Which tool to choose
It works with any AI assistant: the "personality" isn't a special feature, it's an instruction you give in words. If the AI you use lets you save custom assistants or fixed instructions (sometimes called projects, gems, custom GPTs), it's worth creating one for each recurring role: that way you don't rewrite the character every time. For occasional roles, just declare them at the start of the chat.
How to do it
A well-defined personality isn't a cute mask: it really changes what the AI notices and what it ignores. A "cautious lawyer" sees the risks that an "enthusiastic salesperson" skips.
- Decide the profession and a character trait: not "expert", but "fussy reviewer who hates vague sentences".
- Open the chat declaring the role in the first line, before giving the task.
- Add what it should look at first: the editor looks at clarity, the accountant looks at the numbers that don't add up.
- Keep the roles in separate chats: one for the copywriter, one for the reviewer. If you mix them, the responses become lukewarm and average.
- When you need a role often, save it as a custom assistant or paste it from a note.
The operational syntax:
You are a fussy editorial reviewer, irritated by vague sentences and promises without proof. When I give you a text, you don't praise it: you list the weak points frankly and propose a leaner version. Always start from the weakest sentence. Confirm you've understood the role, then I'll paste you the text.
After giving the role, ask a test question: if the AI answers in the right character, the role has taken hold. If it answers lukewarm and generic, reinforce the trait and repeat.
A concrete example
Sara runs a small e-commerce on her own and uses the AI for three very different things. She keeps three fixed chats: a "persuasive copywriter" for the product descriptions, a "patient customer assistant" for replying to difficult messages, a "distrustful analyst" for checking the shipping accounts. When she has to write a description she opens the first, and the AI is already in the right character. She doesn't explain who it should be every time: the role lives in the chat. Three personalities, one subscription.
When it does NOT work (and how to fix it)
If the AI forgets the role after a few messages
In long conversations the personality given at the start fades. Repeat it in short form when you notice: "remember, you're the fussy reviewer". Better still, save the role as a custom assistant: there the instruction stays fixed and doesn't get forgotten between one message and another.
If all the personalities sound the same
You gave roles that are too generic. "Marketing expert" and "finance expert" answer almost identically because the character is missing. Add the human trait: "cautious to a fault", "who cuts all the superfluous", "who always thinks of the angry customer". Character is what really differentiates the voices.
If you can't find where to save a custom assistant
The names change (projects, gems, custom GPTs) and the buttons move. If you can't find it, the solution that works everywhere is a note on your phone with your roles: you copy and paste the character at the top of every new chat. Less elegant, always available.
A tip from someone who actually uses it
Write your recurring roles once and tend them like colleagues. The reviewer that works well today you refine over time: add "never accept a useless adverb", remove what makes it too aggressive. After a month you have a small team of personalities calibrated to your work, and the value isn't novelty, it's that they're already broken in.
Frequently asked questions
Is it the same thing as the "custom instructions" you set once?
Partly. Custom instructions apply to all chats and describe you (who you are, how you want the responses). Per-task roles are more targeted: you change personality depending on the work. The two combine: general instructions about you, specific role for the single chat.
Can I give multiple roles in the same conversation?
You can, but they make the responses confused and average: the AI tries to satisfy all the characters at once. Better one chat per role. If you need the comparison between two points of view, ask for it explicitly: "answer first as an optimist, then as a skeptic".
Does giving the AI a personality make it smarter?
No, and believing it leads to disappointments. The role doesn't add knowledge: it orients the knowledge it already has. A "Nobel laureate in physics" doesn't know more physics than the base model, but answers with the cut and priorities of that profession. The personality is a filter on attention, not an increase in capability.