How to Train a Chatbot on Your Own Data

Vatdi indexes four kinds of content, and most stores use all four:

• Product catalog — names, descriptions, prices, options, stock and categories, synced by the platform plugin and refreshed when products change.
• Website pages — shipping, returns, sizing, about and policy pages, crawled from the address you give.
• Documents — PDF, Word, TXT, CSV and Markdown files up to 10 MB each.
• FAQs — question-and-answer pairs you write yourself or import from a spreadsheet.

Everything you add is stored for your store only. Nothing from one store is ever used to answer a visitor on another.

The chatbot is not retrained as a model. Instead each product, page, file and FAQ is split into short passages and converted into embeddings — numeric fingerprints of meaning — stored in a vector index for your store. When a visitor asks a question, Vatdi finds the passages closest in meaning to the question and hands only those to the language model, which writes the reply from them. This is called retrieval-augmented generation (RAG).

The practical consequence: the chatbot can only say what your content says. When nothing relevant exists, it tells the visitor it does not have that information and offers to hand over to your team, rather than inventing an answer. Where a product or page is the source, the reply links to it so the visitor can check for themselves.

Catalog changes need no work from you: the plugin pushes new, updated and removed products to Vatdi and the index follows. For pages and documents, re-add the page or upload the new version of the file whenever it changes; the previous passages are replaced. You can also switch any item off from the dashboard without deleting it, which is useful for seasonal policies.

The most reliable maintenance habit is to read the conversations once a week. Every question the chatbot could not answer is a gap in your content — add the missing FAQ or page and the next visitor gets a proper reply.