Knowledge Base

Last updated: October 6, 2026

What a knowledge base is

Traditionally, a knowledge base is a help center or an internal wiki: articles written so customers or employees can find answers on their own. In the context of AI chatbots, the term is broader. It means everything the chatbot has been given to answer from, including web pages, documents, spreadsheets, notes and question-and-answer pairs.

The knowledge base is separate from the language model. The model provides language skills and general knowledge. The knowledge base provides your business facts, such as prices, opening hours, return rules and product details.

How a chatbot uses it

Most AI chatbots connect the two with retrieval-augmented generation (RAG):

Ingest: content is collected from websites, files or manual entries and converted to plain text.
Chunk and embed: the text is split into passages, and each passage is turned into an embedding.
Retrieve: for each question, the most relevant passages are found.
Answer: the language model writes a reply based on those passages.

An example

A dental clinic builds its chatbot knowledge base from its website, a PDF price list and a few Q&A pairs covering questions the reception desk hears every day, such as parking and payment in installments. When a visitor asks whether the clinic treats children, the chatbot finds the pediatric page and answers from it. When asked about a treatment the clinic does not list, a well-configured chatbot says it does not have that information instead of guessing.

Why it matters for business chatbots

A chatbot is only as accurate as its knowledge base. Outdated prices, contradictory pages or missing policies lead directly to wrong or incomplete answers. Good practice includes:

Cover the questions customers actually ask, not only what is on the homepage.
Update or remove outdated content, and avoid two pages that say different things.
Write short, direct answers for sensitive topics such as prices and policies.
Review chat logs to find questions the chatbot could not answer, then add the missing content.

How intoCHAT builds it

In intoCHAT, an agent's knowledge base is made of sources: a website crawl (up to 500 discovered URLs, from which you pick pages), uploaded files such as PDF, Word, PowerPoint, Excel, CSV or Markdown, text snippets and Q&A pairs. Q&A answers take priority over other sources, and you can star a source so it ranks higher without retraining.

You can edit, preview, download, delete or retrain each source. Retraining is manual, so retrain a source after the original page or file changes.

Frequently asked questions

What should I put in a chatbot knowledge base?

Start with the content that answers your most common customer questions: product or service pages, prices, policies, opening hours and contact details. Add Q&A pairs for questions your website answers poorly. Leave out drafts and outdated material, because the chatbot will treat it as current.

How is a knowledge base different from training a model?

A knowledge base is searched at answer time, so you can add, change or remove content without changing the model. Training or fine-tuning changes the model's internal weights, which is slower and harder to update. Most business chatbots rely on a knowledge base for their facts.

How often should I update my chatbot's knowledge base?

Update it whenever the underlying information changes, such as new prices, policies or products. In intoCHAT, retraining is manual, so retrain the affected source after you change the original page or file. Reviewing chat logs regularly shows which gaps to fill.