Hi @RandomGuy ,
The model treats the conversation history as a pattern to follow. Sometimes pattern is interrupted and then this happens .
If you are using the Gemini API or AI Studio, you should place the formatting rule in the System Instruction block, not the user prompt. System instructions carry a higher weight and persist throughout the context. or If you are using normal chat interface ,Try reminding the model your rule every few turns like : “Remember to format every response with [title]: before the answer.” or else use Few-Shot Prompting technique .
Please try these methods and let me know if they work as expected.
Thank you.