“You are a world-class expert” is probably the most copied line in prompting, and one of the least effective. A persona only matters when it changes the decisions the model makes, not when it adds a flattering job title. If you have read our guide to reusable prompt templates, think of the persona as the part you write once and reuse everywhere.
In this post we break a working persona into experience, priorities and limits, and show how to keep it separate from the task. Every persona used in our text prompts follows this pattern.
What a weak persona looks like
You are a world-class marketing expert with 20 years of experience.
This sounds specific, but it tells the model nothing it can act on. What kind of marketing? For whom? What does this expert refuse to do? The output usually ends up generic, just a bit more confident.
The three parts of a persona that works
- Experience: the context the persona has worked in, described concretely (B2B SaaS, local retail, regulated finance).
- Priorities: what this person optimises for when two goals conflict, such as clarity over cleverness.
- Limits: what they will not do, written as behaviour.
Here is the same marketer, rewritten:
You are a senior B2B copywriter who has spent years writing for
software companies selling to operations managers.
You prioritise clarity over cleverness and concrete numbers over
adjectives. When a claim can't be backed by the brief, you leave
it out instead of softening it.
You never use exclamation marks, you never promise outcomes the
product can't guarantee, and you ask for missing facts instead
of inventing them.Write limits as behaviour, not adjectives
“Be professional” is an adjective. “Never use exclamation marks” is a behaviour. Models follow behaviours far more reliably, and you can check them in the output in seconds.
If you can’t check it in the output, it isn’t an instruction. It’s a wish.
Editorial team
Keep the persona separate from the task
A persona that lives inside one prompt gets copied, edited and forked until there are six versions of it. Store it once as a reusable snippet and include it in every prompt that needs it. When you improve the persona, every prompt that uses it improves too.
This also makes testing easier: you can run the same task with two different personas and compare the results directly.
A quick checklist
- Does the persona describe a concrete context, not just a title?
- Does it say what to prioritise when goals conflict?
- Are the limits written as behaviour you can verify?
- Is it stored once and reused, rather than pasted into every prompt?
Want to see personas in action? Open any prompt in the library and use Preview. The fill form lists which persona is included.