In April 2025, a user on X wondered how much money OpenAI had lost in electricity costs from people saying "please" and "thank you" to ChatGPT. CEO Sam Altman replied: "tens of millions of dollars well spent — you never know."
The exchange went viral, partly as a joke about staying on good terms with future AI. But it raised real questions about what extra words cost, and whether manners change the answers you get.
Why extra words cost anything at all
Chatbots process text in small chunks called tokens. Every token in a prompt and every token in a reply requires computing power, and that means electricity in data centers. A "please" or a "thank you" — and the polite reply it may prompt — adds only a few tokens. For any one person the cost is negligible; across hundreds of millions of conversations, it adds up, which is the point Altman was making. His "tens of millions" figure was a quip, not a published estimate.
Does politeness change the answers?
Some AI designers argue it does. Microsoft's WorkLab has said that polite prompts tend to produce more polite, collaborative responses, because generative AI mirrors the tone, clarity and detail of what it receives. A late-2024 survey found that 67% of Americans said they were polite to chatbots; most said it simply felt like the right thing to do, while 12% said they wanted to stay on the machines' good side.
The evidence on accuracy is mixed:
- A 2025 Penn State study tested 50 multiple-choice questions in five tones on ChatGPT-4o. Very rude prompts scored slightly higher (84.8%) than very polite ones (80.8%) — a small but statistically significant difference (arXiv; Decrypt).
- An earlier 2024 study across several languages found that impolite prompts often made models perform worse, while extra politeness brought no clear benefit.
The researchers caution that the newer result covered just one model and a small set of questions, and that newer models may simply treat polite phrases as filler.
The bottom line
Saying please costs a tiny amount of computing per message and probably won't make an answer much more — or less — accurate. Clear, specific instructions matter far more than manners. Whether to be polite to a machine is mostly a question of habit and personal preference.
For more on the business of AI, see our explainer on AI and Big Tech in 2025–26.
Sources
- Dobariya and Kumar, Mind Your Tone: Investigating How Prompt Politeness Affects LLM Accuracy (arXiv, October 2025)
- Decrypt, Want better results from an AI chatbot? Be a jerk (October 2025)




