Table of Contents
- Key takeaways
- What is a customer support chatbot?
- Is a customer support chatbot the same as a customer service chatbot?
- How is a chatbot different from live chat?
- How does a customer support chatbot work?
- What does a chatbot need before it can answer?
- What are the types of customer support chatbots?
- Rule-based chatbots
- Intent-based chatbots
- Knowledge-trained AI chatbots
- AI agents
- What can a customer support chatbot handle?
- What can a customer support chatbot not handle?
- A company answers for what its chatbot says
- Customers give a chatbot few chances
- When should a customer support chatbot hand off to a human?
- Worked example: one conversation, three outcomes
- What are the benefits of a customer support chatbot?
- Which type of customer support chatbot do you need?
- Where a knowledge-trained chatbot such as SiteGPT fits
- Common customer support chatbot mistakes
- Frequently asked questions
- Basics
- What is a customer support chatbot?
- What is the difference between a customer support chatbot and a customer service chatbot?
- Is a customer support chatbot the same as live chat?
- How it works
- How does a customer support chatbot know the answers?
- Does a customer support chatbot use AI?
- Can a customer support chatbot give a wrong answer?
- Limits and handoff
- Can a chatbot replace customer service agents?
- When should a chatbot transfer a customer to a person?
- Do you have to tell customers they are talking to a chatbot?
- Is a company responsible for what its chatbot says?
- Cost and setup
- How much does a customer support chatbot cost?
- What do you need before launching a customer support chatbot?
- Which type of chatbot is best for a small business?
- Sources
Created by
Do not index
Created time
Oct 25, 2023 11:33 PM
A customer support chatbot is software that answers customer questions in a chat window, using scripted rules, an AI model trained on the company's own content, or both, and passes the conversation to a person when it cannot help.
TL;DR
- A customer support chatbot is also called a customer service chatbot. The two names mean the same thing.
- It works in five steps: it reads the message, works out the question, finds the answer, replies, and then closes the chat or hands it to a person.
- There are four types: rule-based, intent-based, knowledge-trained AI, and AI agents. They differ in where the answer comes from.
- A chatbot handles questions that have a written, repeatable answer. Refund disputes, complaints, and anything with money or legal risk go to a person.
- In a 2026 Gartner survey, 49% of customers said they would use a company chatbot, but only 27% would try one again after a bad experience. Reliable answers matter more than wide coverage.
Key takeaways
Question | Short answer |
What is it? | Software that answers customer questions in a chat window and hands off to a person when needed |
How does it work? | Read the message, work out the question, find the answer, reply, then close or hand off |
What types exist? | Rule-based, intent-based, knowledge-trained AI, and AI agents |
What can it handle? | Routine questions with a documented answer, at any hour |
What can it not handle? | Disputes, complaints, undocumented questions, and advice that carries legal or financial risk |
When should it hand off? | When the customer asks for a person, the answer is not in its content, a wrong answer is costly, or it has failed twice |
What is a customer support chatbot?
A customer support chatbot is a program that holds a typed conversation with a customer and answers the question without a human agent. It sits in a chat window on a website, inside an app, or in a messaging channel such as WhatsApp or Slack.
The chatbot gets its answers from one of two places. A scripted chatbot follows rules a person wrote. An AI chatbot reads the company's help center, website, and documents, then writes an answer from that content.
A good customer service chatbot also knows its limits. It says when it has no answer and passes the conversation to a person with the messages so far.
Is a customer support chatbot the same as a customer service chatbot?
Yes. A customer support chatbot and a customer service chatbot are the same thing. Vendors and teams use the two words interchangeably.
Other names for the same tool are support bot, help desk chatbot, and customer care chatbot. This guide uses "customer service chatbot" from here on.
How is a chatbot different from live chat?
A chatbot is software that writes the reply. Live chat is a person typing the reply in the same kind of chat window. Many businesses run both: the chatbot answers first and a person takes over when needed. For a full difference table and the same question answered by live chat and by two kinds of chatbot, see live chat vs chatbot.
The guide to customer chat support covers the human side of the chat window, including queue times, staffing models, and outsourced teams.
How does a customer support chatbot work?
A customer service chatbot works in five steps, from the customer's message to the saved transcript. The steps below describe a knowledge-trained AI chatbot.
- The customer sends a message. The customer types a question in the chat window. Some chatbots show starter buttons with common topics.
- The chatbot works out the question. A language model reads the message and identifies what the customer is asking, even when the wording is unusual or has typos.
