How Chatref works

Step 1: Connect your data sources

Chatref starts by connecting to the content you want the chatbot to use. This can include websites, documents, FAQs, and internal knowledge bases. The content you connect is used only to answer questions and remains isolated within your workspace, following the same security principles described in the Security section.

This setup allows businesses to add an AI chatbot to their website without changing how their content is written, as explained in adding an AI chatbot to a website.

Step 2: Content is structured and indexed

Once connected, Chatref breaks the content into smaller, structured sections that are easier to search. Each section is stored with context such as its source and position within the original content.

This structure makes it possible to retrieve only the most relevant information later, which is a key part of how retrieval-augmented generation works.

Step 3: A user asks a question

Visitors or internal team members can ask questions in natural language, similar to how they would interact with a chat interface. There are no predefined flows or fixed commands. Users can ask direct, specific questions about the connected content, such as product details, documentation, or policies.

This experience is commonly described as letting users chat with a knowledge base.

Step 4: Relevant information is retrieved

When a question is asked, Chatref searches only within the connected content to find the most relevant sections. Irrelevant information is ignored, and only the parts that directly relate to the question are selected.

Chatref does not search the internet and does not use unrelated data sources. This retrieval step ensures that answers are grounded in the content you provided, which is why Chatref differs from open-ended chat systems often compared on the comparison page.

Step 5: An answer is generated from retrieved data only

After the relevant information is retrieved, Chatref generates an answer using only that content. The system is designed to stay within clear boundaries, responding strictly based on what exists in your data.

If the required information is not present, Chatref does not make assumptions or fill gaps with guesses. This behavior prioritizes accuracy and reliability, which is a common requirement addressed in the FAQ section.

Step 6: The response is delivered to the user

The final answer is shown to the user through the embedded chatbot. This works consistently across public websites, documentation pages, and internal tools. The same retrieval and answer logic applies regardless of where the chatbot is used.

This approach is also used when answering questions from documents, including PDFs, as described in using an AI chatbot for PDFs and documents.

How Chatref avoids hallucinations

Chatref avoids hallucinations by retrieving information before generating an answer. The system does not respond unless relevant content is found, and it does not rely on general knowledge or public data.

If the connected content does not contain the answer, Chatref clearly indicates that the information is not available. This design choice favors accuracy over completeness and aligns with the principles outlined in why RAG is used.

Supported data sources

Chatref can work with multiple types of content, including:

• Websites and landing pages
• PDF files
• Text and document files
• FAQs
• Internal documentation

These sources can be updated or removed at any time, allowing the chatbot to stay aligned with the most current information.

Security and data isolation

All connected content is isolated at the workspace level. Chatref does not share data across customers and does not use customer content to train public models. Access and handling of data follow the controls described in the Security section.

What happens when an answer is not found?

When Chatref cannot find relevant information in the connected content, it responds by stating that the information is not available. It does not generate speculative answers or attempt to infer details beyond the data provided.

This behavior helps maintain trust and is one of the most common questions addressed in the FAQ.

Summary

Chatref works by retrieving relevant information from your connected content and generating answers strictly from that data using retrieval-augmented generation. This approach ensures accurate, secure, and reliable responses without guessing or hallucinations.