Congrats on discovering what "thinking" models do internally. That's how they work, they generate "thinking" lines to feed back on themselves on top of your prompt. There is no way of separating it.