Training Your Chatbot

Training Your Chatbot

The effectiveness of your chatbot depends on the quality of its training data. Here is how to train your chatbot to give accurate, helpful responses.

Knowledge Sources

Document Upload:

1. Go to Chatbots > [Your Bot] > Knowledge Base.
2. Click Upload Document.
3. Select PDF, DOCX, TXT, or MD files.
4. WRRK processes the document, extracting text and building a searchable index.
5. The chatbot can now reference this content when answering questions.

URL Crawling:

1. Click Add URL.
2. Enter a web page URL (e.g., your product documentation).
3. WRRK crawls the page and extracts content.
4. Optionally enable Follow Links to crawl linked pages (up to 50 pages deep).

FAQ Pairs:

1. Click Add FAQ.
2. Enter a question and its answer.
3. Add variations of the same question to improve matching.

Training Tips

Be comprehensive — The more content you provide, the better the chatbot performs. Cover all common questions your customers ask.

Use natural language — Write FAQ answers the way you would speak to a customer, not in robotic or overly formal language.

Include edge cases — Think about unusual questions or misunderstandings and add FAQ entries for them.

Update regularly — As your product or service changes, update the knowledge base. Remove outdated information promptly.

Testing After Training

After adding training data:

1. Click Test Chat in the chatbot builder.
2. Ask questions as a customer would.
3. If the chatbot gives incorrect or incomplete answers, refine your training data.
4. Repeat until you are satisfied with the responses.

Conversation Review

Periodically review real conversations:

1. Go to Chatbots > [Your Bot] > Conversations.
2. Read through recent chats.
3. Identify questions the bot could not answer.
4. Add those topics to the knowledge base.

Fine-Tuning Behavior

Beyond knowledge, you can adjust how the chatbot behaves:

• Tone — Set to Professional, Friendly, or Custom.
• Fallback Message — What the bot says when it cannot answer.
• Handoff Message — What the bot says when escalating to a human.
• Language — Primary and secondary languages the bot should respond in.