A Chatbot on Your Website: Worth It or Not?

Chat widgets went from novelty to default in about three years, and a great many were installed because everyone else had one. Here is a straight assessment of when one helps and when it quietly does damage.

When a chatbot genuinely earns its place

You get the same handful of questions constantly. Opening hours, delivery times, whether you cover a particular area, whether a service includes something. If eight questions cover most of your inbound messages, a bot answers them instantly at three in the morning and you stop typing the same reply.

Your customers decide outside office hours. Restaurants, clinics, tradespeople, anything people research in the evening. A visitor with a question at 22:00 either gets an answer or goes to a competitor who answers.

You have real volume. A chatbot is worth configuring and maintaining when it handles enough conversations to save meaningful time. Below a few enquiries a week, you are maintaining a system to replace something you could have answered yourself.

Qualification saves you time. If half your enquiries are outside what you do, a few questions up front — what, where, roughly what budget — filter them before they reach your calendar.

When it costs you

When it stands between people and a human. The most damaging pattern is a bot that cannot answer and will not hand over. A visitor who wanted to buy something ends up arguing with a script and leaves. Always offer a visible route to a person; the exit should be as easy to find as the greeting.

When it makes things up. An AI bot connected to a language model without proper grounding will answer confidently about prices you do not charge and services you do not offer. That is not an embarrassment, it is a commercial promise a customer may reasonably hold you to.

When your site has five pages. If everything a visitor needs is one scroll away, a widget popping up to ask if you need help is an interruption, not a service.

When nobody is behind it. A live chat that shows "typically replies in a few minutes" and does not is worse than no chat. It is a promise you are visibly breaking.

Bot, human, or both

Three models, and the choice matters more than the technology.

Rule-based bot. A decision tree with fixed answers. Boring, cheap, entirely predictable, and it cannot invent anything. For a business with eight recurring questions this is often the correct answer, and it is regularly dismissed too fast in favour of something cleverer.

AI bot grounded in your content. A language model constrained to answer only from your actual pages and documents. Much better conversationally, and the grounding is what stops it inventing. Requires your content to be accurate and organised, which is a useful forcing function in itself.

Live chat with a person. The best experience by a distance, when someone is genuinely available. Works well for high-value sales, badly as a background task nobody owns.

Many good setups are a hybrid: a bot handles the common questions and hands over to a person, with a form fallback outside working hours.

The practical requirements people forget

Three things that turn a good idea into a liability.

Privacy. A chat collects personal data — names, emails, whatever people type, and often more than they should. You need a lawful basis, a retention period, and a processing agreement with the vendor. If the widget also sets tracking cookies, it needs consent before it loads, which for most implementations means it should not fire until the visitor accepts.

Accessibility. Since June 2025 the European Accessibility Act applies to your chat widget like everything else on the page. It must be reachable and operable by keyboard, announced properly to a screen reader, and it must not trap focus. A surprising number of popular widgets fail this, and it is worth testing with the Tab key before you commit.

Performance. Chat widgets are among the heaviest third-party scripts on the web. Loading one on every page, immediately, can cost you a second of interactivity on a mid-range phone. Load it lazily, after the page is usable, and only where it is actually needed.

How to decide

Answer four questions honestly:

1. What are the five questions you answer most often? Write them down. If you cannot fill five, you do not need a bot.
2. How many enquiries a week? Under five, handle them yourself. Over twenty, automation starts paying.
3. Who maintains it? Answers go stale. Prices change. If nobody owns it, it will be wrong within a year.
4. What happens when it cannot help? If the answer is not "a human, quickly", fix that before installing anything.

What we would suggest first

Before a chatbot, try the cheaper thing: put the answers on the page.

Most chat conversations are questions whose answers should have been visible. A clear FAQ, prices where people expect them, delivery and coverage stated plainly, and a contact form that is easy to find will remove a large share of the messages a bot would have handled — and unlike a bot, that content also gets you found in search and cited by AI assistants.

If you have done that and the questions keep coming, then a chatbot is solving a real problem rather than covering for a page that does not say enough.