- The chatbot finds the answer. It searches the company's content for the passages that match the question. This search step is called retrieval.
- The chatbot writes the reply. The language model writes an answer from the retrieved passages and usually links the source page. If nothing relevant was found, a well-configured chatbot says so.
- The chat closes or is handed off. The customer can end the chat, ask a follow-up, or ask for a person. The transcript is saved for the team to review.
Steps 3 and 4 together are called retrieval-augmented generation, or RAG. The guide to what RAG is explains the method in detail.
A rule-based chatbot skips steps 2 to 4. It shows buttons, the customer picks one, and the chatbot sends the reply a person wrote for that button.
What does a chatbot need before it can answer?
A chatbot needs written answers before it can answer anything. For an AI chatbot, that means a help center, product pages, policy pages, or documents that already state the answer. For a rule-based chatbot, it means a script for every path.
The chatbot is only as accurate as that content. An out-of-date returns page produces an out-of-date answer. The guide to chatbot training covers how to prepare the content.
What are the types of customer support chatbots?
There are four types of customer service chatbots: rule-based, intent-based, knowledge-trained AI, and AI agents. Each type gets its answer from a different place.
Type | Where the answer comes from | Good at | Weak at | Setup work |
Rule-based | Buttons and scripted replies a person wrote | Fixed flows that must run the same way every time | Any question nobody scripted | Build and maintain every branch by hand |
Intent-based | A model matches the message to one of a list of intents, each with a written reply | A known set of frequent questions in free text | Questions outside the intent list | Define intents and supply example phrases for each |
Knowledge-trained AI | A language model writes the reply from the company's own content | The long tail of questions the content already answers | Questions the content does not cover, and tasks that need another system | Connect the content and test the answers |
AI agent | A language model that also calls other systems to look up or change data | Tasks such as checking an order or updating an account | Higher risk, because a mistake changes real data | Connect each system, set permissions, and review actions |
Rule-based chatbots
A rule-based chatbot follows a decision tree. The customer clicks a button, and the chatbot sends the reply written for that button. It never says anything unexpected, and it cannot answer a question nobody planned for.
Intent-based chatbots
An intent-based chatbot lets the customer type freely. A model matches the message to the closest intent on a list, such as "track my order" or "reset my password", and sends the reply written for that intent. Every new topic needs a new intent.
Knowledge-trained AI chatbots
A knowledge-trained AI chatbot answers from the company's own content with a large language model. Nobody writes the individual replies. The team keeps the source content accurate, and the chatbot writes each answer from it.
This type can still give a wrong answer when the content is missing or contradictory. The guide to chatbot hallucinations covers the causes and the guardrails.
AI agents
An AI agent is a chatbot that can also act. It can look up an order, book a slot, or update a record in a connected system. The comparison of an AI agent vs a chatbot explains where the line sits.
Many products mix the types. A common setup uses knowledge-trained AI for general questions and a rule-based flow for steps that need exact control.
What can a customer support chatbot handle?
A customer service chatbot can handle questions that have a written, repeatable answer. The rule holds for every type: if a new agent could answer from the help center on day one, a chatbot can answer it too.
Task | Can a chatbot handle it? | Condition |
Opening hours, shipping times, return policy | Yes | The policy is written down and current |
Product details and plan differences | Yes | Product and pricing pages are in its content |
How-to and troubleshooting steps | Yes | The steps are in the help center |
Collecting a name and email for follow-up | Yes | Lead or contact capture is turned on |
Routine questions at night and on weekends | Yes | None. A chatbot has no shifts |
Order status and account lookups | Only with a connection to the order or account system | Needs an AI agent or an integration |
Questions in several languages | Depends on the product | Check the vendor's supported language list |
Real deployments follow the same pattern. The roundup of companies using chatbots for customer service shows which tasks named brands gave to a chatbot.
What can a customer support chatbot not handle?
A customer service chatbot cannot handle questions that need judgment, authority, or information nobody wrote down. Five kinds of conversation belong with a person.
- Refund disputes and billing exceptions. The chatbot can state the policy. It should not decide who gets an exception.
- Complaints from upset customers. A frustrated customer wants to be heard by someone who can act.
- Questions with no documented answer. A chatbot that guesses does more harm than one that says it does not know.
- Legal, medical, or financial advice. A wrong answer here can hurt the customer and create liability for the business.
- Account changes that need an identity check. These need a verified, auditable process.
