The context length includes all of GPT4's replies, is that correct? If I want to minimize the context length, should I be asking GPT4 to limit its replies to ju

The context length includes all of GPT4's replies, is that correct? If I want to minimize the context length, should I be asking GPT4 to limit its replies to just the fix that is requested? GPT4 seems to prefer long-winded replies, i.e., when I ask for what amounts to a one-line fix, it repeats the entire block of code with the one line correction. In contrast, Replit's Ghostwriter often gives a concise reply showing only the one-line fix, and I have to ask for the entire block when it isn't clear where the fix is applied.