Does the persona change what it knows?

No. Knowledge comes from your indexed sources. The persona changes how it speaks and what it declines to attempt.

Can I stop it answering certain topics entirely?

Yes — state those categories in the rules. Combined with not indexing material on those topics, that is the reliable way to keep a boundary.

How chatty should it be?

Less than you think. People arriving with a question want the answer and the source. Personality is best kept to the greeting and the refusal, where it softens an unhelpful moment.

Getting started

Set your bot's tone, greeting, and rules

The persona is not decoration. The rules you write here decide what it refuses, which is the setting that matters most.

Start freeNo credit card. 2 minute setup.

A default assistant sounds like everyone else's

Out of the box it is competent and generic. The gap between that and something that sounds like your business is a greeting, a tone, and a short list of rules about what it should decline.

Replies read as generic vendor support rather than as your brand
The greeting invites questions the bot cannot answer
No explicit boundaries, so it attempts things it should decline
Escalation language does not match how your team actually talks

Write a greeting that sets expectations

The greeting shapes what people ask. 'Ask me anything' invites questions outside your content and produces refusals; naming what it is good at — 'Ask about shipping, returns, or sizing' — produces answerable questions instead.

Choose a tone that matches the rest of your site

Read your own homepage aloud and match it. An assistant that is chattier or more formal than everything around it reads as bolted on, which undermines the credibility that citations are meant to build.

The Role and tone settings panel offering four primary roles — customer support, sales assistant, product guide, and custom — with custom selected. — The three preset roles set tone and priorities for you. Custom replaces them entirely, so pick it only if you intend to write the whole persona.

Write the rules about what it must not attempt

This is the important part. State the categories it should decline outright — medical, legal, or financial advice, individual account questions, anything requiring a quote. Explicit boundaries are more reliable than hoping the absence of content is enough.

Name the categories

Be specific: 'do not estimate delivery dates' beats 'be careful'.

Say what to do instead

Point it at the escalation route so a refusal is still useful.

Test the boundaries, not just the happy path

Ask it the things you told it to refuse. A persona is only configured correctly if the refusals fire — and this is much easier to check before launch than after a customer finds the gap.

Common questions

Does the persona change what it knows?: No. Knowledge comes from your indexed sources. The persona changes how it speaks and what it declines to attempt.
Can I stop it answering certain topics entirely?: Yes — state those categories in the rules. Combined with not indexing material on those topics, that is the reliable way to keep a boundary.
How chatty should it be?: Less than you think. People arriving with a question want the answer and the source. Personality is best kept to the greeting and the refusal, where it softens an unhelpful moment.