A company answers for what its chatbot says
A business is responsible for its chatbot's answers. In February 2024, a British Columbia tribunal held Air Canada liable for negligent misrepresentation after its website chatbot gave a customer wrong information about bereavement fares.
Air Canada argued that the chatbot was a separate entity. The tribunal rejected that and wrote that it "makes no difference whether the information comes from a static page or a chatbot".
Customers give a chatbot few chances
Customers stop using a chatbot after one bad answer. Gartner surveyed 3,566 B2B and B2C customers in February and March 2026. Only 27% said they would try a chatbot again after a negative experience.
The same survey found that 49% of customers would use a company chatbot if one were offered. Only 7% used one in their most recent service interaction. That is a gap of 42 percentage points between intent and use.
Gartner's advice to service leaders was to start with issue types that have a proven high resolution rate and widen the scope only after performance is reliable.
When should a customer support chatbot hand off to a human?
A customer service chatbot should hand off to a human in four situations: the customer asks for a person, the answer is not in the chatbot's content, a wrong answer would be costly, or the chatbot has failed twice on the same question.
Signal | What the chatbot should do | Why |
The customer asks for a person | Hand off at once | 87% of customers in Gartner's 2026 survey called access to a human agent essential |
The answer is not in its content | Say so and offer a person | A guess can be wrong, and the business answers for it |
The topic involves money, safety, or a legal commitment | Route to a person or a fixed scripted flow | The cost of a wrong answer is high |
Two failed answers, or a frustrated customer | Hand off and pass the transcript | A third attempt rarely helps and the customer should not have to repeat the problem |
A handoff works only if the context travels with it. The agent should see the full conversation, so the customer does not have to explain the problem again. Gartner advises that a chatbot should give a visible path to a human agent and pass along the information it has already collected.
The chatbot best practices guide covers handoff rules, bot disclosure, and testing in more depth. For tools compared on this one feature, see AI chatbots with human handoff.
Worked example: one conversation, three outcomes
This example uses a hypothetical online store with a knowledge-trained AI chatbot. It shows the same chatbot answering, declining, and handing off in one conversation.
Turn | Customer message | What the chatbot does | Outcome |
1 | "How long do I have to return shoes?" | Finds the returns page and states the return window with a link | Answered |
2 | "Do you ship to Norway?" | Finds no shipping page that mentions Norway and says it does not know | Declined, person offered |
3 | "My pair arrived damaged and I want my money back now." | Recognizes a refund dispute and a frustrated customer | Handed off with the transcript |
All three outcomes are correct behavior. The chatbot resolved one question, avoided one guess, and sent one dispute to a person who can approve a refund.
What are the benefits of a customer support chatbot?
The benefits of a customer service chatbot are instant replies, cover outside office hours, consistent answers, and less routine work for agents. Each benefit has a condition.
Benefit | What it means | Condition |
Instant replies | The customer does not wait in a queue | The answer has to be right, or the wait moves to a second contact |
Cover at any hour | Routine questions get an answer at night and on weekends | Urgent cases still need a route to a person |
Consistent answers | Every customer gets the same policy wording | The source content must be current |
Less routine work | Agents spend their time on disputes and complex cases | The team must review transcripts and fix content gaps |
Which type of customer support chatbot do you need?
The type of customer service chatbot you need depends on where your answers live and whether the chatbot has to act. Use this table to match the type to the need. Once you know the type, the ranking of AI chatbot platforms for customer service compares seven vendors on price, handoff and G2 rating.
If you need | Pick | Why |
A fixed flow that never deviates, such as an eligibility check | Rule-based | Every step is scripted and auditable |
Answers to a short, stable list of questions in free text | Intent-based | Each intent has an approved reply |
Answers to hundreds of questions your help center already covers | Knowledge-trained AI | No replies to write. The chatbot answers from existing content |
A chatbot that looks up orders or changes account data | AI agent | Only this type connects to other systems and acts |
General answers plus one strict step | Knowledge-trained AI with a rule-based flow | Open questions get AI answers and the strict step stays scripted |
A chatbot for employees and IT requests | An internal help desk chatbot | The IT support chatbot guide covers that use case |
Where a knowledge-trained chatbot such as SiteGPT fits
SiteGPT is a knowledge-trained AI chatbot, the third type in the table above. It learns from website URLs, sitemaps, files, and connected sources such as Google Drive, Notion, GitHub, and YouTube. It answers in 95+ languages.
Escalation to a person is available on every plan. By default the chatbot escalates when the visitor asks for a person, when the situation is urgent, or when the visitor is frustrated after an answer that did not help. A team can replace those defaults with its own rules.
