How to Train a Chatbot on Your Own Documents

What Types of Documents Work

The system accepts plain text and PDF files. You can also paste text directly into the knowledge base interface without uploading a file. Common sources include:

• FAQ documents with questions and answers about your products or services
• Product documentation covering features, specifications, and usage instructions
• Policy pages like return policies, terms of service, shipping information
• Internal knowledge base articles your team uses to answer customer questions
• Training manuals for employee onboarding or process documentation
• Pricing information including plans, tiers, and what each tier includes

The content should be written in clear, factual language. The chatbot performs best when the source material directly states the information rather than implying it. If your FAQ says "We ship within 3-5 business days," the chatbot will quote that exactly. If the information is buried in marketing language, the chatbot may struggle to extract the specific detail a visitor needs.

How to Upload Documents

How Chunking and Retrieval Work

When you upload a document, the platform does not feed the entire document to the AI model on every question. That would be slow and expensive. Instead, it breaks the content into chunks and converts each chunk into a vector embedding, which is a mathematical representation of the chunk's meaning. These embeddings are stored in a searchable index.

When a visitor asks a question, the system converts that question into an embedding too, then finds the chunks whose embeddings are most similar. Typically the top 3-5 most relevant chunks are included in the prompt sent to the AI model. This process is called Retrieval Augmented Generation (RAG), and it is what allows the chatbot to give specific, sourced answers without hallucinating.

Chunking costs 3 credits per chunk as a one-time cost. A 10-page PDF might produce 20-40 chunks depending on content density, costing 60-120 credits (under $0.12). Once the chunks are created, they persist in your knowledge base until you delete them, and there is no ongoing storage cost.

Best Practices for Training Data

Be Specific and Factual

Write content that states facts directly. Instead of "Our shipping is incredibly fast and reliable," write "Standard shipping takes 3-5 business days. Express shipping takes 1-2 business days. Free shipping is available on orders over $50." The chatbot can only be as specific as its source material.

Cover Common Questions First

Start with the 20-30 questions your customers ask most often. Look at your support inbox, chat history, or FAQ page for patterns. Training the chatbot on these first gives you the biggest immediate impact. You can always add more content later.

Keep Content Current

If your pricing, policies, or product details change, update the knowledge base too. You can delete old chunks and upload new versions at any time. The chatbot will immediately use the updated information. See How to Keep Your AI Training Data Up to Date for a maintenance strategy.

Separate Topics Clearly

If you upload one large document that covers many topics, the chunking may mix related but different information in the same chunk. It is often better to upload several smaller documents, each covering a single topic, so each chunk stays focused. See How to Organize Training Data for Best Results for more on this.