I still hang up on automated phone systems that can’t understand a straight sentence. Ask most people and they’ll admit the same thing: bad voice recognition, and a bot that won’t give you a straight answer, are two of the fastest ways to lose someone’s patience. Chatbots on websites do the same thing, and most of the website chatbot mistakes I see come down to one habit: building the bot to serve the business first and the visitor second.
I saw this clearly last year reviewing a chatbot for a client, Coastal Solar. The pop-up numbers told a story I’ve come across more than once with small business sites, and it wasn’t really about the bot at all. It was about what the bot had been built to do.
A Chatbot Built to Filter Leads, Not Help Visitors
The chatbot had been set up by a third party who, from what I could tell, promised a lot and tested very little. The script pushed for contact details early, before it had answered anything the visitor actually asked. It didn’t feel like a tool there to help. It felt like a filter, sorting people into “worth talking to” or “not,” before they’d had a chance to find out anything about the business.
That’s a subtle distinction, but visitors feel it. A chatbot that’s there to serve you reads completely differently to one that’s there to qualify you, even when the wording looks similar on the surface.
The Numbers Didn’t Lie
The data backed up the feeling. Over the period we looked at, the chatbot pop-up had appeared 567 times. Only 9 of those turned into an actual chat, a 1.59% engagement rate. Of those 9 conversations, 4 produced a lead.
Four leads isn’t nothing, and it’s worth saying the bot wasn’t a total write-off. But a 1.59% engagement rate on something sitting on nearly every page is a sign the opening message wasn’t earning attention, let alone trust. Most visitors were looking straight past it, and the small number who did click in mostly didn’t convert once they were there.
Why We Removed It Instead of Trying to Fix It
We could have spent time rewriting the script, softening the opening line, testing new prompts. Instead we looked at what was actually stopping people getting where they needed to go. There were two separate paths on the site: a web-based enquiry form, and a second form that handed visitors off to ServiceM8, a third-party trade software platform most people had never heard of, mid-enquiry.
Neither of those problems was going to be solved by a better chatbot script. The friction wasn’t in the bot, it was in what happened after someone decided they wanted to get in touch.
Two Clear Buttons Beat One Clever Bot
So we removed the chatbot and rebuilt the path instead. The homepage now has two plain buttons: one for solar enquiries, one for general electrical work. Each leads straight to a proper form, on the site, with no chat window to work through and no unexplained redirect to a third-party booking tool.
It’s a far less “clever” solution than a chatbot. It’s also a lot easier for a visitor to use without thinking about it.
More Leads, Better Data, Less Guesswork
Since the change, leads have increased, and the quality of the data coming through has improved too. Splitting the enquiry into solar versus general electrical at the button stage means Coastal Solar now knows what someone wants before the form is even filled in, instead of working it out from a chat transcript afterwards.
None of that came from the site looking sleeker. It came from a visitor understanding what to click and where it would take them, in under a second, rather than being asked to type a question into a box first.
The Website Chatbot Mistakes Worth Learning From
Strip away the specifics of this one client and the pattern holds up more broadly. Most of the website chatbot mistakes I see aren’t really about chatbots. They’re about tools that got installed because a vendor sold the idea convincingly, not because anyone checked whether it actually worked for the people using the site. Swap “chatbot” for a pop-up quiz or a booking widget and the same mistake shows up just as often.
If you’re going to run a chatbot, or any interactive tool like it, test it properly before you trust it. Put it in front of a small audience first, watch what actually happens, and refine it before it’s live for everyone. Don’t assume it works because someone told you it would.
The other thing I feel strongly about, and this is a personal opinion as much as a professional one, is disclosure. If a visitor is talking to AI, they should be told plainly that they’re talking to AI. Given how much of this work is about building trust with a visitor, hiding what they’re actually talking to undermines the exact thing you’re trying to create.
It’s Not About How Many Leads You Capture
The test I’d put to any business owner considering a chatbot, or auditing one that’s already live, is simple: is this here to help the person using it, or just to get something out of them? It’s not all about getting leads for the business. It’s about how the tool actually helps the person on the other end of it. Get that the right way round and the lead numbers tend to follow.
If you’re not sure whether something on your own site, a chatbot, a quiz, a booking widget, is helping visitors or just filtering them, that’s exactly the kind of thing a proper Website Conversion Review turns up. Most website chatbot mistakes are easy to spot once someone’s actually looking for them, and cheaper to fix than most business owners expect. If you’d rather talk it through first, a Clarity Call is a low-pressure way to start.
FAQs
Do chatbots actually hurt website conversion rates?
Not automatically, but a poorly built one can. If a chatbot is designed to extract contact details before it’s given the visitor anything useful, it reads as a filter rather than a help, and most visitors will ignore it or bounce rather than engage.
Should small businesses use a website chatbot at all?
It depends on whether it’s solving a real visitor problem. A chatbot that answers genuine, common questions quickly can work well. One installed because it seemed like a modern thing to have usually doesn’t.
How do you know if a chatbot is costing you leads rather than generating them?
Look at the pop-up-to-engagement rate, not just leads captured. A very low engagement rate, well under 5%, usually means the opening message or the premise of the bot isn’t earning attention in the first place.
Does a chatbot need to disclose that it’s AI?
There’s no way around this one in my view: yes. Visitors should always know when they’re talking to AI rather than a person. Hiding it undermines the trust you’re presumably trying to build with the interaction in the first place.
What should you test before launching a chatbot?
Run it past a small group first, ideally people who match your actual visitors, and watch where they hesitate or drop off. Refine the opening message and the flow based on what you see, rather than launching to everyone straight away.
What’s a simpler alternative to a chatbot for capturing enquiries?
Often it’s just a clear, well-labelled form with an obvious next step. If your business has a couple of distinct enquiry types, separate buttons or forms for each can do more for lead quality than a chat window ever will.
Should a chatbot or form hand visitors off to a third-party platform?
Be cautious with this. A redirect to a booking or trade platform a visitor doesn’t recognise, mid-enquiry, introduces a moment of doubt at exactly the point you want them to feel confident. Keeping the enquiry on your own site removes that friction.
What’s the biggest website chatbot mistake overall?
Installing something to solve the business’s problem rather than the visitor’s. A chatbot, or any tool like it, only earns its place on a site if it makes things easier for the person using it, not just easier to capture their details.