Do you tend to be courteous with AI tools, throwing in a “please” or “thank you” while chatting with ChatGPT?
If so, you’re not alone, but your politeness might be adding to OpenAI’s bill.
Sam Altman, the chief executive of ChatGPT’s owner OpenAI, has revealed that expressions of politeness such as “please” and “thank you” are costing the company tens of millions of dollars.
Although the idea may come across as amusing, there is a technical reason behind the high price tag.
In this explainer, we break down why polite language is proving expensive for the company, what exactly Sam Altman had to say, and whether it is time to change how you talk to AI.
How ‘please’ and ‘thank you’ to ChatGPT are costing OpenAI millions
It started with a question on X by user @tomieinlove: “I wonder how much money OpenAI has lost in electricity costs from people saying ‘please’ and ‘thank you’ to their models.”
OpenAI boss Sam Altman responded by saying that such polite phrases were costing the company “tens of millions of dollars.”
Still, he called the cost “well spent,” and even hinted that the actual figure could be higher, adding, “You never know.”
Impact Shorts
More ShortsSo, why is simple courtesy proving so costly?
Each polite message requires a fresh reply from the AI. Generating these responses in real time uses powerful computing systems, which consume large amounts of electricity.
That electricity powers not only the data processing but also the cooling systems that keep the servers from overheating.
According to Goldman Sachs Research, energy use is climbing due to the heavy demands of generative AI models like ChatGPT.
“The primary driver of this increased power demand is the growing computational intensity of AI workloads, particularly those related to generative AI and large language models,” their study said.
With huge volumes of data being processed and stored, energy consumption is rising fast. By 2028, AI is expected to account for roughly 19 per cent of all data centre electricity use.
Goldman Sachs estimates that each query to ChatGPT-4 uses around 2.9 watt-hours of electricity, nearly ten times the energy of a typical Google search. With over a billion requests made every day, that is about 2.9 million kilowatt-hours daily.
For premium users, who pay based on tokens, essentially the number of words, those extra “thank yous” may be making a noticeable dent in their monthly bill.
Should you stop using polite words with ChatGPT?
While OpenAI may be footing a large bill for small courtesies, Sam Altman never suggested users should stop being polite.
He said, “Tens of millions of dollars well spent…”
A survey from 2024 found that 67 per cent of Americans say they use good manners with AI because it still matters, even with machines. More than half said it is simply “the right thing to do,” while 12 per cent admitted they are just playing it safe in case of an AI uprising.
Though saying “please” to a chatbot might feel pointless, experts say such habits help set the tone for smoother, more useful interactions.
Microsoft’s WorkLab memo notes that it is not about AI appreciating the gesture. Instead, polite prompts encourage “collaborative” responses from the system.
Kurtis Beavers, a design director for Microsoft Copilot, explained in the memo: “Using polite language sets a tone for the response.”
“Doing so not only ensures you get the same graciousness in return, but it also improves the AI’s responsiveness and performance,” he added.
How social media reacted to Altman’s disclosure
Sam Altman’s revelation that polite exchanges with ChatGPT are racking up tens of millions in electricity costs set off a mix of jokes and genuine concern on the social media platform X.
“That’s a small price to pay to feel safe, when the apocalypse comes,” one user wrote.
Another questioned, “Is being polite to the AI keeping the coming apocalypse at bay?”
A more practical suggestion came from a third user: “Can we put a thanks button like thumbs up in chat? With no response from ChatGPT…It helps user continue to be polite and additional response generation may also be avoided in process.”
“Is trained on user input, the ‘thank you’ is positive feedback that helps it know when it was doing a good job and could affect the weightings of the new model. Not wasted,” a fourth user wrote.
)