How it works: the bot answers with your content

Your bot doesn't invent answers. It searches through the content you provided and answers with what it finds. This is the most important rule to understand.

How it works

A customer writes: "are you open on Saturdays?". The bot doesn't "know" the hours by heart. It does this:

1. It takes the question and searches, among all your content, for the parts that talk about hours.
2. It finds the right paragraph: "We are open from Monday to Saturday, from 9 am to 7 pm".
3. It builds the answer using only that part, in the customer's language.

This method is called RAG: the bot retrieves your content and then answers. The advantage: if the answer is in your documents, it gives it; if it's not there, it says it doesn't know instead of inventing.

The steps, in brief

• Indexing: when you upload content, it is prepared for search.
• Chunk: long documents are divided into blocks.
• Embedding: each block is transformed into searchable "meaning".
• Semantic search: the question finds the closest blocks.
• Generation: the bot writes the answer using those blocks.

What you need to do

Give the bot good content. A bot is only as good as the content you give it: see Writing content.

Limits

• If the answer is not in your content, the bot doesn't know it. This is not a flaw: it is the guarantee that it doesn't invent.
• The bot does not reason across dozens of documents at once and does not perform complex calculations: see Limits.

From the app

From the app you can add a FAQ or photograph a document to include in the sources. See Knowledge base from the app.