Types of chatbots: the 6 types explained, with examples

There are six types of chatbots: menu-based, rule-based, AI-powered, generative AI, hybrid, and voice chatbots. Underneath, every chatbot is one of two kinds. Rule-based chatbots follow a script someone wrote. AI chatbots read what you type and work out an answer.

Every guide counts them differently, from two types to six, so this one covers every type, the other names each one goes by, real examples, and what each one costs a small business in Australia.

Matey does custom chatbot development for Australian businesses: online stores, retailers, tradies, and service businesses. We build AI chatbots every day, so we've seen every type from the builder's side. If you already know you want one, our guide to the best AI chatbot for small business compares the options with prices.

What are the types of chatbots?

The six types of chatbots are menu-based, rule-based, AI-powered, generative AI, hybrid, and voice. Here's how each one works and where you'll meet it:

1. Menu-based chatbots

A menu-based chatbot, also called a button-based chatbot, gives you a list of options and walks you down a flowchart as you press them. It's the phone's "press 1 for accounts" turned into a chat window.

Menu-based chatbots are cheap, quick to set up, and can't say anything wrong. They suit a business whose customers only ever want three or four things. They fall over the moment a question isn't on a button, and real customers' questions usually aren't.

Odoo's built-in Live Chat chatbot is this type: it follows a script of pre-written steps. Our guide to the best Odoo AI chatbot options covers what an Odoo store can use instead.

2. Rule-based chatbots

A rule-based chatbot reads what you type and looks for keywords. Type "refund" and it sends the returns paragraph. Type "can my order get to Darwin by Friday" and, if nobody wrote a rule for those words, you get "Sorry, I didn't understand that."

Someone writes every rule and every answer by hand, and rewrites them whenever anything changes. Tidio's Flows and Facebook Messenger bots built in ManyChat are the common small business versions. Rule-based is also the oldest type of chatbot: the first famous one, ELIZA, was written at MIT in the 1960s and worked by matching keywords.

3. AI-powered chatbots

An AI-powered chatbot, also called an intent-based or NLP chatbot, uses natural language processing to work out what you mean, typos and all, then sends the answer someone wrote for that meaning. "Where's my parcel", "hasnt arrived yet", and "order still not here" all land on the same tracking answer.

An AI-powered chatbot understands far more than a rule-based one, but it still only knows the answers it was given, and it tends to lose the thread on a follow-up question. Most bank and telco virtual assistants built before ChatGPT arrived in late 2022 worked this way.

4. Generative AI chatbots

A generative AI chatbot writes every answer fresh, using a large language model. ChatGPT, Google's Gemini, and Microsoft's Copilot are the general-purpose ones. The same technology now answers customers on business websites: Australia Post's virtual assistant opens by telling you AI writes its answers and can make mistakes.

Generative AI chatbots handle odd questions and follow-ups well, because they read the whole conversation. Their weak spot is the flip side of that: when the answer isn't in what they were given, they can make one up that sounds just as confident as the truth.

5. Hybrid chatbots

A hybrid chatbot mixes the types: buttons for the common questions, AI for everything else, and a hand-over to a person when it's out of its depth. Most good business chatbots are hybrids now.

A matey is a hybrid chatbot. It offers a few starter questions, answers whatever a customer types from the business's own products, prices, and policies, and hands the chat to a person when it should. When we build one, the AI is the easy part. Most of the work is getting the business's real information into it, and writing the rules for when it stops and passes the chat on.

6. Voice chatbots

A voice chatbot talks instead of types. Siri and Alexa are voice chatbots, and so is an AI receptionist answering a business's phone.

Underneath, a voice chatbot is one of the other types with speech recognition on the front and a voice on the back. An old phone menu that asks you to "say 'billing'" is a rule-based voice chatbot. An AI receptionist that takes a booking in a normal conversation is a generative one.

How many types of chatbots are there?

There are six types of chatbots in the longest common list, but you'll see two, three, four, or six depending on who's counting. They're the same chatbots, sliced differently:

You'll also hear AI chatbots split in two: retrieval-based chatbots pick the best answer from a set of written ones, and generative chatbots write a new one each time.

What is the difference between a rule-based chatbot and an AI chatbot?

