Why should the response be better just because there is "more data"?

Should I be adding extra random tokens to my prompts to make the LLM "smarter"?