Configuring the AI Bot Persona

This guide walks through setting up an AI WhatsApp bot persona — its behavior, its knowledge, and the auto-reply rules that sit alongside it.

1. Create a bot persona

From the AI bot settings, create a new persona. A persona is the identity the AI bot operates under when it replies to contacts — you can have more than one, for example if different numbers or use cases need different behavior.

2. Write the system prompt

The system prompt is where you describe the bot's role and tone — who it's representing, what it should help with, how formal or casual it should sound, and any boundaries on what it shouldn't do or claim. This is the instruction set the AI follows for every reply it generates, so it's worth being specific rather than generic.

3. Build the knowledge base

The knowledge base is what grounds the bot's answers in your actual business information instead of relying on general knowledge alone. Add entries in whichever form is easiest for the source material:

• Pasted text — type or paste content directly, useful for short facts, policies, or FAQs.
• Uploaded document — upload an existing document so its content becomes searchable knowledge for the bot.
• URL — point to a web page, letting the bot draw on content that's already published (like a product page or help article) without copying it manually.

4. Manage and reorder entries

Knowledge base entries can be reordered and managed as your information changes. Each entry can also be toggled active or inactive individually — useful for temporarily disabling outdated information (like an expired promotion) without deleting it outright.

5. Set up auto-replies alongside the AI layer

Beyond the AI-generated responses, configure the standard auto-reply types that run alongside it:

• Keyword replies — a fixed response triggered by a specific incoming keyword.
• Welcome message — sent to a contact reaching out for the first time.
• Away message — sent outside your configured availability, if applicable.
• Fallback reply — sent when nothing else matches, so a contact never gets silence.

These give you predictable, exact-wording responses for specific situations, while the AI layer handles everything more open-ended.

6. Consider a handoff to a human

Mersal's data model supports handing a conversation off from the bot to a human agent when the AI isn't the right fit to continue — for example, once a conversation reaches a certain complexity or a contact explicitly asks for a person. Set this up if your team needs a way to step in on conversations the bot starts.

7. Test the persona

Send test messages that mirror how real contacts phrase questions, and check both the tone (does it match the system prompt) and the accuracy (does it pull the right facts from the knowledge base) before relying on it for live conversations.

Next steps

• AI WhatsApp Bot — the full feature reference for how personas, knowledge bases, and AI-generated replies fit together.
• Building an Automation Workflow — combine the AI bot with workflow nodes for more structured, multi-step conversations.