Designing AI Assistant Personalities for Consistent BehaviourDesigning AI Assistant Personalities for Consistent Behaviour
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Every AI assistant has a personality. Even those nobody bothered to design.
These are the assistants that tend either to be excessively cheerful, demonstrably bored... or just systematically repetitive: "Sorry, I didn't understand that."
These agents are what happens when an organisation gives a language model only the bare bones of a brand guideline doc - and then just asks it to 'get on with things'.
The result is a personality by accident. Flat, one-dimensional and misrepresentative of the brand identities they've been positioned to represent Brands can often convince themselves that a tone of voice amounts to a fully-rounded, operational personality. And, while a conventional tone-of-voice document might describe a brand as friendly, confident, approachable or knowledgeable, that's only really useful as long as language remains static.
It's considerably less helpful when an AI assistant needs to decide whether to challenge a user's assumption, acknowledge uncertainty or interrupt a conversation to clarify something important.
An agent needs more than adjectives. It needs behavioural principles.
It needs to know:
> When to be concise and when to explain.
> How to handle uncertainty without becoming evasive.
> When humour is appropriate — and when it emphatically isn't.
> How to respond to frustration, ambiguity and repeated failure.
> When to hand control back to a human.
These are decisions about interaction, not simply writing style.
A useful starting point is to establish the agent's role, boundaries and behavioural priorities.
Is it an efficient problem-solver? A patient guide? A knowledgeable specialist? What happens when these characteristics conflict?
Consider an assistant that has been instructed to be both reassuring and transparent. If it cannot complete a request, should it reassure the user first or explain the limitation?
The answer depends on the context. That's why agent personality design needs scenarios, evaluation criteria and testing - not just a colourful persona description.
A strong framework should also define what the agent must never pretend to know or do.
Consistency is the real test, and the measure of a good agent personality isn't whether it sounds charming in a demonstration.
It's whether it behaves coherently when the conversation becomes difficult.
That means testing edge cases, ambiguous instructions, emotional cues, interruptions and failures. It means checking that the assistant remains recognisably itself without allowing personality to compromise clarity or trust.
A brand's AI personality shouldn't be whatever falls out of the model. It should be a deliberate part of the experience.
And that starts long before anyone writes the welcome message.

I work across conversational UX, AI persona design, brand-language systems and agent behaviour frameworks,: helping organisations build assistants that don't just sound more human, but behave more usefully.
Post image
Back to feed
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started