Positive instructions tell AI what to do. Negative examples tell it what to avoid. Both together produce dramatically better output than either alone.
When you get an AI output that's close but not right, the fastest way to fix it is to identify the specific thing that's wrong and name it explicitly. Instead of "make it less salesy," try: "Do not include phrases like 'I'd love to connect,' 'circle back,' or 'reach out.' Do not list features. Do not end with a question."
This technique is especially powerful for cold email, LinkedIn content, and anything where generic AI output stands out immediately as AI-generated. The more specifically you can name what you don't want, the more the output will sound like a human wrote it.
Examples that work well in sales contexts: