AI chatbot examples: 12 good conversations for small businesses, and why each one works

This is for owners who want to see what a good AI chatbot conversation actually looks like before buying one. Below are twelve short example chats across common small business situations, each followed by one line on why it works, plus the bad patterns to avoid.

By Ritchie, ReplyKit · Published 15 October 2026 · 10 min read

Quick answer

A good AI chatbot conversation answers the actual question from the business's own information, gives a price or range when one exists, offers a clear next step such as real booking times, admits when it does not know and passes the chat to a person, and replies in the customer's language. The twelve examples below show each pattern.

What a good conversation hasBooking examplesQuestions and qualificationVoice notes, photos, after hoursWhen the AI should step backBad examples to avoidWhere ReplyKit fitsFAQ
About these examples. Every conversation on this page is an illustrative example written by us to show a pattern. None is a real customer chat. The businesses, names, times and prices are made up.

What makes a good AI chatbot conversation?

A good conversation leaves the customer with a correct answer and a next step, in as few messages as possible. Five habits separate good chatbots from annoying ones.

What do good booking conversations look like?

The best booking chats answer the question, then offer specific open times instead of a link, and confirm in the same chat.

1. Dental clinic: price and booking in one message

Illustrative example.
Customer: Hi, do you do scaling and polishing? How much, and can I come this week?
Agent: Yes, we do. Scaling and polishing is $90 and takes about 45 minutes. This week I have Thursday at 10:00 am or 3:30 pm, or Saturday at 9:00 am. Which suits you?
Customer: Thursday 3:30
Agent: Booked for Thursday at 3:30 pm. Could I have your full name for the appointment? You will get a reminder the day before.

Why it works: it answers all three questions in one reply and offers real times, so booking takes two messages, not ten.

2. Restaurant: a "no" turned into a booking

Illustrative example.
Customer: Table for 6 this Saturday at 8pm?
Agent: Saturday at 8:00 pm is fully booked for six, sorry. I can do 7:15 pm or 8:45 pm for six people. Would either work?
Customer: 8:45 is fine
Agent: Done, a table for six on Saturday at 8:45 pm. What name should I put it under?

Why it works: instead of a dead end, it offers the nearest alternatives, which is what a good host would do.

3. Tutoring center: a trial class for a child

Illustrative example.
Customer: Do you have maths classes for Year 5? Can my son try one first?
Agent: Yes. Year 5 maths runs on Tuesdays and Thursdays, 5:00 to 6:30 pm, and the first trial class is $15. To place him in the right group, how is he finding maths at school at the moment, comfortable or struggling a bit?
Customer: Struggling with fractions
Agent: Thanks, that helps. The Thursday group is working on fractions this month. Shall I book him for a trial this Thursday at 5:00 pm?

Why it works: one natural question qualifies the lead and makes the recommendation feel personal.

What do good answers to price and product questions look like?

Good answers give the figure from the price list, explain what changes it, and then offer the next step without pushing.

4. Hair salon: a price that depends on the customer

Illustrative example.
Customer: How much for highlights? My hair is shoulder length.
Agent: Highlights for shoulder-length hair start at $120 and take about two and a half hours. The final price depends on thickness and the look you want, and your stylist confirms it before starting. Would you like to see this weekend's open times?

Why it works: it gives a real starting price instead of "contact us", and is honest about what can change it.

5. Property agent: availability plus qualification

Illustrative example.
Customer: Is the 2-bedroom on Park Road still available? Can I view?
Agent: Yes, the 2-bedroom on Park Road is available at $1,800 a month. Viewings run weekday evenings after 6:00 pm and on weekends. Before I book one, when are you hoping to move in, and is it for you alone or with family?
Customer: Next month, me and my partner
Agent: Great. I have Wednesday at 6:30 pm or Saturday at 11:00 am. Which works?

Why it works: the agent learns timing and need in one friendly question, so the human agent knows the lead is warm before the viewing.

6. Online store: honest about what it cannot see

Illustrative example.
Customer: Where is my order #4471? It's been a week.
Agent: Sorry for the wait. I cannot see live order tracking from this chat, so I have passed order #4471 to the team and they will reply here. For reference, standard delivery takes 3 to 5 working days from dispatch, and dispatch can take up to 2 days.

Why it works: it does not invent a tracking status; it captures the order number, hands over, and still answers what it does know.

How should a chatbot handle voice notes, photos and late-night messages?

Customers on messaging apps send voice notes and photos, and they message at midnight. A good agent treats all three as normal.

7. Car workshop: a voice note

Illustrative example.
Customer: [Voice note, 0:24: says the car makes a grinding noise when braking and asks if someone can look at it today]
Agent: Thanks for the voice note. A grinding noise when braking is worth checking soon. We can see the car today at 4:00 pm or tomorrow at 10:00 am. The brake inspection is free, and any parts are quoted before we start. Which time works?

Why it works: it understood the voice note and replied in text with a clear choice. More in handling voice notes with AI.

8. Plumber: a photo

Illustrative example.
Customer: [Photo of water dripping from a pipe joint under a kitchen sink] How much to fix this?
Agent: Thanks for the photo. It looks like the leak is at a joint under the sink. Repairs like this usually start at $80 for the call-out plus parts, but the plumber needs to confirm the cause on site. The earliest slot is tomorrow between 9:00 and 11:00 am. Shall I book it?

Why it works: it uses the photo to give a useful range but does not pretend to diagnose from a picture. See how AI agents understand photos.

9. Physiotherapist: a message at 11:48 pm

Illustrative example.
Customer (11:48 pm): Is it too late to book for tomorrow morning? My back is killing me.
Agent: Not too late. I'm sorry about your back. Tomorrow I have 9:00 am or 11:30 am for a first assessment, which takes 45 minutes. Shall I book one of those?
Customer: 9 please
Agent: Booked for 9:00 am tomorrow. Please arrive 10 minutes early to fill in a short form.

Why it works: the booking is made while competitors are asleep. Harvard Business Review found firms that contacted leads within an hour were nearly seven times as likely to qualify them as those that waited even an hour longer (HBR, 2011).

When should a chatbot hand over to a person?

When the answer is not in its knowledge, when money or a complaint is involved, or when the customer asks. Stepping back well is part of a good conversation, not a failure.

10. Yoga studio: the answer is not in its knowledge

Illustrative example.
Customer: Are your classes safe in the third trimester of pregnancy?
Agent: I don't have that information, and it's not something I should guess about. I've asked the team to confirm and they'll reply here. Could I take your name in case they want to call you?

Why it works: no invented answer on a safety question; the owner is alerted and the gap can be filled once. A Canadian tribunal held Air Canada responsible when its chatbot gave wrong advice (CBC News, 2024), which is why "I don't know" beats a guess.

11. Home goods shop: a complaint handed to a human

Illustrative example.
Customer: This is the second time my order arrived broken. I want a refund.
Agent: I'm sorry, that's frustrating, especially twice. Refunds are handled by our team, so I've passed this to them now. Could you send a photo of the damage so they have it ready?
[The owner takes over the chat and the AI pauses for this conversation.]
Owner: Hi, it's Mei from the shop. I've seen the photos. I'm refunding you in full today and I'm sorry again.

Why it works: it apologizes, promises nothing it cannot decide, gathers evidence, and gets a person in fast. See how an agent decides when to hand over.

12. Cleaning service: a customer writing in Malay

Illustrative example.
Customer: Boleh saya tahu harga pakej pembersihan rumah 3 bilik?
Agent: Boleh! Pakej pembersihan rumah 3 bilik ialah RM 180 dan mengambil masa lebih kurang 4 jam. Slot terdekat ialah hari Selasa pukul 10 pagi atau hari Rabu pukul 2 petang. Yang mana sesuai?

Why it works: the customer asked the price of a three-bedroom cleaning package, and the agent answered in Malay with the price, the duration and two open slots, without asking them to switch to English.

What do bad chatbot conversations look like?

Bad chatbots fail in predictable ways. If a tool you are testing does any of these, it will cost you customers.

