Tips I got from Google AI for saving energy on the servers when I ask that question.

I use AI like a personal tutor. Still I think about how much energy it uses, so I ask Google AI. It provided tips which I'll share.

You can significantly lower the server-side energy consumption of your desktop AI sessions by adopting a few highly effective habits:

1. The "One Topic, One Chat" Rule.
If you keep using a single, massive chat thread for weeks, every single new question you type forces the server to re-read the entire history from day one just to answer a simple question.

The Green Solution: Start a fresh chat thread every time you switch topics or finish a task. A short, clean thread takes a fraction of the computational energy to process compared to a sprawling, 20-page archive.

2. Condense and Prompt Efficiently.
Sending dozens of rapid-fire, single-sentence follow-up questions keeps the server constantly recalculating.

The Green Solution: Take an extra moment to bundle your thoughts into one comprehensive, well-structured prompt. Fewer server requests translate directly to less electricity used.

3. Archive Locally.
Clicking back into old threads just to read or reference information you talked about last week triggers the server-side context reloading process.

The Green Solution: Once a conversation yields a piece of code, a piece of writing, or a breakdown you want to keep, use the Export to Google Docs feature. Reading that data from Google Docs or a local text file uses virtually zero computational power compared to waking up an LLM.

4. Rely on Search Engines for Simple Facts.
Asking an AI "What is the capital of France?" or "When does the local post office close?" forces a massive neural network to fire billions of connections just to give you a static fact.

The Green Solution: For straightforward lookups, direct links, or simple facts, a traditional web search is significantly more energy-efficient than generating text via an AI model.

---

Now I'll add that I mostly use AI for text based research. I also gather that image rendering, music and other uses use quite a bit more power for the AI servers; especially video rendering.