AI Based Chatbot Knowledge Audit: Which of Your Pages and Documents to Feed It, and How to Fix Them First

AI based chatbot answers are only as good as the material behind them. Feed it an outdated price list, a vague returns page and three contradictory FAQs, and it will confidently repeat the confusion to your customers. This audit helps you choose what to include, fix what’s wrong before the bot sees it, and organise your facts so answers stay accurate.

๐Ÿ“š All your content
โ†“
โœ… Keep
โ†“
Current, clear, customer-facing
๐Ÿ› ๏ธ Fix first
โ†“
Vague, outdated, contradictory
๐Ÿ—‘๏ธ Leave out
โ†“
Internal, personal, irrelevant

โšก In short

  • List every source you might feed the bot, then sort each into keep, fix first or leave out.
  • Contradictions are the biggest danger: two pages with different return windows produce random answers.
  • Rewrite key facts as short, specific statements: numbers, dates, conditions, links.
  • Repeat the audit every few months, and whenever a price or policy changes.
Ai based chatbot 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.

Why audit first

The bot repeats whatever you give it, including mistakes

When people set up an AI chatbot, the tempting shortcut is to point it at everything: the whole website, every PDF, old FAQ documents, a few email templates. More material feels like it should mean better answers. In practice, it often means worse ones.

An AI based chatbot that answers from your content treats that content as the truth. If your shipping page says delivery takes 3โ€“5 days and an old blog post says 2โ€“3 days, the bot may quote either, depending on how the question is phrased. If your returns page is vague (“returns accepted in most cases”), the bot’s answer will be vague too. And if an internal document with staff notes or discount codes slips in, the bot may share things customers were never meant to see.

A content audit before training avoids all of this. It takes an afternoon for most small businesses, and it pays off twice: your bot answers better, and your website usually gets clearer for human visitors as well. The guide on training your AI chatbot covers the setup itself; this page is about getting the raw material right.

Step 1 ยท List your sources

Everything that might contain an answer

Start by listing every place where answers to customer questions currently live. Don’t judge yet; just collect. A typical small business finds sources like these:

  • Website pages: home, about, services or products, pricing, delivery or shipping, returns, contact, FAQ.
  • Product information: descriptions, size guides, specifications, care instructions.
  • Documents: brochures, price lists, terms and conditions, course outlines, menus.
  • Saved messages: answers you often paste on WhatsApp or email, and any saved replies in your chat tool.
  • Old content: blog posts, announcements, past offers.
  • Internal notes: staff instructions, supplier details, discount codes.

Write each one in a simple table with three columns: the source, what customer questions it answers, and when it was last updated. You’ll use this table for the next step.

Step 2 ยท Sort

Keep, fix first, or leave out

Go through your list and put each source in one of three groups.

โœ… Keep

Current, clear and meant for customers.

  • Your pricing page, if it’s up to date
  • A specific delivery page with times and costs
  • Product descriptions with real details
  • A FAQ written in plain language

๐Ÿ› ๏ธ Fix first

Useful but flawed: vague, outdated, or in conflict with something else.

  • Returns page that says “in most cases”
  • A price list from last year
  • Two pages with different opening hours
  • Answers buried in long paragraphs

๐Ÿ—‘๏ธ Leave out

Not for customers, or no longer true.

  • Staff notes and internal procedures
  • Discount codes and supplier terms
  • Old offers and expired announcements
  • Anything with personal data

Step 3 ยท Find contradictions

The single most important check

Contradictions are the most common reason AI chatbots give wrong answers, because the bot has no way to know which version is current. Before feeding anything in, check the facts customers ask about most, across all your “keep” and “fix” sources.

Facts to cross-check

– Prices and fees – Delivery times, areas and charges – Return and refund windows and conditions – Opening hours, including holidays – Booking or cancellation rules – Payment methods (UPI, cards, cash on delivery, EMI) – Contact details

How to check quickly

Search your website for each key number or word, for example “days” for delivery, “โ‚น” for prices, “refund” for returns, and note every place it appears. If two places disagree, decide which is right, fix the other, and make a note. Many businesses find at least one contradiction they didn’t know about, often on an old page nobody visits but a bot would happily read.

Step 4 ยท Rewrite for clarity

Turn vague content into answerable facts

AI answers well from short, specific statements. Compare these.

Returns
๐Ÿšฉ VagueWe accept returns in most cases, subject to our policy and the condition of the item.
โœ… Bot-readyUnused items can be returned within 7 days of delivery. Start a return from the link in your order email. Refunds reach your account 5โ€“7 working days after we receive the item.
Delivery
๐Ÿšฉ VagueWe aim to deliver quickly across India.
โœ… Bot-readyDelivery takes 2โ€“4 working days to metro cities and 4โ€“7 to other areas. It’s free over โ‚น999; otherwise โ‚น60.
Hours
๐Ÿšฉ VagueOur team is available during business hours.
โœ… Bot-readyWe’re open Monday to Saturday, 10 am to 7 pm IST. Closed on Sundays and national holidays.

