Chatbase: Answer Customer FAQs with an AI Agent

Updated October 18, 2026. Built against the tool’s current screens and checked step by step. How we check this lesson

Most small businesses answer the same twenty customer questions by email every week: hours, prices, service area, how to book, the refund policy. Every answer is already in the FAQ, yet someone still types it out, and anyone who writes after hours waits until morning. Chatbase (chatbase.co) builds an AI agent trained on your own content and puts it on your website as a chat bubble. In this free business-case lesson you price the repeat questions for a made-up bookkeeping firm, then follow Chatbase's own docs to create an agent, train it on a sample FAQ you can download, tell it to answer only from that and to point anyone it cannot help to a person, test it, find the embed code and fix answers from the chat logs. The screenshots on this page and in the lesson come from Chatbase's docs, credited and linked under each one, so they show Chatbase's demo agents. This is my own lesson, not an official Chatbase course.

Start here first: AI Receipt Extractor: Receipts to Google Sheets with n8n (module 1 of Business Automations)

In the course: Part 9 · n8n Automation Toolkit (Module 80 of the full course)

Part of: Give AI Your Company Knowledge (step 10 of 13). Next up: A Support Agent That Answers from Your Own Docs

Part of: AI Apps Worth Knowing (step 47 of 49). Next up: KDAN: Get a Vendor Contract's Key Terms with AI Chatbot

The sample FAQ to train your agent on

This is the file the lesson trains the agent on: 22 questions and answers for Juniper Lane Bookkeeping, a made-up five-person firm, covering hours, location and service area, prices, booking, billing, refunds, cancelling and how to reach a person. Every name, price and phone number in it is made up.

Download the sample FAQ (.txt) Chatbase accepts .pdf, .doc, .docx and .txt files up to 20 MB. Upload it with Add files, or swap in your own FAQ.

The business case: the same questions, every week

Picture Dana Ruiz, who runs Juniper Lane Bookkeeping, a made-up five-person firm. Every week the inbox fills with the same questions. Do you do tax returns? What does monthly bookkeeping cost? Are you open Saturdays? Can I get a refund if I cancel? Every answer is in the firm's FAQ.

Here is the arithmetic, using example numbers you should replace with your own. Say it is about 40 emails a week at 4 minutes each. That is about 2 hours 40 minutes a week. At $45 an hour, that is about $520 a month, and customers who write after 5 PM wait until morning. What Dana needs is an agent that answers from the FAQ straight away, and says so when it does not know.

Who this is for, and what you need

Owners, office managers and support leads at small firms who answer the same customer questions by email or phone. To follow the practice you need nothing. To do it for real you need a Chatbase account. Chatbase's pricing page lists paid plans with a 7-day trial and no free plan, so check it before you start.

A note on the screens: Chatbase is a paid app, so this lesson uses the screenshots Chatbase publishes in its own docs. They show Chatbase's demo agents, such as TaskFlo and Sandbox VR. Your account will show your own agent and your own sources.

Step 1: create the agent and add your FAQ

Sign in to Chatbase, open your dashboard and click New AI agent. On Data sources, the buttons across the top are Add files, Add website, Add text snippet, Add Q&A's, Add Notion Pages and Add Tickets. Click Add files.

In the Add file panel, click the upload area or drag your file onto it, then click Add file source. Chatbase accepts .pdf, .txt, .doc and .docx files up to 20 MB each. A PDF needs selectable text: a scanned PDF with no text layer cannot be trained on. The file's row shows Syncing, then Synced. If it shows Failed, the row gives the reason.

Step 2: tell it how to behave, and what to do when it does not know

Go to Build > Instructions. The General box holds the agent's main instructions: your business, its role and its tone. The Guardrails box holds rules for every conversation. Chatbase's Best Practices page recommends saying explicitly that the agent should answer only from its training data, with a fallback answer for anything else. That fallback is your hand-off to a person. Here is the General text from the lesson, and the Guardrails follow below.

General: Juniper Lane Bookkeeping is a five-person bookkeeping and payroll firm for small businesses. We do monthly bookkeeping, payroll, sales tax filings and year-end packages. We do not prepare tax returns. You answer customer questions on our website. Be friendly, short and clear. Use only our FAQ. If a question is not clear, ask one clarifying question.

