Every few months a client asks whether they should stick an AI chatbot on their website. It is a fair question, and the honest answer is: it depends entirely on your site, your traffic and how the bot is built. A good one can quietly capture leads while you sleep. A bad one irritates the few visitors you have and sends them elsewhere. Let me give you the buyer's guide I would want if I were making the decision.
When a chatbot earns its keep
Chatbots pay off in specific conditions. If your site matches most of these, one is probably worth trying.
- High enquiry volume. If you are fielding the same questions dozens of times a week, a bot that handles the routine ones frees up real time.
- Repetitive questions. Opening hours, pricing ranges, "do you cover my area", "how long does X take". Predictable questions are exactly what a bot handles well.
- Out-of-hours traffic. A good chunk of web visits happen in the evening and at weekends when nobody is at the desk. A bot can answer and, more importantly, capture the enquiry so it is waiting for you on Monday rather than gone.
- Enough traffic to matter. A bot only helps if people arrive to use it. On a busy site, small improvements in capture add up.
In short, a chatbot is a volume tool. It shines when there is a steady stream of visitors asking answerable questions, and it captures the ones who would otherwise leave without making contact.
When it annoys people
The flip side is just as real, and I see it ignored constantly.
On a low-traffic site, a chatbot solves a problem you do not have. If you get a handful of visitors a day, a pop-up bot adds friction and clutter for almost no capture upside. You would be better off with a clear phone number and a good contact form.
For complex, consultative sales, a bot can actively hurt. If your work involves nuance, a proper conversation and building trust, funnelling a serious prospect through a scripted widget cheapens the first impression. Some buyers want a person, and forcing a bot on them signals the opposite of the care they are paying for.
And a badly trained bot is worse than no bot at all. We have all hit the widget that loops the same three canned answers, cannot understand a plain question, and offers no escape to a human. That traps the user, and a trapped user does not become a customer. If you cannot do it well, do not do it.
Scripted widget versus real assistant
This is the distinction that matters most, and it is where a lot of the disappointment comes from. There are two very different things both sold as "chatbots."
A scripted widget follows a fixed decision tree. You click a button, it shows the next set of buttons. It is really a menu wearing a chat costume. It is cheap, predictable, and fine for the simplest routing, but it cannot handle anything you did not script, and users hit its walls quickly.
An LLM-powered assistant trained on your own content is a different animal. It understands questions phrased in the visitor's own words, and it answers from your actual material: your services, your pricing approach, your FAQs, your policies. Trained properly and kept on a tight leash, it feels like talking to someone who has read your whole website. The critical phrase there is "on your own content." A general AI bot that has not been grounded in your information will improvise, and an improvising bot on your site is a liability because it can confidently say things that are not true.
Lead capture that actually works
Whatever you build, the job is not to have a clever conversation. It is to capture a lead or point the person to the next step. The patterns that work are unglamorous.
- Capture the contact detail early and naturally, before the visitor loses interest, ideally in exchange for a genuinely useful answer.
- Always offer a clear route to a human. "I can get one of the team to call you back, what is the best number" often converts better than trying to close in the chat.
- Hand off cleanly. The bot should qualify and route, then get out of the way. It does not need to do everything.
- Log every conversation. The transcripts are a goldmine, showing you the real questions people ask, which should feed straight back into your website content.
So, worth it?
For a busy site drowning in repetitive questions, an LLM assistant trained on your content and built to capture leads can genuinely pay for itself. For a quiet site or a considered, high-trust sale, it is often a gimmick that gets in the way, and your money is better spent elsewhere.
If you are weighing one up, we build website assistants grounded in a client's own content as part of our custom software and web app development, and we are equally happy to tell you when you do not need one.