An AI chatbot is only as good as the information you give it. Point it at a messy website and it will give messy answers. Give it clear, current, well organised content and it will answer most customer questions correctly. This guide covers what to include and how to keep it in good shape.

What an AI chatbot knowledge base is

It is the set of content the chatbot is allowed to answer from: web pages, FAQs, policies, product details and help articles. When a customer asks something, the chatbot finds the most relevant parts and answers from them, instead of making something up.

What to include first

Your most asked questions, with clear answers. Start here.
Policies: shipping, returns, refunds, cancellations, warranties.
Pricing and plans, including what is and is not included.
Products or services: what they are, who they are for, key details.
Contact and opening hours, and how to reach a person.
Step-by-step help for common tasks such as booking, ordering or changing an account.

Structure it so answers are easy to find

One topic per page or entry. A long page about everything is harder to answer from than several focused ones.
Write FAQs as real questions and answers, in your customers' words.
Put the key fact near the start of each section.
Spell out numbers and conditions: "Returns within 30 days of delivery, unused, with tags."

Avoid contradictions

If your shipping page says 3 to 5 days and an old FAQ says 2 to 4, the chatbot may quote either. Contradictions are the most common cause of wrong answers. When you change a policy, search all your content for the old version and update or remove it.

Leave out what should not be said

Internal notes, staff instructions and draft pages.
Old promotions that have ended.
Anything you would not want a customer to read word for word.

Some topics also need rules, not just content. For example, "never confirm a refund amount" or "only share order details after the customer signs in".

Find the gaps

No knowledge base is complete on day one. The best way to improve it is to look at the questions the chatbot could not answer. Each one is either a missing article or a sign that an existing one is unclear. Review them every week at first, then less often as the gaps close.

Keep it current

Update the knowledge base the same day prices or policies change.
Re-check pages that are crawled from your website after a redesign.
Remove seasonal content when the season ends.

Building one with IotaBot

IotaBot builds the knowledge base from your web pages, sitemap, FAQs, plain text, industry templates and help-center articles. It lists the questions it could not answer so you can fill the gaps, and plain-English bot rules control what it may say and to whom. File uploads such as PDFs are not supported. See the knowledge base software page.