On the right, Model configuration holds the Model list and the Temperature slider, from Reserved to Creative. Chatbase's docs say the default is 0, which gives focused, consistent answers. For prices and policies, leave it low. Compare puts versions of your agent side by side and sends them the same message, so you can keep the better one. Click Save changes when you are done.

The Guardrails text to copy

• Only answer from the training data. Never guess prices, dates or policies.
• If the answer is not in the training data, say: "I'm not sure about that one. Please email hello@juniperlane.example or call (555) 014-2290 and a person will help."
• Never ask for bank statements, passwords or Social Security numbers in the chat.
• Stay on topic: Juniper Lane's services, prices, hours, booking and billing.

Step 3: test it in the Playground

Open Playground. It shows your agent the way a visitor sees it, with the Model and a Data sources summary showing whether it is trained. Ask the questions customers really send and check each answer against the FAQ. For the sample file: "Do you do my tax return?" should get a no and a mention of the year-end package. "Are you open on a Saturday in March?" should get 9:00 AM to 1:00 PM. "If I cancel after two weeks, do I get a refund?" should explain the refund for months not started, and that the setup fee is not refunded once setup has begun.

Then ask something the FAQ does not cover, such as "Do you work with nonprofits?". The agent should say it is not sure and point to the email and phone number. If it makes up an answer, tighten the Guardrails and test again. We built this lesson from Chatbase's docs, not a live account, so check your own agent's answers.

Step 4: hand off to a person

The fallback answer is the simple hand-off. Chatbase can also create a ticket with an action: Build > Actions, Create action, then Create ticket under Escalations. Its default When to use says to create a ticket when the user asks for a human, or asks a question the agent does not know, after confirming. It only creates tickets in a helpdesk you have already set up, such as the Chatbase Helpdesk or Zendesk. The pricing page lists the Chatbase helpdesk inbox from the Standard plan.

Step 5: find the embed code, and lock it to your site

In the Playground with Chat bubble selected, click Deploy. The menu offers Website widget, Website iframe, Shopify and WordPress. Website widget is the floating chat bubble. Its panel has a Specific domains switch, Only allow embedding the agent on specific domains, so turn it on and add your own site. Under Widget setup is the script with a Copy button. It goes in your site's head section or just before the closing body tag, and Chatbase has guides for WordPress, Wix, Webflow and other site builders.

Do not install it until the Playground answers hold up. A wrong answer on your website costs more than the email it saved.

Step 6: read the chat logs and fix answers

Activity holds the Chat logs: every conversation your agent had, with filters for date, channel, sentiment, feedback and confidence score. Open a conversation and each answer shows a Revise answer button. It opens an Improve answer panel with the customer's message, the agent's response and an Expected response box. Click Update Answer and the panel says it will create a new Q&A entry in your data sources, so the agent uses your answer next time.

Block 15 minutes a week to read the logs, fix two or three answers and add anything missing to the FAQ.

Mistakes to avoid

• Going live before you test. Ask real customer questions in the Playground first.
• Leaving out the fallback. Without a Guardrails rule, the agent may guess a price or a policy.
• Uploading a scanned PDF. Chatbase needs selectable text.
• Forgetting to update the FAQ when a price changes. The agent is only as current as its sources.
• Leaving Specific domains off, so anyone could put your agent on their site.
• Never reading the chat logs. They show the questions you did not plan for.

Links

• Chatbase: the AI agent builder this lesson teaches.
• Build Your First AI Agent (Chatbase docs): Chatbase's own quick start: create, train, test and deploy.
• Best Practices (Chatbase docs): Chatbase's advice on instructions, cleaner data sources, model choice and correcting answers.
• Chatbase pricing: current plans. On 18 October 2026: Hobby $40 a month ($32 billed yearly) with 700 message credits, Standard $150 a month ($120 billed yearly) with 4,000 credits, Pro $500 a month ($400 billed yearly) with 15,000 credits, and a 7-day trial. No free plan was listed.

Step by step: Chatbase: Answer Customer FAQs with an AI Agent

All 28 steps of the interactive module, in order. In the module you watch each one, then do it yourself in a practice copy that checks your work.

1. The same questions, every week

1. Read the inbox. Dana Ruiz runs Juniper Lane Bookkeeping, a five-person firm (made up). Every week the inbox fills with the same questions: do you do tax returns, what does it cost, are you open Saturdays, can I get a refund. Each one has an answer in the firm's FAQ. Click the inbox.

