Can a chatbot answer customer questions from my own FAQ?

Yes. A chatbot can answer customer questions from your own FAQ, policies and documents, replying only from that material and passing anything it can't answer to your team. How well it works depends mostly on how accurate and current your FAQ is.

How does a chatbot actually use my FAQ to answer questions?

A chatbot answers from your FAQ by matching what a customer asks to the closest information you have written down, then composing a reply from it. Modern ones do this by meaning rather than exact keywords, so "where's my parcel?" and "has my order shipped yet?" can both lead to your delivery answer.

In practice the process looks like this:

You supply your FAQ, policies and any other reference documents.
The system splits them into small chunks and indexes them.
When a customer writes in, it finds the chunks most relevant to the question.
It drafts a reply using only that material.

The key instruction is to answer only from your content. Without that limit, a chatbot may fill gaps with plausible-sounding guesses, which is the last thing you want when the topic is refunds or safety.

Your FAQ therefore becomes the chatbot's entire knowledge. If an answer isn't in it, a well-set-up bot should say so and pass the question to a person rather than improvise. So source material quality matters more than tool choice.

What should my FAQ contain before I connect a chatbot?

Start by checking that your FAQ is current, specific and written the way customers actually ask. A chatbot repeats what it finds, so out-of-date answers become out-of-date replies at speed.

Before connecting anything, go through these checks:

Dates and prices: remove or update anything that has changed, such as delivery times, opening hours or fees.
Vague wording: replace "usually" and "may" with the real rule and its exceptions.
One answer per question: if two pages say different things, decide which is right.
Real customer phrasing: copy questions from your inbox, chat logs and messaging threads rather than inventing them.
Policies in full: include returns, cancellations and complaints procedures, not just summaries.

It also helps to add the questions your FAQ doesn't cover yet. Look through a month of messages and note anything you answered more than a few times. Those repeat questions are the strongest candidates to write up.

Finally, decide what the bot must never handle alone, such as legal complaints, medical advice or payment disputes, and write that down as a clear rule for handover to a person.

What happens when the chatbot doesn't know the answer?

A properly set-up chatbot admits it doesn't know and hands the conversation to a person. This is the most important behaviour to test, because a confident wrong answer does more harm than a short delay.

Good handover usually includes:

A clear message: something like "I'm not sure about that, so I'll pass you to the team."
The conversation so far: your staff see what was asked, so customers don't repeat themselves.
A way to reach the customer: an email address, phone number or the same messaging thread.
Honest timing: tell people when you'll reply, especially outside working hours.

You can also set triggers for automatic handover. Examples include an angry tone, words like "complaint" or "refund", a customer asking for a human, or the same question being asked twice without success.

Review the handed-over questions weekly. Each one shows a gap in your FAQ. Write the missing answer, add it to the source material, and the same question is handled automatically next time.

Over a few weeks, this loop steadily shrinks the enquiries reaching your inbox.

Can I use one FAQ across email, website chat and messaging apps?

Yes. The same set of answers can serve every channel, which is one of the main advantages of keeping a single source of truth. What changes is the format, not the facts.

Think about each channel separately:

Website chat: customers expect quick, short replies and often want a link to the right page.
Messaging apps: messages should be brief and conversational. Long paragraphs are awkward on a phone.
Email: replies can be fuller and more formal, and a signature and greeting are expected.

Some channels have their own rules. Business messaging apps, for instance, often have requirements around consent and message templates, so check the current terms before sending anything proactive.

Keep one master FAQ and update it in one place. If your opening hours change, you shouldn't be editing three different documents.

Also consider data protection. Customers may share names, order numbers or addresses in a chat. Under UK GDPR you should be clear about what you collect, why, and who processes it, and your privacy notice should mention the chatbot. Avoid asking for payment card details in any chat channel.

How do I test and keep the chatbot accurate?

Test it with real questions before any customer sees it, then keep checking after launch.

Start with a test list of 30 to 50 questions taken from your actual inbox. Include awkward ones: typos, two questions in one message, and questions your FAQ doesn't answer. Mark each reply as correct, partly right or wrong, and fix the source material behind any miss.

Once live, build a simple routine:

Weekly: read a sample of conversations and the handed-over questions.
After any change: if prices, policies or services change, update the FAQ the same day.
Monthly: look for the most common questions and check the answers are still clear.
Always: make it easy for customers to say a reply wasn't helpful.

Be open that customers are talking to an automated assistant, and give them an easy route to a person. This builds trust and is good practice.

Set a modest first goal, such as handling your five most repeated questions reliably. Expand once you're confident, rather than switching on everything at once. A small, accurate bot is more useful than a broad, unreliable one.

Frequently asked questions

Will the chatbot make things up?

It can if left unrestricted, so set it to answer only from your content and hand over when unsure. Test it with real questions before launch.

Do I need technical skills to prepare my FAQ?

Not really, as it is mainly clear writing. The technical setup is usually handled by a tool or provider.

How long should my FAQ be?

Accuracy matters more than length. Start with your most repeated questions and add more over time.

Related on B.A.I.S.

B.A.I.S. builds custom support chatbots that answer your customers 24/7 on your website, WhatsApp or Telegram, using your own FAQs, manuals and policies. Anything the chatbot can't answer is passed to your team.

Need this done for you? Custom AI agent or support chatbot