Artificial Intelligence
Use artificial intelligence responsibly!
- First, check whether you really need AI for your task. Avoid making unnecessary AI enquiries.
- You might be able to find the answer to your question yourself, or a simple internet search might be all you need. If each of us replaces one AI query with a traditional search query every day, we can save around 0.1 kg CO2e per person per year (Kompetenzzentrum Nachhaltiger Konsum, 2026). With 14,000 university staff and students, that amounts to around 1.4 t CO2e.
- Do you rely on AI for help? If so, think carefully about how to phrase a prompt so that you receive the information you want as directly as possible. Be specific and tell the AI exactly what you want it to do. This will help avoid unnecessary follow-up queries and additional AI requests.
Using AI responsibly
Thanks to the rapid advances in artificial intelligence (AI), acquiring knowledge quickly now seems to be just a prompt away. Most users are now well aware of the typical pitfalls of AI, which sometimes takes a rather loose approach to the truth. However, many users are less familiar with the resource consumption associated with using AI.
Every prompt requires computing power, which in turn consumes energy and results in greenhouse gas emissions. At the same time, AI is becoming increasingly efficient: a study by Google shows that the energy consumption of a typical Gemini text prompt has fallen by a factor of around 30 within a year (Elsworth et al., 2025). In May 2025, such a prompt required an average of 0.24 Wh and produced 0.03 g CO2e. That amount of energy is enough to watch television for about 10 seconds.
Of course, it often doesn’t stop at just one prompt. Furthermore, actual energy consumption depends heavily on the AI model used and the specific task at hand. Many providers of AI tools, for example, offer a special mode that can be activated, in which the AI thinks about a problem for longer and more thoroughly or conducts more extensive research to generate a more accurate answer. When the AI is used in this ‘reasoning’ mode, energy consumption can quickly increase by more than a factor of 10 (Oviedo et al., 2026). The generation of images and videos, particularly long or complex queries, the uploading of attachments, and the use of voice control and output can also be significantly more resource-intensive.
In addition to energy consumption, the water consumption of data centres also plays a role. However, when breaking down the cooling of data centres to the water consumption per AI interaction, the scientific literature reveals widely varying results, ranging from 0.26 ml (Elsworth et al. 2025) to 10–50 ml (Li et al., 2025) per AI response.
Training large AI models also requires considerable resources, which are often not taken into account in such calculations. For the training of GPT-3 alone, for example, total water consumption was estimated at around 5.4 million litres (Li et al., 2025).
With the number of AI queries rising sharply worldwide, even small amounts of consumption per query can quickly add up to a very significant total. It is therefore worth using AI in a targeted manner – rather than automatically generating a prompt for every minor query.
Keep reading on
Elsworth et al. (2025): Measuring the environmental impact of delivering AI at Google Scale, https://doi.org/10.48550/arXiv.2508.15734.
Oviedo et al. (2026): Energy use of AI inference, efficiency pathways, and test-time scaling, Joule, 10, 8, https://doi.org/10.1016/j.joule.2026.102430.
Li et al. (2025): Making AI Less 'Thirsty', Commun. ACM, 68, 7, https://doi.org/10.1145/3724499.
Kompetenzzentrum Nachhaltiger Konsum (zuletzt aufgerufen 30.09.2026), https://nachhaltigerkonsum.info/klimawaage-annahmen-berechnungen-quellen#strom