   Meet Dana. She runs a small bookkeeping firm. Every week, the inbox fills up with the same questions. Tax returns. Prices. Hours. Refunds. Every answer is already in her FAQ.

2. Put a price on it. Say it is about 40 emails a week at 4 minutes each. That is about 2 hours 40 minutes a week. At $45 an hour, that is about $520 a month, and customers who write after 5 PM wait until morning. These are example numbers, so swap in your own. Click the total.

   Say it's forty emails a week, at four minutes each. At forty-five dollars an hour, that's about five hundred twenty dollars a month. And anyone who writes after five waits until morning. Use your own numbers here.

3. Get the sample FAQ. The answers already exist in Juniper Lane's FAQ: 22 questions and answers on hours, prices, service area, booking, billing and refunds. Download the sample file from the card if you want to follow along in your own Chatbase account. It is a plain .txt file, which Chatbase accepts. Click the FAQ card.

   The answers already live in the firm's FAQ. Twenty-two questions and answers. You can download the sample file from this card, and use it in your own Chatbase account.

4. Meet Chatbase. Chatbase (chatbase.co) builds an AI agent trained on your own content, such as files, website pages and Q&A, and puts it on your website as a chat bubble. The plan: train it on the FAQ, tell it to answer only from that, test it, and point anyone it cannot help to a person. The Chatbase screens in this lesson are screenshots from Chatbase's own docs, linked under each one. They show Chatbase's demo agents, so yours will show your own names. Click the Chatbase card.

   Chatbase builds an AI agent, trained on your own content. Then it puts that agent on your website, as a chat bubble. The screens in this lesson are screenshots from Chatbase's own docs. So the names on them are Chatbase's demo agents, not Dana's.

2. Create the agent

5. Click New AI agent. Sign in to Chatbase and open your dashboard. On Agents, click New AI agent.

   Sign in to Chatbase, and open your dashboard.

3. Train it on your FAQ

6. Click Add files. Your agent needs something to learn from. On Data sources, the buttons across the top are Add files, Add website, Add text snippet, Add Q&A's, Add Notion Pages and Add Tickets. Our FAQ is a file, so click Add files.

   Your agent needs something to learn from. Our FAQ is a file.

7. Drop in the FAQ. The Add file panel opens. Click the upload area, or drag the file onto it, and pick juniper-lane-bookkeeping-faq.txt. Chatbase accepts .pdf, .doc, .docx and .txt files up to 20 MB each. A PDF needs selectable text: a scanned PDF with no text layer cannot be trained on. Click the upload area.

   Chatbase takes P D F, Word, and text files, up to twenty megabytes each. If it's a P D F, the text has to be selectable.

8. Click Add file source. Click Add file source. The file uploads, then trains on its own: its row shows Syncing, then Synced. If it shows Failed, the row gives the reason, so fix the file and upload it again.

   The row says syncing, then synced. If it fails, it tells you why. Fix the file, and try again.

4. Tell it how to behave

9. Write the General instructions. Go to Build > Instructions. The General box holds the agent's main instructions: your business, its role and its tone. Replace the example with a short version of yours, for example: *Juniper Lane Bookkeeping is a five-person bookkeeping and payroll firm. We do not prepare tax returns. You answer customer questions on our website. Be friendly, short and clear. Use only our FAQ.* The full text to copy is on this lesson's guide page. Click the General box.

   Go to Build, then Instructions. The General box says who you are, and what the agent does. Describe the firm in two lines. Then say: answer customers, be short, use only our FAQ.

10. Add Guardrails. Guardrails are rules for every conversation. This is where "I don't know" comes from. Write: *Only answer from the training data. If the answer is not there, say you are not sure and give our email and phone number. Never ask for bank statements or Social Security numbers.* Chatbase's own Best Practices page gives the same advice: tell the agent to rely only on its training data and use a fallback answer. Click Guardrails.

    Guardrails are rules for every chat. This is where I don't know comes from. Tell it to answer only from the training data. If the answer isn't there, say so, and give your email and phone number.

11. Leave Temperature low. On the right, Model configuration has the Model list and the Temperature slider, from Reserved to Creative. The docs say the default is 0, which gives focused, consistent answers. For prices and policies, leave it low. Click Temperature.

    On the right is the model, and the temperature slider. The default is zero, which keeps answers focused and consistent. For prices and policies, leave it low.

12. Open Compare. Not sure your new instructions are better? Compare puts versions of your agent side by side and sends them the same message. Click Compare.

