A chatbot answers on one page from a script. An agent answers on every channel from your records, and can act on what it learns.
Published November 12, 2025
The word chatbot usually means a widget in the corner of one page, running a decision tree somebody drew a year ago. It knows what is in the tree. Ask it anything outside that and it offers you a contact form, which is the same as telling you to come back during office hours.
An agent is a different shape, and the difference is worth being precise about, because the two get sold under the same word and they do not do the same job.
A prospect who finds your listing at nine at night might call, might text, might email, or might type into the box on your site. A chatbot covers exactly one of those. An agent covers all of them, and treats them as the same conversation rather than four unrelated events.
That matters more than it sounds, because people move between channels without thinking about it. Somebody starts a web chat, gets a text confirmation, then calls the next day with a follow-up question. If those three are handled by three disconnected systems, the person ends up explaining themselves three times, and your team gets three partial records instead of one useful one.
This is the part that separates something useful from something embarrassing. A scripted bot can only repeat what was typed into it. An agent reads from your actual system: which units are open, what they rent for, what your pet policy says, what is on the calendar this Saturday.
The consequence is that its answers stay correct when your business changes. You raise a rate, and the next caller hears the new one. A unit goes under application, and it stops being offered. Nobody has to remember to go and update the bot, which is the step that never actually happens.
Answering is half the job. The other half is what happens next, and this is where most automations quietly stop being useful.
An agent that answers beautifully but leaves a human to re-enter everything afterwards has moved the work rather than removed it.
This is the part buyers underrate and then care about most once something goes wrong. A general-purpose model asked about a neighbourhood will happily answer, and in real estate that answer can be a fair housing problem. A narrow, well-scoped agent refuses, and routes the question to a person.
The same applies to facts it cannot source. If a rate is not in your rate sheet and a balance is not in your system, the right behaviour is to say so and hand off, not to produce something plausible. A confident wrong answer costs more than a missed call, because the missed call does not create an expectation you then have to walk back.
The interesting part of an AI agent is not what it says. It is what it refuses to say.
Three questions separate them quickly. Ask what happens when someone asks a question the script does not cover. Ask where the answer about availability comes from, and what happens when that changes. Ask what it does when a caller asks for a human.
If the answers are a fallback message, a spreadsheet someone updates manually, and a contact form, you are looking at a chatbot with better marketing.
This is not an argument that widgets are worthless. A small operation with one property, a handful of questions that never change and somebody who answers the phone reliably may genuinely not need anything more.
The point at which the difference starts to matter is when your answers change often, when inquiries arrive on more than one channel, or when something has to happen after the conversation rather than during it. That is when a script stops keeping up with the business it is describing, and it is worth being honest with yourself about which situation you are in before buying either.
On the site
Keep reading