The AI chatbot site visitors trust is the one that answers from your own content and admits when it does not know. That trust is built in the knowledge base, not in the chat window. On the levelpartner.ai platform, each AI agent has a knowledge base you fill from your website pages, written FAQs and documents. This article answers one question: how do you build and maintain the knowledge behind your site's chatbot so its answers are accurate, current and specific to your firm?
Why answers go wrong
A chatbot that answers badly almost always has one of three knowledge problems. The answer is missing, so it guesses. The answer is there but buried in a long page, so it finds the wrong passage. Or two sources disagree, such as an old services page and a new pricing page, and it picks the stale one.
Each problem has a fix, and none of them involves the AI model. They involve tidying what you feed it.
Three kinds of source
Website pages give the chatbot broad coverage quickly. You add your site's URLs and the platform reads the content. They are good for general questions about services and location, weaker for precise answers because web copy is written to persuade, not to answer.
Written FAQ entries are the most reliable source. Each is a question and a direct answer you wrote and approved. Use them for anything the chatbot must get exactly right: what is included, how onboarding works, which areas you cover.
Documents such as service guides and policies fill the gaps. Keep them short and current. A forty-page brochure is a poor source; a two-page summary of the same content is a good one.
Before you start
- An AI agent connected to your site's chat widget.
- A list of the questions visitors ask most, from chat history, emails and sales calls.
- Approved answers to each, checked by someone who knows the detail.
- A list of website pages that are current, and a list of those that are out of date.
- An owner for the knowledge base, with time each month to update it.
Short documents, such as a two-page service overview per industry, go in through + Add Source as a file upload once the pages and FAQs are in.
How to set it up
- Open AI Agents and select the Knowledge Base tab.
- Create a new knowledge base or edit an existing one.
- Click + Add Source and select Web Crawler.
- Choose Exact URL so you add only current pages.
- Enter the URL and click Extract Data.
- Click View All Pages and tick the URLs you want.
- Click Train Bot.
- Open the FAQ tab and add your top questions with answers.
- Open the Knowledge Base Retrieval Tester and ask a realistic question.
- Return to the bot's Bot Training tab and select the knowledge base.
Writing FAQ entries the bot can use
Write the question the way a visitor would type it, not the way your website names the service. "Do you work with mortgage brokers?" beats "Industry coverage". Add common variations as separate questions if the answer is important.
Put the fact first. "Yes, we work with mortgage brokers in the UK, the US, Canada and Australia. Most start with lead follow up and booking." The bot can lift that answer whole. A paragraph that starts with your mission statement forces it to dig.
Say the consultancy's chatbot answers from 140 web pages, including old blog posts with outdated service names, so visitors hear about a service the firm no longer sells. The clean-up would remove 90 pages and add 35 FAQ entries plus four two-page service summaries. With the outdated sources gone, wrong answers in a monthly sample should shrink toward none, because the bot can no longer find the old material.
What this replaces
A maintained knowledge base replaces the scattered answers in people's heads and sent folders. Once written, the same FAQ entries are useful to new staff, in email templates and on your website's FAQ page. The chatbot is the first user, not the only one.
The usual mistake is adding every page on the site because more data feels safer. More sources mean more chances to pick a stale one. Add fewer, better sources, and prefer written answers over long pages.
Time to build and maintain
The first version takes a day, mostly writing and checking FAQ entries. Maintenance takes about an hour a month: export the questions the bot handed over or answered poorly, write answers to the new ones and retire anything out of date. Any time you change a service, price or policy, update the knowledge base the same day.
When the knowledge base is not the answer
Some questions should never be answered from a knowledge base, because the right answer depends on the person. Fees that depend on scope, mortgage product suitability and property valuations are examples. For those, the best knowledge entry tells the bot to explain why it cannot answer and to offer a call.
Troubleshooting
The bot quotes an old price or service. An outdated page is still in the knowledge base. Search your sources for the old term and remove or update every match.
Answers are vague. The bot is drawing on long marketing copy. Write a direct FAQ entry for that question.
It answers questions you never covered. It is filling the gap from general knowledge. Tighten the instruction to answer only from your sources and to hand off otherwise.
Two answers contradict each other. Two sources disagree. Pick the correct one and delete or fix the other.
Frequently asked questions
How much content does an AI chatbot site need to start?
Twenty to forty FAQ entries covering your most common questions, plus your main service pages, is enough for a solid start. Add more as real questions arrive. Quality beats quantity at every stage.
Does the knowledge base update automatically when my website changes?
Only if you ask it to. Crawled web pages can be set to refresh on a daily, weekly or monthly schedule, and you can refresh a single page by hand after an edit. FAQ entries and uploaded documents never change on their own, so a monthly review still matters.
Can the same knowledge base serve chat and voice?
Yes, but write voice answers shorter, because callers cannot scan. Many firms keep one FAQ set and add a short spoken version for voice. See knowledge based agent in artificial intelligence for the theory, and AI chatbot website and chat bot site for the wider set-up.