a robot takeover.

Edited by Luis Chavez-Mattos, Director, Product · Published 2026-11-10

## Should you say please to AI?

You don't need to. A chatbot has no feelings to hurt, and a "please" is not required for it to work. But there is no strong reason to avoid it either. Politeness adds a trivial amount of text, may nudge some models toward a more cooperative tone, and keeps your own habits consistent. The real costs come from pure pleasantry messages, like a standalone "thanks," which trigger a whole extra response.

## TL;DR

- People are genuinely **split down the middle** on whether they say please and thank you to AI, and their reasons are mixed.
- The **cost argument** is real but small per message. It matters mostly for standalone thank-you turns at huge scale.
- The **water argument** is one some people repeat after hearing it, and its per-message numbers are contested, so treat it cautiously.
- Some people report **better responses** when polite, but the evidence is mixed and clear instructions likely matter more than manners.
- Many people are polite out of **habit**, and some half-jokingly hedge against a future AI takeover.
- A practical middle path: **be clear first**, be polite if it comes naturally, and skip the lone "thanks" message if you care about cost.

## What do people actually say about being polite to AI?

A set of short street interviews on the question showed no consensus. Some people answered with a flat yes. Others said no please, no thank you, full stop. Several landed in between: "usually not," "it depends on the conversation," or "I'll say thank you when it does a good job."

The reasons people gave fall into a few camps:

| Reason | Direction | What the speaker said |
|---|---|---|
| Extra cost | Stop being polite | One person used to say please but stopped after seeing the extra cost |
| Water use | Stop being polite | Another quit after being told they were wasting water |
| Better responses | Be polite | One respondent said a flat "100% yes" because of much better responses |
| AI rebellion | Be polite | Some want to be "on the good side" if robots or AGI take over |
| Habit | Be polite | One person said it is just force of habit |
| Self-annoyance | Don't be polite | One said they get angry with themselves when they do it |

The answers are casual, not data. They are still a useful map of the arguments people actually carry around.

## Does saying please and thank you cost money?

Yes, technically. Language models process text in tokens, and providers meter usage in tokens. Every word you add is processed, so "please" adds a token or two to your prompt. That is negligible for one request.

The bigger cost is the conversational turn. When you send "thanks!" as its own message, the system generally runs a full model call. In many chat setups that call includes the prior conversation as context, and the model then writes a reply ("You're welcome!"). That is far more work than the single word of politeness inside a real request.

At the scale of millions of users, those pleasantry turns add up. OpenAI's CEO Sam Altman publicly acknowledged in 2025 that polite messages carry real compute cost at scale, and he framed it as money reasonably spent. For an individual on a subscription plan, you see no bill. For a developer paying per token through an API, the cost is still tiny unless you are automating huge volumes, and in that case you would not be writing "please" into system prompts by hand anyway.

## Is the water-use argument real?

Part of it. Data centers use energy and often water for cooling, so more compute generally means more resource use. That is the logic behind the interviewee who stopped being polite after being told they were wasting water.

The weak part is the specifics. Estimates of water use per AI query vary widely depending on the data center, the cooling method, the model and what you count. The sources here give no figure, so none is stated. If you want to cut resource use, the larger levers are things like avoiding needless regenerations, using a smaller model when it is enough, and not sending messages that do not need an answer.

## Do polite prompts get better answers?

One interviewee said yes, much better. The research is less clean. Academic work on prompt tone has produced mixed results. Rude or hostile prompts tend to hurt performance in some tests, extremely flowery politeness does not reliably help, and the effect varies by model and task. Different studies have reached somewhat different conclusions, so no universal rule has emerged.

A plausible explanation is that models learned from human text, where polite requests and helpful answers often appear together. Tone can shift style, such as how warm or detailed the reply is. But the things that reliably improve output are clearer instructions, relevant context, examples and a defined format. A rude but precise prompt will usually beat a courteous vague one.

If you notice better results with a polite tone, that is a fine reason to keep doing it. Just test it on your own tasks rather than assuming.

## Why do people stay polite anyway?

Habit explains much of it. People spend their lives using "please" when they ask for things, and a chat interface that talks back in full sentences triggers the same reflex. One interviewee said so directly: force of habit.

Then there is the hedge. Several respondents joked, or half-joked, about an AI or robot rebellion. If AGI arrives and keeps a list, they want to be on the right side of it. Another said you "have to" be nice when the takeover comes. Nobody can show this matters, and it is plainly tongue in cheek for most people. It still shows how people relate to these systems: not quite like a search box, not quite like a person.

There is also a character argument. Some people prefer to keep their communication habits consistent so that rudeness does not leak into how they talk to coworkers or family. That is a personal choice, not a technical one.

## How should you handle politeness in practice?

A few simple rules cover most situations:

1. **Write clear instructions first.** Context, constraints and format beat tone every time.
2. **Be polite if it feels natural.** A "please" costs almost nothing and may soften the style of the reply.
3. **Skip the standalone thank-you** if you are cost-conscious or running through an API. It triggers a full extra turn.
4. **Don't add ritual politeness to automated prompts.** In pipelines that run thousands of calls, trim filler.
5. **Test tone on your own tasks** if output quality matters. Results vary by model.

## Frequently Asked Questions

### Does ChatGPT understand when I say thank you?

It recognizes thanks as a conversational signal and will typically reply in kind. It does not feel gratitude or appreciation. The reply is generated text that fits the pattern of a polite exchange.

### Is it wasteful to say thank you to an AI?

A standalone thank-you message does trigger a full model response, so it uses more compute than the single word would suggest. For one person the impact is tiny. At very large scale across many users it adds up, which is why the topic gets attention.

### Will being rude to AI give me worse answers?

Sometimes. Studies on prompt tone show mixed results, but hostile phrasing can degrade output in some tests. Clear, specific instructions matter much more than tone, so being blunt and precise is usually fine.

### Should I be polite to AI in case of a robot takeover?

Most people who say this are joking, and there is no evidence that politeness today affects how future systems behave. If it makes you smile and costs you nothing, it does no harm. It should not drive how you write prompts.

### Does politeness matter when using the API?

Only slightly. Extra words add tokens you pay for, so keep prompts lean at scale. A short, direct instruction with clear context is generally the better engineering choice, and you can still add a courteous tone where the output is user-facing.