Bad patternWhat the customer seesWhat good looks like
Menu loop"Please choose 1, 2 or 3" after asking a real questionReads the question and answers it
Invented answerA price or policy that is not yours"I don't know, I've asked the team"
Half an answerAnswers the price, ignores "can I come Friday?"Answers both parts
Link dump"Book here: [link]" with no timesOffers two or three real slots
Form in a chatAsks for name, email, phone, address up frontAnswers first, asks only what is needed
No way outCannot reach a personHands over when asked
Wall of textFive paragraphs on a phone screenTwo or three short sentences

Most of these come from scripted bots that follow a menu; the chatbot vs AI agent explainer covers why. Some come from AI that answers from the open internet; our guide to AI hallucination explains the fix. If someone asks whether they are talking to a bot, answer honestly; in the EU, Article 50 of the AI Act requires people to be told they are interacting with an AI system unless it is obvious, from 2 August 2026 (EU AI Act, Article 50).

Where ReplyKit fits

ReplyKit is the AI agent we build, and every pattern above is one it is designed for on WhatsApp and website chat. It answers only from your content (website, PDFs, price lists, Q&A pairs), offers real open slots and books them with double-booking protection, sends a reminder about 24 hours before, understands voice notes and photos, replies in over 30 languages, and escalates when it does not know, listing the question so you can teach it once. You can take over any chat in one click and hand it back. WhatsApp connects through Quick Connect, an unofficial QR pairing that is not sanctioned by Meta, so use a dedicated number.

What it does not do: look up live orders in Shopify or another store system (example 6 is exactly how it would behave), take payments in the chat, or send broadcasts. To see these patterns with your own content, try the live demo on our homepage or start a 7-day free trial with no credit card. Plans are on the pricing page.

Frequently asked questions

What is an example of an AI chatbot?

A chatbot on a clinic's website or WhatsApp that answers "how much is a cleaning and can I come Thursday?" with the price from the clinic's list and two open times, then books the slot. AI agents such as ReplyKit work this way on WhatsApp and website chat.

What makes a good chatbot conversation?

It answers the actual question from the business's own information, offers a clear next step, admits when it does not know, hands over to a person when needed, and keeps replies short. The fewer messages to a correct answer, the better.

Are these real customer conversations?

No. Every example here is illustrative, written by us to show a pattern. The businesses, names, prices and times are made up.

How do AI chatbots handle questions they cannot answer?

Good ones say they do not know, pass the chat to a person and alert the owner. Poor ones guess, which can mean quoting a price or policy the business never set.

Can an AI chatbot understand voice notes and photos?

Some can. Customers on WhatsApp send both all the time, so check this during any trial. ReplyKit understands voice notes and photos on every plan.

Should a chatbot say it is an AI?

Yes, if asked, and it is good practice to make it clear anyway. In the EU, the AI Act requires people to be told they are interacting with an AI system unless it is obvious, from 2 August 2026.

See these patterns with your own content

Add your website and price list, then message your agent like a customer would. 7-day free trial, no credit card needed.

Start free
Written by Ritchie, ReplyKitPart of the small team in Malaysia that builds and runs ReplyKit. Writes about WhatsApp automation, AI customer service and small business lead response.

How we know this. We wrote these examples ourselves, based on the situations we design and test our own agent against: two-part questions, voice notes, photos, after-hours bookings, missing knowledge and handoffs. They are illustrations, not customer chats, and no figure in them is a real price. The outside sources below were opened in October 2026. We make ReplyKit, so weigh our view accordingly; any well-built AI agent should handle these patterns, and if your inquiries are rare and simple, a good FAQ page may be all you need. About ReplyKit.

Sources and further reading

  1. Harvard Business Review, The Short Life of Online Sales Leads (Oldroyd, McElheran, Elkington, 2011) · firms contacting leads within an hour were nearly seven times as likely to qualify them
  2. CBC News, Air Canada found liable for chatbot's bad advice (February 2024) · a business is responsible for what its chatbot says
  3. EU AI Act, Article 50: Transparency obligations · telling people they are interacting with an AI system, applying from 2 August 2026

ReplyKit is an independent product and is not affiliated with or endorsed by Meta or WhatsApp. All example conversations are illustrative and written by ReplyKit; they are not real customer chats, and the businesses, names and prices are made up. Third-party sources and any prices mentioned were checked in October 2026 and may change. This article is general guidance, not legal advice.