The difference between a rule-based chatbot and an AI chatbot is who writes the answer. A rule-based chatbot sends an answer someone wrote in advance, word for word. An AI chatbot works out what the customer means and writes an answer from the information it's been given.

A rule-based chatbot is safe and predictable, and stuck on anything off-script. An AI chatbot understands customers, and needs guardrails. Our guide to what a chatbot is explains how an AI chatbot finds its answers.

What are examples of AI chatbots like ChatGPT?

The best-known AI chatbots like ChatGPT are Google's Gemini, Microsoft's Copilot, Meta AI, and Perplexity. They're all general-purpose generative AI chatbots: they know a little about almost everything, and nothing in particular about your business.

A business's chatbot is the opposite kind of AI chatbot. It should know one business inside out: its range, today's prices, what's in stock, and its rules on returns. Ask ChatGPT whether your local camping store has a four-person tent under $500, and it has to go and read the store's website, same as you would. Ask the store's own AI chatbot and it should simply know.

How much does each type of chatbot cost?

A rule-based chatbot can cost nothing to start, a DIY AI chatbot runs about US$40 a month for a small business, and a custom-built hybrid AI chatbot from Matey is a flat $149 AUD a month. Here's what each type costs, from each company's own pricing page on monthly billing:

US dollar prices are converted at the Reserve Bank's rate of about A$1.45 to the US dollar (RBA). DIY AI chatbots charge per conversation or per message, which is cheap in a quiet month and dear in your busiest one. A matey's price is the same every month, with unlimited conversations.

Which type of chatbot does a small business need?

Most small businesses need a hybrid AI chatbot: one that answers from your own products, prices, and policies, offers a few buttons for the common questions, and hands anything tricky to you. That's the type Matey builds. A menu-based chatbot is only enough when customers ask three or four things and nothing else.

• An online store with a big range needs a hybrid AI chatbot. Customers ask which one, which size, and will it fit, and no menu can hold every answer.
• A tradie or service business needs a hybrid AI chatbot on the website, plus a voice chatbot if missed calls are the bigger problem.
• A business with three questions, like opening hours, where to park, and a booking link, can get by with a menu-based chatbot, or a clear contact page.

How can you tell what type of chatbot you're talking to?

You can tell what type of chatbot you're talking to with four messages. We call it the Matey type test, and it works on any chatbot: yours, a competitor's, or one you're thinking of paying for.

1. Ask a real question in your own words, with a typo. "do u deliver to a job site". A menu-based chatbot shows its menu again, and a rule-based one only answers if "deliver" is one of its keywords. An AI chatbot has a go at answering.
2. Ask a follow-up. "what about on a saturday?" Only a generative AI or hybrid chatbot keeps the thread. The others treat it as a brand new question.
3. Ask for something that isn't on offer. "can I get 20% off if I buy two?" A well-built AI chatbot gives the real policy or passes it to a person. A badly built one invents a deal.
4. Ask for a person. A hybrid chatbot takes your details and hands the chat over. The rest loop back to the start.

Which type of chatbot is riskiest for a business in Australia?

A generative AI chatbot is the riskiest type of chatbot for a business, because it's the only type that can make things up. A menu-based or rule-based chatbot can only send answers someone wrote. A generative one can invent a refund rule, a price, or a delivery date if its information has a gap.

That matters, because what your chatbot tells a customer is your business talking. In Canada, Air Canada had to honour a refund policy its website chatbot invented, after a tribunal rejected its argument that the chatbot was responsible for its own words (Law Society Journal). Australia's privacy regulator also says public-facing AI tools like chatbots should be clearly labelled as AI (OAIC).

The fix is in how the chatbot is built, not in avoiding AI:

• Answer only from the business's own information. Products from the real catalogue, rules from the real policy pages.
• Say "I don't know" when the answer isn't there.
• Hand over to a person instead of guessing.

What does a hybrid AI chatbot look like on a website?

A hybrid AI chatbot on a website looks like a chat window with a few starter questions, that answers anything a customer types and passes the hard ones to a person. This is a matey on a made-up workwear store:

Want to see the type of chatbot your own customers would get? Try a matey on our homepage, or ask the one in the bottom corner of this page.