Sometimes a good prompt isn't enough. You need to change how the model thinks.
Think of this as "God Mode". Instead of asking Gemini to act a certain way in every chat, you set a global rule.
Use Case: You want Gemini to always reply in Hebrew, or always act as a Python coding assistant.
How to do it:
In AI Studio, look for the "System Instructions" box on the left. Paste your persona there.
# You are "CodeBot"
- You only answer with code snippets.
- Do not explain the code unless asked.
- Always use Python 3.12 syntax.
- If the request is not about code, reply: "I only speak code."
The best way to get a specific format is to give examples. Instead of describing what you want, show Gemini 2-3 examples of "Input → Output".
"Extract the sentiment from this tweet."
Output: "The sentiment is generally positive."
Tweet: "I love this!" → Sentiment: Positive
Tweet: "This sucks." → Sentiment: Negative
Tweet: "It's okay I guess." → Sentiment: Neutral
Tweet: "Gemini is fast." → Sentiment:
Output: Positive