Chatbot Trained on Knowledge Base: Rewrite Help Articles as Short Answers a Bot Can Use (Free Article-vs-Answer Examples)

Chatbot trained on knowledge base is the tidy idea: your help articles already exist, so let the bot read them. In practice, articles written for people browsing a help centre are not the same as answers for someone in a hurry. This guide shows how to rewrite articles into short, safe answers, with side-by-side examples, so the bot sounds helpful instead of vague.

2websites on the free plan
100free visitors per website each month
0per-seat charges
1 hrto rewrite your top ten articles

โšก In short

  • Help articles are written to be browsed. Bot answers need to be short, direct and complete in themselves.
  • Rewrite each article as a question plus a two or three line answer. Keep one fact per answer.
  • Remove phrases like ‘see above’ or ‘click the link below’, which mean nothing to a bot.
  • In LiveChatBot you add this knowledge to the AI Agent yourself; there is no automatic import of a help centre.
Chatbot trained on knowledge base in LiveChatBot: AI Agent knowledge sources and a sample chat answered from them
How it looks in LiveChatBot: AI Agent knowledge sources and a sample chat answered from them.

The gap

Why articles don’t work as bot answers

A help article has a title, an introduction, steps, screenshots and related links. A person skims it and finds the line they need. A bot has no eyes. It looks for passages that match a question and writes from them.

That difference creates three problems.

  • Context is spread out. The answer to “can I return a sale item?” might sit in step four of an article titled “Returns overview”.
  • References break. “See the table above” and “as shown below” point at things the bot never sees.
  • Introductions dilute meaning. Three lines of welcome text can outrank the one line that matters.

A chatbot trained on knowledge base content works best when each entry stands alone. If you read an entry with no other page open and it still makes sense, it is ready.

Try it

Article versus answer: five rewrites

Each pair shows an article-style line and the short answer that works better in a bot.

Returns
๐Ÿšฉ Article styleSee our returns section above for details on eligibility.
โœ… Bot-ready answerYou can return unused items within 7 days with the bill. Sale items are not returnable.
Delivery
๐Ÿšฉ Article styleDelivery times vary depending on a number of factors.
โœ… Bot-ready answerDelivery takes 3 to 5 working days in most cities. Remote areas can take up to 8.
Payment
๐Ÿšฉ Article styleWe offer a range of flexible payment options for your convenience.
โœ… Bot-ready answerYou can pay by UPI, card or cash on delivery. Cash on delivery is not available above โ‚น10,000.
Support
๐Ÿšฉ Article styleOur friendly team is always happy to help.
โœ… Bot-ready answerOur team replies between 10 am and 7 pm, Monday to Saturday. After that, leave your number.
Warranty
๐Ÿšฉ Article stylePlease refer to the warranty document for full terms.
โœ… Bot-ready answerProducts carry a 1 year warranty on manufacturing faults. Damage from misuse is not covered.

Method

A five-step method for a chatbot trained on knowledge base articles

Take your ten most-read articles, or if you have no analytics, the ten topics customers email about.

Step 1: write the question first

Use the words a customer would type, such as “Can I change my order?” not “Order modification policy”.

Step 2: answer in the first sentence

Put the yes, no or number up front. Details follow in one or two lines.

Step 3: keep one fact per answer

If an article covers returns, refunds and exchanges, split it into three answers. Smaller pieces are easier to match and easier to update.

Step 4: add the limit

Say what is not covered. “Sale items are not returnable” prevents a cheerful wrong answer.

Step 5: add a handover line

For anything involving money or a complaint, end with “A team member will confirm.”

That is it. Don’t polish the prose; clarity is enough.

In LiveChatBot

How this works in LiveChatBot

LiveChatBot’s AI Agent answers from the knowledge you give it. You add the rewritten answers there, and you decide what the bot may say.

There is no automatic import of a help centre and no link between an external knowledge base and the bot. That sounds like a drawback, but it forces a useful cleanup. Your bot gets ten good answers rather than two hundred tired articles.

If the monthly AI allowance is used up, the free keyword bot answers from your saved replies, so writing both at the same time is smart: one answer can serve as both.

Our guide on training your AI chatbot covers adding and testing the knowledge.

Worked example

Splitting one long article into four answers

Take an invented article called “Orders, changes and cancellations” with six paragraphs. A customer asks, “Can I cancel my order?” The bot must find the right paragraph among six.

Split it into four entries:

  • Can I cancel my order? Yes, before it ships. After that, refuse the cancellation and offer a return.
  • Can I change my address? Yes, before dispatch. Message us with the new address and order number.
  • How do I know it shipped? You receive an SMS with a tracking link.
  • What if I ordered by mistake? Contact us within 24 hours and a team member will help.

Each entry now has a clear question, a first-sentence answer and a limit. The bot no longer has to choose a paragraph; it finds a match. As a bonus, your team can reuse the same lines as saved replies, which keeps human and bot answers consistent.

Keep it right

Keeping the answers true after launch

A rewritten answer is a copy, and copies drift. When a price or policy changes in your help article, the bot’s version stays old until you update it.

Give each answer an owner and a check date. A simple rule works: whoever changes a policy updates the bot the same day. Review the full set every quarter.

Also read real chats. When a customer asks something the bot cannot answer, that is a missing entry. Write it once and add it.

For cases that need a person, turn the chat into a ticket so nothing is lost, and use chat flows to ask for contact details after hours.

Honest limits

What a chatbot trained on knowledge base content can’t promise

Be realistic about both the technology and the product.

  • A bot can still misread a vague entry, so ambiguity is your risk to remove.
  • LiveChatBot has no help-centre import and no automatic sync.
  • It has no PDF or file upload for training.
  • “Powered by LiveChatBot” shows on every plan.

What it does give you: knowledge you control, a keyword bot fallback, Flows, an away form, an Android app, Zapier, webhooks and a REST API. See pricing for the free plan, then sign up at livechatbot.in.

Frequently asked questions

Can I connect my help centre to LiveChatBot?

Not automatically. You add the knowledge to the AI Agent yourself, usually as short question-and-answer entries.

How long should each bot answer be?

Two or three lines. Lead with the yes, no or number, then add the one detail that matters.

Do I need to rewrite every article?

No. Start with the ten topics customers ask about most. Add others only when real chats show a gap.

Why avoid phrases like ‘see below’?

A bot answering in chat cannot show a section below. Each entry must make sense on its own.

Is a chatbot trained on knowledge base content safe for prices?

Only if prices rarely change. Otherwise tell the bot to hand price questions to a person.

Keep reading

Related guides

๐Ÿ“˜

Chatbot Trained on PDF Documents: Brochures, Price Lists, Manuals and Scans, and What to Do With Each One (Free Decision Tree)

Read the guide โ†’

๐Ÿงญ

AI Chatbot Trained on Website URL: What Happens When You Paste a Link, How to Test a Vendor’s Claim, and What to Do Instead (Free Quick Quiz)

Read the guide โ†’

๐Ÿ› ๏ธ

Chatbot With File Search: What It Can Really Find in Your Documents, and What to Do With Each File Type (Free File Decision Tree)

Read the guide โ†’

Give your bot answers it can use

Free plan for two websites, no per-seat charges and an AI Agent that answers from knowledge you control. Start on livechatbot.in.