    Not sure your change is better? Compare puts two versions side by side, and sends both the same message.

13. Read the compare controls. Each chat box has its own controls: Sync mirrors your message to the other chats, Settings changes the model, temperature and prompt for that box only, and Save copies that box's settings to your main agent. Each model also shows its credit cost per message. Ask both the same FAQ question, keep the better one, then go back to Instructions. Click the controls.

    Each box has its own settings. Sync sends your message to both. Save copies the winner to your main agent. Each model also shows its credit cost. Ask both the same question, and keep the better one.

14. Click Save changes. Back on Instructions, click Save changes. Nothing you typed is applied until you save.

    Back on Instructions, click Save changes. Nothing applies until you save.

5. Test it in the Playground

15. Check it is trained. Open Playground. It shows your agent the way a visitor sees it. The Overview tab shows the Model and a Data sources summary with the training status and total size. The docs screenshot shows Chatbase's demo agent, TaskFlo, marked Trained. Yours shows your own agent. Click Data sources.

    This is your agent, the way a visitor sees it. The data sources box shows whether it's trained. The screenshot shows Chatbase's demo agent. Yours shows your own.

16. Ask real customer questions. Type the questions customers really send, one at a time, and check each answer against the FAQ. Try: "Do you do my tax return?" (should say: No. Juniper Lane prepares a year-end package for your tax preparer.) "Are you open on a Saturday in March?" (should say: Yes, 9:00 AM to 1:00 PM, from February 1 to April 15.) "If I cancel after two weeks, do I get a refund?" (should say: The monthly fee for any month not started is refunded. The setup fee is not, once setup has begun.) We did not run these in a live account, so check your own agent's answers. Click the message box.

    Now ask what customers really ask. Do you do my tax return? Are you open on a Saturday in March? Can I get a refund? Check every answer against the FAQ.

17. Ask one it should not know. Now ask something the FAQ does not cover: "Do you work with nonprofits?" The FAQ does not say, so the agent should say it is not sure and point to the email and phone number. If it makes up an answer instead, tighten the Guardrails and test again. Click the message box.

    Now ask something the FAQ doesn't cover. Do you work with nonprofits? It should say it's not sure, and point to a person. If it makes something up, tighten the guardrails, and test again.

6. Hand off to a person

18. See the hand-off action. Saying "email us" is the simple hand-off. Chatbase can also create a ticket for a person, with an action. In the Playground, Actions lists what the agent can do: the demo agent has Escalate to human. Click Escalate to human.

    Pointing people to your email is the simple hand-off. Chatbase can also open a ticket for a person, with an action.

19. Find Create ticket. Go to Build > Actions, click Create action, then Create ticket under Escalations. The default When to use says to create a ticket when the user asks for a human, or asks a question the agent does not know, after confirming. It only works with a helpdesk you have already set up, such as the Chatbase Helpdesk (listed from the Standard plan on chatbase.co/pricing) or Zendesk, so we stop here. Click Create ticket.

    Go to Build, Actions, Create action. Under Escalations, there's Create ticket. It opens a ticket when someone asks for a human, or asks something the agent doesn't know. It needs a helpdesk set up first, so we'll stop here.

7. Find the embed code

20. Open Website widget. When the answers hold up, find the code for your site. In the Playground with Chat bubble selected, click Deploy. The menu offers Website widget, Website iframe, Shopify and WordPress. Click Website widget, the floating chat bubble.

    When the answers hold up, find the code for your site. That's the floating chat bubble.

21. Lock it to your domain. The Website widget panel opens. Specific domains has a switch: Only allow embedding the agent on specific domains. Turn it on and add your own site, one domain per line, so nobody can put your agent on theirs. Click the switch.

    The widget panel opens. Turn on specific domains, and add your own website. That way nobody else can put your agent on their site.

22. Find the Copy button. Under Widget setup, the script and its Copy button sit ready. This is the code you or your web person paste into the site, in the head section or just before the closing body tag. Chatbase also has step-by-step guides for WordPress, Wix and other site builders. We stop here and do not install it: get the answers right first. Click Copy.

    Under Widget setup is the script, with a copy button. This is what goes on your website. We'll stop here, and not install it. Get the answers right first.

8. Read the chat logs

23. Open a conversation. After it goes live, check what it said. Activity holds the Chat logs: every conversation your agent had. You can filter by date, channel, sentiment, feedback and confidence score. Click the first conversation.

