What does a simple “thank you” cost? More than you might think – at least with ChatGPT. When millions of people speak politely to AI every day, it adds up to millions in costs for OpenAI. A viral Instagram reel gets to the heart of the matter and hits a raw nerve in the AI world. Because there is far more to politeness than good manners: Each of these little expressions consumes tokens that cost energy, water and ultimately money. What seems banal at first glance becomes a symbol of a larger tension between human tone and technical efficiency. This article looks at why this is relevant and what it says about us in our dealings with machines.

How many tokens do I need?

Every input in ChatGPT generates tokens, including a polite please and thank you. But what do these tokens actually mean? They are the basic unit with which OpenAI calculates the length of texts and thus the price of the programming interface. How many tokens an input requires depends on four aspects:

1. The number of characters
2. The text language
3. The punctuation marks and emojis
4. The GPT model used

This can lead to a text being shorter than the tokens used and therefore costing more than you might initially think. Punctuation marks usually require one token, while emojis require two to three. However, there are significant differences in text language. To give an example, in English a word uses one token, while in German a word needs 1.8 tokens and in Spanish even two tokens (OPENAI GPT TOKEN GUIDE – TOKENSYSTEM ERKLÄRT, n.d.).

In addition, the new GPT-4 model consumes fewer tokens than the previous version, thus saving energy and costs.

Now that you can count your tokens, the only remaining question is how much they cost.

The latest model GPT-4o charges five dollars for 1 million tokens through requests. To limit the costs, there is a maximum number that can be used per request, which is 128,000 tokens for the current model (OPENAI GPT TOKEN GUIDE – TOKENSYSTEM EXPLAINED, n.d.).

Experts therefore estimate the cost of a question to ChatGPT at 0.02 euros (Was Kostet ChatGPT? – Digital, 2023). That may not sound like much at first glance, but with 4.5 billion visits per month, it adds up (Kaiser, 2025).

This price is calculated on the basis of two key aspects:

• The consumption of kilowatt hours. As an example, ChatGPT requires 0.14kWh for a 100-word text, which is equivalent to the energy consumption of 14 LED lamps for one hour (“Please” and “Thank you” cost ChatGPT several million dollars, 2025).
• Water consumption to cool the data centers. The amount is estimated at 0.5 liters of water for a chat with 10 to 50 questions (Fett, 2024).

This means that even a polite please and thank you costs money, as it requires tokens, which in turn cost money due to the electricity and water consumption required. According to Sam Altman, CEO of OpenAI, these costs run into the millions just for kindness (Petereit, 2025).

Why are people polite to a computer?

Why do people even say please and thank you to a computer? A survey in 2024 found that two-thirds of ChatGPT users in the US are polite to their chatbot (Petereit, 2025). And there are two main reasons for this:

1. Out of habit
2. Out of fear of an AI takeover

But what difference does this behavior make to the answer, apart from significantly higher costs? ChatGPT is trained to respond in a similar tone as it receives the question. Therefore, a respectful interaction with the AI leads to an equally respectful response (“Please” and “Thank you” costs ChatGPT several million dollars, 2025).

Moreover, this politeness is a reflection of our attitude towards technology. Who “Please” and ‘Thank you’, the system not only treats you like a tool, but almost like a conversation partner. This may seem irrational, but it creates a form of relationship that influences our usage behavior.

Understandable from a human perspective, but an expensive habit in terms of server costs. Because the more friendly the world speaks to AI, the more tokens are generated and therefore the higher the costs. Whether this is sustainable in the long term or whether we will soon have to learn to communicate more efficiently remains to be seen. A balancing act between politeness and resource consumption.

This is how OpenAI can save money:

A costly problem like this, caused by the repeated use of polite phrases, could be mitigated by technical solutions. This would optimize the use of resources without compromising the quality of the interaction:

• Polite phrases such as “please” and “thank you” could be automatically recognized and ignored during token counting. In this way, the friendly tone of voice would no longer become a cost trap, at least not for OpenAI. However, this approach is not entirely without side effects. Language is nuanced, and what looks like politeness can carry an important meaning in context. If too much is filtered out, the quality of the response may suffer.
• A subtle user note in the interface could alert users to the problem, such as: “Polite phrases increase your token count slightly.”
• Intelligent prompt handling in the backend could recognize that it is a matter of politeness, but not technically incorporate it into the processing.

Ultimately, it remains a balancing act between efficiency and human expression. Because in the end, it’s not just about costs, but also about how we want to talk to machines.

Politeness has its price – at least when it is addressed to an AI. What we hardly notice in everyday life has a clear impact in the OpenAI system: Every “please” and “thank you” costs computing power, water, electricity and therefore money. And yet it would be too short-sighted to focus solely on efficiency and sacrifice human expression. Rather, the question is how we want to speak with AI in the future. Between resource awareness and social interaction, new solutions are needed, both technically and culturally. Because in the end, it is not just the algorithm that shapes interaction, but also how we decide to communicate with technology.

Conclusion: courtesy or efficiency?

Politeness has its price – at least when it is directed at an AI. What we hardly notice in everyday life has a clear impact in the OpenAI system: Every “please” and “thank you” costs computing power, water, electricity and therefore money. And yet it would be too short-sighted to focus solely on efficiency and sacrifice human expression. Rather, the question is how we want to speak with AI in the future. Between resource awareness and social interaction, new solutions are needed, both technically and culturally. Because in the end, it is not just the algorithm that shapes interaction, but also how we decide to communicate with technology.