It's 9:40 on a Sunday night. A customer is on your website with one question: are you open tomorrow, and do you take walk-ins? The answer is on your Contact page, but they didn't see it. They're not going to email you and wait. They're going to try the next business on the list.
Now picture a small chat window in the corner of the page. They type the question and get an answer in seconds. They book a time. You find out about it Monday morning with your coffee.
That window is a chatbot. If you've been wondering what a chatbot is, how it works and whether your business needs one, this is the plain version, including the parts vendors tend to skip.
What is a chatbot?
A chatbot is software that has a conversation with a person, usually by text, sometimes by voice. The person asks or requests something. The bot replies without a human typing each message.
For a business, that conversation might happen:
- In a chat window on your website
- Inside a messaging app or social media inbox
- In your own customer app or client portal
- On your phone line, as a voice assistant
What it does depends on how it's built. A bot might answer questions about hours and services, collect a name and phone number for a callback, check where an order is, or schedule an appointment. The common thread is that it handles a request instantly, at any hour, and passes the ones it can't handle to a person.
It's worth separating chatbots from the things they get confused with. A contact form takes a message and waits. A search box finds pages. A chatbot goes back and forth with the visitor, which is why it feels more like asking a person.
How chatbots work
Under the hood, every chatbot does the same two jobs:
- Understand the request. What is this person asking for?
- Respond. Give an answer, ask a follow-up question, take an action or hand off to a human.
What differs is how the first job is done, and that splits chatbots into two broad kinds.
Rule-based chatbots
These follow a script you write. The customer taps a button ("Hours," "Pricing," "Book a visit") or types a keyword, and the bot sends back the answer you prepared. Think of a decision tree with a friendly face.
They're predictable. You know exactly what they'll say, because you wrote every line. That makes them a good fit for narrow, repeatable jobs like store hours, service areas or a quote request that needs the same five details every time. The limit is that they only know what you scripted. Ask something unexpected and you'll get "Sorry, I didn't understand that."
Bots built on language models
Most newer chatbots run on a large language model, the same kind of technology behind the text tools you've probably tried. Instead of matching keywords, they interpret what a person means, even when it's phrased in a way nobody planned for. "Can I bring my dog in this weekend?" and "Do you allow pets Saturday?" are the same question, and the bot treats them that way.
On their own, though, these models know nothing about your business. A well-built business chatbot is given your information (your services, policies, hours and FAQs) and told to answer from that and nothing else. It can also connect to your systems, so it can look up an order or check an open appointment slot.
This is the part that's worth being honest about. A language model writes fluent answers whether or not they're correct. If it isn't kept to your approved content, it can state something wrong with total confidence: a price you don't charge, a policy you don't have. Good builds limit what the bot can say, test it against real questions and give it a clear way out when it doesn't know. We cover how that works in practice in our guide on how to build a chatbot for business.
What chatbots are good at
Chatbots earn their keep on volume and repetition. The jobs that suit them best:
- Answering common questions. Hours, locations, service areas, how a process works, what to bring.
- Capturing leads outside office hours. A visitor who'd have left a form blank might answer three conversational questions and leave their number.
- Qualifying inquiries. Asking a few questions up front so your team spends time on people who fit.
- Routing. Sending a billing question to billing and a sales question to sales, instead of everything landing in one inbox.
- Simple tasks. Booking, rescheduling, checking status and sending a confirmation.
If you could write down the ten questions that make up most of your inbox, those are your chatbot's job description.
Where chatbots fall short
Here's the other side, because it matters when you're deciding.
They're frustrating when they can't help. Nearly everyone has been stuck in a chat loop with no way to reach a person. A bot that traps customers does more harm than having no bot, and the Nielsen Norman Group publishes usability research on chat interfaces if you want evidence beyond our opinion.
They don't handle nuance well. An upset customer, an unusual situation or a high-stakes decision needs a person.
They raise privacy questions. People will type personal details into a chat window without thinking. Decide up front what the bot should collect, where those conversations are stored and who can read them, and tell customers plainly. Never ask for card numbers or passwords in a chat.
They need upkeep. Prices change, services get added, policies shift. A bot full of last year's answers is worse than a clear FAQ page.
They're only as good as what you feed them. If your website is vague, your bot will be too. A chatbot can't cover for missing information. It tends to expose it.
What good looks like to a customer
The best chatbots feel almost boring. They say up front that they're a bot, so nobody feels tricked. They answer quickly, in plain words, and don't pretend to know things they don't. When they hit a limit, they say so and offer a person: a phone number, a callback or a message that goes straight to your team. If a customer finishes a chat feeling helped rather than handled, the bot is doing its job.
Is a chatbot right for your business?
A quick gut check:
- Do you get the same questions over and over?
- Do customers reach you outside business hours?
- Do you lose leads because replies take too long?
- Could most of your inquiries be answered from information you already have written down?
- Do you have someone who'll review conversations and keep it current?
Three or more yeses and a chatbot is worth a serious look. If most of your inquiries are one-of-a-kind, a quick personal reply will do more for your business than any automation.
It's also worth asking whether a chatbot is the right tool at all. Sometimes the real fix is a clearer website, a better contact page or a booking tool that does the job without a conversation. If your site gets traffic but few inquiries, start with why a website isn't generating leads before adding anything new.
Off the shelf or custom
You have two broad routes. Ready-made chatbot products are quick to start and fine for general questions and lead capture. A custom chatbot, built around your own systems, makes sense when the bot needs to do real work inside your business: pull a client's project status from your portal, check live availability in your scheduling system or write into your records without anyone copying details across.
That's the same buy or build question that comes up with all business software, and we walk through it in build vs buy software. If your bot needs to talk to the tools you already run, our integrations work connects them. If it needs to be a proper part of a custom system, that's application development.
Start with the question, not the bot
The best way to begin is to forget the technology and collect your customers' questions. Read the last month of emails, texts and voicemails. Count the repeats. That list tells you whether a chatbot would save you real hours or just add one more thing to maintain.
If it would, the next step is deciding how to build it, which is what our guide on how to build a chatbot for business covers. And if you'd like a second opinion on whether a bot fits, we're happy to talk it through, including when the answer is that you don't need one yet.