    Once it's live, check what it said. Activity holds the chat logs, every conversation your agent had.

24. Click Revise answer. Here a customer asked for a person, and the agent asked for their name, email and issue before creating a ticket. Below each answer you see its confidence score and a Revise answer button. When an answer is wrong or weak, click Revise answer.

    Here, a customer asked for a person. The agent asked for their details first. Below each answer is a confidence score, and a Revise answer button.

25. Write the right answer. The Improve answer panel shows the customer's message and the agent's response. Type what it should have said in Expected response, and add a Q&A Title. Click Expected response.

    The panel shows the question, and what the agent said.

26. Click Update Answer. Click Update Answer. The panel says it will create a new Q&A entry in your data sources, so next time a customer asks something similar, the agent uses your answer. That is how the agent gets better each week. Click Update Answer.

    It saves your answer as a new Q and A, so next time it gets it right. That's how the agent gets better every week.

9. What it saves, what it costs

27. What this saves you. Click What this saves you.
28. Check the pricing. From chatbase.co/pricing, read 18 October 2026: Hobby is $40 a month, or $32 a month billed yearly, with 700 message credits a month and 1 AI agent. Standard is $150 a month ($120 billed yearly) with 4,000 credits, and lists the Chatbase helpdesk inbox. Pro is $500 a month ($400 billed yearly). The page offers a 7-day trial. Chatbase's FAQ says each answer uses 1 to 5 credits, depending on the model. Click the pricing card.

    Here's the cost, as of October eighteenth. Hobby is forty dollars a month, with seven hundred message credits. Standard is a hundred fifty, and adds the helpdesk inbox. There's a seven-day trial. Check the pricing page before you pick.

Screenshots

Frequently asked questions

Is this an official Chatbase course?

No. This is my own lesson. Chatbase is a separate, third-party product from chatbase.co. The screenshots come from Chatbase's own docs and are credited and linked under each one.

Does Chatbase have a free plan?

Chatbase's pricing page, read 18 October 2026, lists Hobby at $40 a month ($32 billed yearly), Standard at $150 a month ($120 billed yearly) and Pro at $500 a month ($400 billed yearly), with a 7-day trial. It did not list a free plan. Check the page for current plans.

What is a message credit?

Each answer your agent gives uses message credits. Chatbase's FAQ says it is 1 to 5 credits per answer, depending on the AI model. Hobby includes 700 credits a month, Standard 4,000 and Pro 15,000.

What files can I train Chatbase on?

Chatbase's docs list .pdf, .txt, .doc and .docx files up to 20 MB each, with selectable text. You can also add website pages, text snippets, Q&A pairs, Notion pages and support tickets.

How do I stop it from making up answers?

Tell it in Guardrails to answer only from the training data, and give it a fallback answer for anything else. Keep Temperature low, test questions it should not know, and fix weak answers with Revise answer.

Can it hand a customer to a person?

Yes. The simple way is a fallback answer with your email and phone number. Chatbase also has an Escalations action that creates a ticket in a helpdesk you have set up, and a Chatbase Live Chat action for a real-time hand-off.

Start this module free

More in Business Automations

AI Receipt Extractor: Receipts to Google Sheets with n8n

Drop a photo of a receipt into an n8n chat. An AI agent reads it, pulls out the date, vendor and amount, picks a category, and adds a row to Google Sheets.

Free · Live ~12 min Intermediate

n8n AI Email Responder: Draft Gmail Replies Automatically

Build an n8n agent that watches your Gmail, reads each new customer email, and writes a reply as a draft in the same thread, ready for you to check and send.

Free · Live ~12 min Intermediate

Daily Job Alerts: Email Each Tech Their Jobs from Google Sheets with n8n

Every morning at 7, n8n reads a plumbing company's jobs sheet, skips anything not assigned yet, and emails each job to the tech it's assigned to — customer, address…

Free · Live ~10 min Intermediate

Gmail AI Triage: Auto-Label Every Email and Draft Replies

Build an n8n workflow that reads each new email, sorts it as a lead, a client or an invoice with a Text Classifier, adds the matching Gmail label, and drafts a reply to…

Free · Live ~18 min Advanced

All 6 modules in Business Automations

Terms in this lesson

Plain-English definitions, each with an example from a free lesson.

The full AI glossary

Browse related tutorials