How to train your AI agent in TotalChat with documents, URLs and FAQs

Deploying a generic chatbot that answers with mechanical phrases can frustrate your prospects. In TotalChat, the AI architecture follows a strict Source of Truth principle: your agent only answers based on the official information you provide.

1. The primary prompt: define your agent’s role

The first step in TotalChat is setting the operating guidelines in the AI Training Campaign section:

• Identity: Instruct the agent to speak in the first person plural (“our company”, “our catalog”).
• Clear policies: Specify what to do for pricing, discounts and warranty questions.
• Tone of voice: Define whether communication should be executive, friendly or technical.

2. The three training methods

TotalChat offers three complementary ways to load your information:

A. Structured documents (PDF, TXT, Word)

Upload product manuals, terms and conditions or catalogs. You can choose between:

• Raw Response: Processes the whole document for exhaustive, context-rich answers.
• Generate FAQ: Splits the content into short questions and answers for speed and token savings.

B. Web link (smart URL scraping)

Enter the URL of your services or documentation page. With CSS selectors by class or ID (Class / ID), you can capture only the relevant block (for example #pricing or .specs), excluding irrelevant headers and footers.

C. Frequently asked questions (manual FAQ)

Enter your customers’ usual questions directly with exact answers and no room for error.

3. Enabling “No Match Reply” and fallback

To guarantee continuous support:

1. Go to Bot Manager > Action Buttons > No Match.
2. Check AI Reply and select your configured campaign.
3. Turn on Enable No Match Reply in the general settings.

If a customer asks something that does not match your buttons or preset flows, the AI steps in and answers from the uploaded training. And if human help is needed, the agent hands the chat over to your team in the Live Chat Inbox.