Large language models (LLMs) are becoming deeply embedded in everyday human-AI interaction. However, the environmental costs associated with using those tools remain elusive. Here, we address the question of whether providing users with information about the environmental costs of using LLMs tempers their use. Such information highlights an ethical tension: acting pro-environmentally requires people to forgo some of the benefits AI tools provide. We study how people solve this trade-off in tasks that are relatively easy to tackle without using LLMs and further assess if the impact of providing such information has a stronger effect on people who have stronger environmental concerns. In two preregistered experiments (total N = 1200), we studied participants' use of ChatGPT in a text-editing task after they learned about the tool's environmental costs. Results revealed that (i) raising awareness of ChatGPT's environmental costs reduced its use and usage volume; (ii) interacting with a sycophantic model, one that encourages continuous use, did not attenuate the awareness effect, although it led to greater overall use; and (iii) raising awareness was especially effective among people with high environmental concerns. The findings highlight how raising awareness concerning LLM's environmental costs can encourage responsible consumption.
