Search for a way to answer customer questions on your website and you will be sold four things with four names. AI chatbot. AI agent. Virtual assistant. Live chat. Half the vendors use two of these words for the same product on the same page.
The words do mean different things, and the difference matters when you are choosing. Here is the plain version.
Live chat
A person on your side, typing.
There is a widget on your site, a visitor sends a message, and it arrives in an inbox where a member of your team picks it up. Everything else in this article is software answering. Live chat is a human answering, with software delivering the message.
The tools people mean when they say live chat — Intercom, Tidio, Crisp, Zendesk, Freshchat — are really support platforms. The chat window is the front door; behind it is the inbox, the ticket queue, the routing, the reporting on how quickly your team responds.
Who it is for: businesses with somebody whose job is answering. If nobody is rostered to watch the inbox, live chat is a widget that collects unanswered messages.
Rule-based chatbot
The older kind, and still very common.
Somebody draws a flowchart. If the visitor clicks Delivery, show the delivery message. If they type a word from this list, show that message. If nothing matches, offer a menu or a contact form.
It does exactly what it was told, which means it never invents anything and never surprises you. It also cannot answer a question nobody anticipated, and every new answer is a new branch somebody has to build.
Who it is for: simple, stable, high-volume flows. Order lookups, opening hours, routing to the right department. Also anyone who genuinely cannot risk an unscripted answer — some regulated situations really do need a script.
AI chatbot
Software that reads the question and writes an answer, working from a body of knowledge you gave it.
No flowchart. You point it at your website, upload your documents, add some questions and answers, and it responds to whatever comes in by finding the relevant part and phrasing a reply. It handles a question worded three different ways without three branches.
The important constraint, and the thing to check when you are comparing tools: a properly built AI chatbot for a website answers from your content, not from what the underlying model knows about your industry. If it is answering from general knowledge, it will eventually tell a customer something about your business that is not true.
Who it is for: most small businesses. The questions are varied but the answers are all somewhere on your site already.
AI agent
An AI chatbot that also does things.
This is where the marketing gets loose, so here is a usable test: does it only produce text, or does it take actions? If the software can decide to collect a contact, put a button in the conversation, hand over to a person, or call another system — and decides when on its own, based on what the conversation is about — that is agent behaviour. If it only ever replies with words, it is a chatbot however it is labelled.
The second test is less obvious and matters more if nobody is supervising: does it know when it does not know?
Ask any chatbot something your website has never mentioned. A weak one produces a fluent, confident, wrong answer, and you find out weeks later from a customer who acted on it. A stronger one says it does not have that information. An agent does both — says so to the visitor, and records the question for you, so the thing that is wrong today is fixed tomorrow rather than compounding.
That is self-assessment, and it is the trait that makes an unsupervised chatbot survivable rather than a slow-motion liability. It is also the easiest thing to test in five minutes with any trial account, and the least likely to appear on a feature list.
Who it is for: anyone who needs the conversation to end somewhere — a booking, a contact, a purchase — and anyone who will not be watching it.
Virtual assistant
Mostly a marketing word. Sometimes it means a chatbot, sometimes an agent, occasionally an actual person in another country. Read the feature list rather than the label.
The bit that trips people up
The two words are not describing the same thing, which is why so many products use both.
Chatbot describes the form: a chat window on a website that a visitor types into. Agent describes the behaviour: software that works out what is being asked and decides what to do about it, rather than following a script somebody drew.
On capability there is a hierarchy. An agent does what a chatbot does and takes actions on top. On form there is not, because an agent still reaches your customer as a chat window — there is no separate agent-shaped thing sitting on a website.
Which is why a product calling itself an AI agent can show you a chat widget without being dishonest, and a product calling itself an AI chatbot can collect leads and hand over to a person under the more familiar name. Neither word tells you where the boundary is.
Judge them by what they do.
A short buying checklist
Whatever the label, these are the questions that separate the products:
- Where do the answers come from?Your content only, or the model's general knowledge as well? Only the first is safe to leave unattended.
- What happens when it does not know? Guess, deflect, or tell you? Being told is what makes an unattended chatbot survivable.
- Does anyone have to watch it? If the tool assumes somebody picks up conversations, it is a support platform and you need staff for it.
- What does it do besides answer? Collect a contact, show a button, connect the visitor to a person — or nothing?
- How does the price behave when it works well? Per resolution means the bill grows with success. Flat means it does not.
- How long from signing up to live? Some tools train themselves from your website in a minute; others expect you to build a knowledge base first.
Which one do you need
If somebody on your side is going to answer chats, you want live chat and a support platform. Tidio and Intercom are the usual choices and they are good at it.
If the answers are already on your website and nobody is going to sit in an inbox, what you want is the combination: a chatbot in form and an agent in behaviour. A chat window your visitors recognise, with something behind it that judges intent, acts when the moment calls for it, and admits what it does not know.
That is the shape most small businesses need, and it is the one the terminology obscures — because it gets sold under both names and neither name says which you are getting. The checklist above is how you find out.
If you need tightly scripted, unvarying responses in a regulated context, a rule-based bot is still the right answer.
Disclosure, and our own answers to the checklist. We make AI Chatbot — a chat widget on your site, with an agent behind it. Since we just told you which six questions to ask, here is how it answers them:
- Where do the answers come from? Your content only— your pages, your files, your Q&A pairs. Nothing from general knowledge about your industry.
- What happens when it does not know? It says so, and the question lands on your gaps list with the conversation attached.
- Does anyone have to watch it? No, and there is nowhere to watch it from. There is no shared inbox, because it is built for businesses that do not have anyone to staff one.
- What does it do besides answer? Collects contacts, puts buttons in the conversation, and offers your contact card when someone wants a person.
- How does the price behave when it works well? Flat monthly. No per-seat charge, no per-resolution charge.
- How long from signing up to live? Paste your address and it drafts itself in about a minute.
Where it is the wrong choice: if you want a shared inbox with live handover, or branching multi-step workflows. We say so on the comparison pages too.
See what it writes about your business — free, no card.