SiteGPT has limits. It has no visual flow builder for rule-based scripts. Chat integrations such as Slack, Crisp, and Messenger start on the Growth plan, and Zendesk needs Scale. There is no Intercom or HubSpot integration today, and there is no free plan.
Plan | Billed monthly | Billed yearly | GPT-4.1-mini messages a month | Chatbots |
$59 a month | $39 a month ($468 a year) | 4,000 | 1 | |
$129 a month | $79 a month ($948 a year) | 10,000 | 2 | |
$429 a month | $259 a month ($3,108 a year) | 40,000 | 3 |
Each plan gives one tenth as many GPT-4.1 messages as GPT-4.1-mini messages. Starter on yearly billing costs $468 a year against $708 on monthly billing, a saving of $240 or 34%. Every self-serve plan has a 7-day free trial that requires a credit card.
Common customer support chatbot mistakes
The most common customer service chatbot mistakes are about scope and handoff. Four show up most often.
- Launching with every topic at once. Start with the questions the chatbot resolves reliably. Only 27% of customers in Gartner's survey would retry a chatbot after a bad experience.
- Hiding the route to a person. A chatbot that blocks the way to an agent loses the customer's trust in the whole channel.
- Letting the chatbot guess. Set the chatbot to say it does not know when its content has no answer.
- Not saying it is a bot. Article 50 of the EU AI Act has applied since 2 August 2026. It requires that people are informed they are interacting with an AI system, unless that is obvious.
Frequently asked questions
Basics
What is a customer support chatbot?
A customer support chatbot is software that answers customer questions in a chat window. It uses scripted rules, an AI model trained on the company's content, or both, and it passes the conversation to a person when it cannot help.
What is the difference between a customer support chatbot and a customer service chatbot?
There is no difference. Both names describe a chatbot that answers customer questions, and vendors use the two words interchangeably.
Is a customer support chatbot the same as live chat?
No. A chatbot is software that writes the reply. Live chat is a person typing the reply. Many chat windows use both, with the chatbot answering first.
How it works
How does a customer support chatbot know the answers?
A customer support chatbot knows the answers because someone wrote them down. A rule-based chatbot uses scripted replies. An AI chatbot searches the company's help center, website, and documents, then writes a reply from what it finds.
Does a customer support chatbot use AI?
Some customer support chatbots use AI and some do not. Rule-based chatbots follow buttons and scripts with no AI. Intent-based, knowledge-trained, and agent chatbots use machine learning or large language models. To see what each machine learning job does inside the chatbot, read how machine learning is used in a chatbot.
Can a customer support chatbot give a wrong answer?
Yes. An AI chatbot can give a wrong answer when its content is missing, outdated, or contradictory. Grounding answers in the company's own content and letting the chatbot decline reduce the risk.
Limits and handoff
Can a chatbot replace customer service agents?
No. A chatbot handles routine questions with documented answers. Disputes, complaints, and exceptions still need a person. In Gartner's 2026 survey, 87% of customers said access to a human agent is essential.
When should a chatbot transfer a customer to a person?
A chatbot should transfer a customer when the customer asks for a person, when the answer is not in its content, when a wrong answer would cost money or create a legal commitment, or when it has failed twice on the same question.
Do you have to tell customers they are talking to a chatbot?
In the EU, yes. Article 50 of the EU AI Act has applied since 2 August 2026 and requires that people are informed they are interacting with an AI system, unless that is obvious.
Is a company responsible for what its chatbot says?
A company can be. In Moffatt v. Air Canada (2024), a British Columbia tribunal held the airline liable for negligent misrepresentation over wrong information from its website chatbot.
Cost and setup
How much does a customer support chatbot cost?
The cost of a customer support chatbot depends on the type and the pricing unit. Vendors charge per seat, per message, per conversation, or per resolution. The AI chatbot pricing comparison lists current prices side by side.
What do you need before launching a customer support chatbot?
You need four things before launch: current written answers to your common questions, a list of topics the chatbot should cover, a rule for when it hands off, and a person or inbox to receive the handoffs.
Which type of chatbot is best for a small business?
A knowledge-trained AI chatbot suits most small businesses that already have a website or help center, because nobody has to script the replies. A rule-based chatbot suits a business with a few fixed questions and no written content. When you are ready to buy one for your site, the guide to choosing a customer service AI chatbot for a website compares the four solution types and their first-year costs.
Sources
All sources were opened on October 11, 2026.
Written by
