The "persona" prompt is dead. Here is how to actually get precise AI outputs
If you are still starting your prompts with "you are an expert marketer" or "you are a world class copywriter," you are wasting your time. That advice worked two years ago when models were basic. Today, it just forces the AI to output generic, stereotypical jargon. It makes the model try too hard to sound like what it thinks a marketer sounds like, which is usually a terrible caricature full of words like "synergy" and "delve."
Instead of telling the AI who it is, you need to tell it what it knows, what it must ignore, and how it must evaluate its own work. This is how you get professional results.
Instead of a generic persona, try giving the AI a negative constraint profile. Tell it exactly what mistakes a junior worker would make in this task, and ban those specific mistakes. For example, if you want a product description, do not ask for a professional writer. Tell the AI to avoid passive verbs, never use more than two adjectives per noun, and completely avoid exclamation marks.
Another technique that actually works is few-shot framing with critique. Do not just give the AI one example of what you want. Give it one good example and one terrible example. Explain exactly why the good one works and why the bad one fails. This forces the model to understand the underlying logic of your preference, rather than just copying the surface style.
When you do this, you stop getting that shiny, hollow AI tone. You get something that actually looks like it was written by a thoughtful human who understands the nuances of your specific task.
What is the weirdest or most specific negative constraint you have had to give an AI to make it stop writing like a robot?