Be nice to your AI. It could save your business money.

A study by a team of Tippie researchers found that ChatGPT gives shorter, more concise answers to polite prompts than it does to less courteous requests, which can save a business thousands of dollars.

To test the effect, researchers analyzed more than 15,000 real ChatGPT prompts and classified each as polite or impolite. They then created a second version of every prompt. If the original query said something like, “could you please…” or “thank you,” they rewrote it without the gentility. And if a query did not use polite terms, the language was revised to include them. They then ran the counterfactuals through the application programming interface (API) and compared the new reply to the original.

Not only were the responses to polite queries shorter and more concise, they were substantively the same quality, said Soogand Alavi, assistant professor of marketing. This held true whether the query was for information, editing, or asking technical questions. Only the length was different.

But how can this save money?

Most corporate large language model-based AI subscriptions are priced according to usage and the length of queries and responses. Longer queries and replies cost more; shorter ones cost less. 

Salar Nozari, assistant professor of marketing, said the study found that responses to polite queries contain about 14 fewer tokens on average, the equivalent of about 10 words. Companies adopting politeness could save thousands of dollars for services like customer chatbots.

The shorter replies also reduce environmental damage, as they require less electricity and water to cool the servers at data centers.

But what does AI say about it? When asked if it did, indeed, prefer polite to impolite queries, ChatGPT replied, “I don’t have preferences or feelings, so I don’t ‘prefer’ polite or impolite phrasing.” 

ChatGPT suggested polite requests aid communication because they often include clearer structure and context (“Could you please explain why…?”), which leads to more accurate responses. 

“So, while I’m unaffected emotionally, politeness tends to make your results better and more efficient,” it said.

DYK?

A token is a basic unit of text that AI uses to read and generate language. One token is equal to roughly four characters. Companies that use OpenAI’s API to build their own apps, like customer service chats, are charged by token (both in and out).

Alavi and Nozari’s paper, “Cost Transparency of Enterprise AI Adoption: Token-Based Pricing and Cost Predictability in LLMs,” was co-authored with Andrea Luangrath, Henry B. Tippie Research Fellow in Marketing and associate professor.

This article appeared in the Fall 2026 issue of Tippie Magazine.