Rules for bot-ready facts

– One fact per sentence where possible. – Numbers, not adjectives: “2โ€“4 working days”, not “quickly”. – State conditions plainly: “unused items”, “over โ‚น999”, “before 1 pm”. – Include the next step: the link, the form, who to contact. – Use customers’ words too: if people say “COD”, mention it alongside “cash on delivery”.

Step 5 ยท Structure

Organise knowledge so the bot finds the right piece

Once the content is clean, organise it. An AI based chatbot finds answers more reliably when each topic lives in its own clearly titled section rather than mixed into long pages.

One topic, one section

Create separate entries for Delivery, Returns, Payments, Opening hours, Booking, and each major product or service. Give each a plain title that matches how customers think: “Delivery times and charges”, not “Logistics”.

Short is good

Each section can be a few short paragraphs. If a topic needs a lot of detail, split it: “Returns: how to start one” and “Returns: refund timing”.

Add the questions

At the top of each section, list two or three ways customers ask about it: “How long does delivery take? When will my order arrive? Do you ship to Kerala?” This helps both AI and keyword matching connect questions to the right answer.

Keep a master copy

Keep your organised knowledge in one document outside the chat tool too. It’s your single source of truth for the website, the bot, and your team’s saved replies.

Step 6 ยท Mark the no-go areas

Tell the bot what it must never answer

Part of the knowledge is knowing where knowledge stops. Write a short list of topics where the bot should not answer from content at all and should instead hand over to a person:

  • Medical, legal or financial advice
  • Refund disputes and complaints
  • Custom quotes and exceptions to policy
  • Anything about a specific customer’s account or order details it can’t verify
  • Topics outside your business entirely

Add these as instructions or handover rules in your chat tool. LiveChatBot’s AI Agent lets you give guidance alongside your knowledge, and handovers send these conversations to your inbox. The chatbot customer service rulebook shows how to sort topics into what the bot answers and what goes to people.

Special cases

Products, menus and things that change often

Some content needs extra care because it changes frequently or is very detailed. Here’s how to handle the common cases when preparing an AI based chatbot.

Large product catalogues

You don’t need every product on day one. Start with best-sellers and the categories people ask about most, plus general facts that apply to all products (sizes, materials, care, warranty). Add more as conversations show demand.

Daily menus and stock

If availability changes daily, don’t put it in the bot’s knowledge, because it will be wrong by tomorrow. Instead, give the bot a general answer (“Our menu changes daily; today’s specials are on our Instagram”) or hand stock questions to a person.

Seasonal offers

Add offers with clear start and end dates in the text, and set a reminder to remove them. An expired offer quoted by the bot creates awkward conversations.

Prices that vary

If prices depend on size, quantity or location, give ranges and explain what changes them: “Custom cakes start at โ‚น1,100 for 1 kg; designs and sizes change the price.”

Audit checklist

Run this before you train, and every few months after

Knowledge audit

After the audit

Testing, and keeping it current

With clean knowledge in place, test the bot from your phone using your customers’ real wording, typos included. Ask about each key fact and check the answer matches your master copy word for word on numbers. Then ask something you deliberately left out; the bot should say it doesn’t know or offer a person.

When to re-audit

– Whenever a price, policy or hours change. Update the master copy, the website and the bot together. – Before busy seasons, such as festivals or admissions, when special hours or offers apply. – Every three months as a routine, even if nothing obvious has changed.

Watch the conversations

Weekly, read a sample of chats. Wrong answers almost always trace back to a gap or an outdated line in the knowledge. Fix the source, not just the bot.

You can set up the AI Agent with your audited knowledge on LiveChatBot’s free plan, which includes a small monthly allowance of AI replies per website, with the keyword bot answering from saved replies when the allowance is used. Start on livechatbot.in, and see the pricing page for details.

Frequently asked questions

What should I feed an AI based chatbot first?

Your most-asked facts: prices, delivery, returns, hours, payments and booking, rewritten as short, specific statements, plus your main product or service details.

Can I just point the bot at my whole website?

You can, but audit first. Old pages, contradictions and vague wording lead to wrong or vague answers.

How often should I update the bot’s knowledge?

Whenever a price, policy or hours change, before busy seasons, and as a routine every few months.

Should internal documents go into the bot?

No. Staff notes, discount codes, supplier terms and anything with personal data should be left out.

What if the bot still gives a wrong answer?

Find the source of the wrong fact, fix it there, update the bot, and test again. Most errors trace back to the content.

Keep reading

Related guides

๐Ÿ“˜

Advanced Chatbot Setup: 9 Power-User Techniques Most Small Businesses Never Switch On

Read the guide โ†’

๐Ÿงญ

Advanced AI Chatbot or Simple One? A Capability Ladder to Find the Level Your Business Actually Needs

Read the guide โ†’

๐Ÿ› ๏ธ

About Chatbots: 40 Plain Answers to the Questions Small Business Owners Actually Ask

Read the guide โ†’

Audit this afternoon, train tonight

Clean knowledge makes every answer better. Set up your AI Agent free on livechatbot.in.