AI in customer service: build and deploy your own chatbot
Many businesses now use AI chatbots to answer customer questions day and night.
- Explain the difference between a rule-based and an AI (LLM-based) chatbot
- Write a system prompt that sets a bot's role, tone, scope and fallback
- Build a knowledge base and test a chatbot for wrong answers and hand-off
- Apply the EU AI Act duty to tell users they are talking to AI
Read the lesson
Many businesses now use AI chatbots to answer customer questions day and night. In this lesson you learn how a modern chatbot works, how to build and test one without coding, and why the law says customers must know they are talking to AI.
01Customer service never sleeps
A chatbot answers routine customer questions in plain language, day or night, so people can focus on the hard cases.
Eurostat (2025): 55% of EU businesses used AI, cloud or analytics in 2023.
02What a customer-service chatbot does
A chatbot is an AI assistant that answers customer questions in plain language, day or night.
- Answers common questions. Opening hours, order status, returns, 'how do I…': the questions a business hears hundreds of times, answered in seconds.
- Works around the clock. Customers get help at midnight or on a holiday, without a person waiting by the keyboard.
- Frees people for hard cases. Staff stop repeating the same answers and focus on complex, sensitive or high-value conversations.
Tip: let the bot handle the routine questions so people can focus on the tricky ones.
03Two kinds of chatbot
Not all chatbots are equal. Knowing the difference helps you build a better one.
Rule-based (older)
- Follows fixed buttons and scripts
- Predictable, but limited
- Gets stuck if a question isn't in its menu
AI / LLM-based (modern)
- Understands questions in natural language
- Answers from the knowledge you give it
- Far more flexible, but can 'hallucinate'
Best practice: a knowledge-grounded bot answers only from your documents.
04How a modern AI chatbot works
Behind the scenes, a knowledge-grounded chatbot follows four simple stages.
- Feed it knowledge. Upload FAQs, policies and product info. This is its source of truth.
- Set its behaviour. Give it a name and a system prompt that defines its tone, scope and limits.
- It answers customers. It matches each question to your content and replies in seconds, 24/7.
- You review and improve. Check where it fails, refine the prompt and fill gaps in its knowledge.
Quality comes mostly from the knowledge and the prompt you give it. Builds toward EX 4.1.
05Writing a good system prompt
The system prompt is the bot's job description. Write it as if briefing a new employee on day one.
| Part | What it does | Example (a small bakery) |
|---|---|---|
| Role and tone | Says who the bot is and how it should sound | 'You are the friendly support assistant for a small bakery. Be warm, concise and polite.' |
| Scope and limits | Says what to cover and what to refuse | 'Only answer questions about our products, hours and orders. For anything else, offer a human.' |
| Fallback | Says what to do when it doesn't know | 'If you are unsure, say so and suggest contacting the team. Never guess.' |
| AI disclosure | Makes sure users know it is an AI | 'Start each chat with: You're chatting with our AI assistant.' |
A clear system prompt is the difference between a good bot and a bad one.
06What to put in the knowledge base
Your chatbot is only as good as the information you give it. Cover these four areas.
- Frequent questions. The real questions customers ask most often, with clear, correct answers.
- Policies. Returns, refunds, delivery, privacy: the rules customers need to understand.
- Products and services. What you offer, key details, prices and what makes each option different.
- Contact and escalation. How and when to reach a human, so the bot can hand over smoothly.
Tip: write your FAQ the way customers actually ask, not in formal company language.
07Build a chatbot, step by step
You can build a working chatbot in under an hour, with no coding.
- Create the bot. Sign up to a no-code tool like Chatbase and start a new chatbot project.
- Upload your knowledge. Add a FAQ document, policies or a website link so the bot answers from real content.
- Name it and prompt it. Give it a friendly name and a system prompt: its role, tone, what it should and shouldn't do.
- Test, then deploy. Ask real customer questions, fix weak spots, then embed it on a website with the code snippet it gives you.
Tool: Chatbase.co (free tier). This is exactly what you will do in EX 4.1.
08Test your chatbot before customers do
A chatbot must be tested like any product. Probe it before it goes live.
- Everyday and edge cases. Run the ten most common questions, then awkward, vague or out-of-scope ones to see how it copes.
- Hunt for wrong answers. Confident but incorrect replies do the most damage. Fix these first.
- Check the hand-off. Make sure it offers a human when it should, and never invents information.
Tip: keep a simple log of questions it failed. That list is your improvement plan.
09Good chatbot vs bad chatbot
The same technology can delight or annoy customers. Behaviour is what separates them.
A good chatbot
- Answers accurately from real company info
- Clearly says it is an AI assistant
- Admits when it doesn't know
- Offers a human for complex issues
A bad chatbot
- Invents confident but wrong answers
- Pretends to be a human
- Traps customers in dead-end loops
- Never lets you reach a real person
Aim for the left column: it builds trust instead of frustration.
10Your legal duty: disclose the AI
Under Article 50 of the EU AI Act, people must be clearly told when they are interacting with an AI system, unless it is obvious. This applies from 2 August 2026.
- Add a visible line such as 'You're chatting with our AI assistant'
- Never let the bot pose as a human
- Build the habit of disclosing AI into everything you make now
AaKey terms (6)
Chatbot
An AI assistant that answers questions in natural language, often for customer service.
LLM
A 'large language model': the AI behind modern chatbots that predicts natural-sounding text.
System prompt
The instructions that set a chatbot's role, tone, scope and limits.
Knowledge base
The documents (FAQs, policies, product info) a chatbot answers from.
Escalation
Handing a conversation over to a human when the bot can't or shouldn't answer.
Transparency duty
The EU AI Act rule to tell people clearly when they are dealing with AI.
Practise
?Check your understanding
Answer these in your own words — out loud, in a notebook, or with a study partner.
- What is the difference between a rule-based and an AI (LLM-based) chatbot?
- Why does grounding a chatbot in your own documents reduce wrong answers?
- What three things should a good system prompt always define?
- Under the EU AI Act, what must you tell customers who chat with your bot?
Key takeaways
- Modern AI chatbots answer natural questions from the knowledge you give them
- Ground the bot in your own documents to keep answers accurate and on-brand
- A clear system prompt (role, scope, fallback) is what makes a bot reliable
- Always test for wrong answers and a clean hand-off to a human
- By law (EU AI Act, Art. 50) you must tell users they are talking to AI
Your next steps
- 1
Lesson quiz
5 questions, 70% to pass, retake as often as you like.
Create a free account to take the quiz - 2
Hands-on exercise EX 4.1
You build a customer-service chatbot for a fictional business, configure it with a name and system prompt, and test it with 10 real customer questions. This shows an employer or client that you can build and check a working AI tool.
Open the exercise →