Skip to content

How the response is created

Discover how glacom.AI processes information to answer your customers and how it ensures it doesn't invent content.

How it works

When a customer asks a question, for example "What are the opening hours on Saturday?", glacom.AI doesn't answer from memory. The system searches your documents for the most relevant pieces of text, reads your instructions on the tone of voice, and composes a response using only that information. If you wrote that you close at 1:00 PM on Saturdays, the assistant will say exactly that, citing the source if requested.

The process follows these steps: 1. It receives the customer's question. 2. It retrieves the most suitable chunk via semantic search. 3. It applies the customization rules you set. 4. It generates the final text based exclusively on the data found.

What you need to do

Define the instructions: tell glacom.AI how it should speak (formal, friendly) and which rules to follow. Upload content: make sure the answers are present in your documents, as explained in writing content.

What glacom® does

The system analyzes the request, filters out useless information, and synthesizes a coherent response. If the uploaded documents do not contain the answer, glacom.AI doesn't invent anything but informs the customer that it doesn't know, as described in I don't know.

Limits

The assistant cannot use knowledge external to the documents you provided: if a piece of information is true but not written in your files, glacom.AI won't know it. The quality of the response depends entirely on the clarity of the uploaded texts.

Frequent problems

  • Why does the assistant say it doesn't know if the information is on the site? → The content probably wasn't processed correctly or is written ambiguously.
  • Can the assistant invent a price? → No, if the price is not present in the text blocks, the system will not generate it.
  • How do I change the way it responds? → You need to modify the chatbot customization instructions.

From the app

You can monitor the answers given and report imprecise ones to refine the system.