Automation · 5 min read
AI chatbots on business websites: what they answer well, what they get wrong
Most chatbot pitches oversell the part where it sounds human and undersell the actual job: answering the same fifteen questions correctly, every time, without making anything up.
Every second sales call we take now includes the phrase "we want a chatbot like ChatGPT on our site." What people actually want, once you ask what problem it should solve, is usually much narrower: stop losing enquiries at 9pm, stop answering the same five questions by email, stop losing the visitor who almost booked but had one question left. A chatbot can do all three. It cannot do what the pitch decks imply, which is replace a person who actually understands the business.
What they are genuinely good at
Strip away the marketing and a website chatbot is doing one job well when it works: matching a visitor's question to information you already have, instantly, at any hour. That is a narrower job than it sounds, and it is worth more than it sounds.
- Repeat questions. Pricing ranges, turnaround times, what's included, do you serve my area. The same fifteen questions account for most of the traffic on a support inbox, and a scoped bot answers them correctly every time.
- Out-of-hours capture. A visitor at 11pm who has a question is a visitor who leaves if nobody answers. A bot that qualifies them and books a callback converts a dead browser tab into a lead.
- Pre-qualifying before a human joins. Asking budget, timeline and scope before a call saves the person on your team from spending fifteen minutes finding out it was never a fit.
~60%
of routine customer questions can be resolved without a human when a chatbot is scoped to a defined knowledge base
Source: Internal review across client support-ticket logs, Lipsum Technologies, 2026
Where they get it wrong
The failure mode nobody puts in the demo is hallucination, and it is the one that actually damages a business. Give a general-purpose model open access to answer anything, and it will answer anything, including questions it has no data for. It will quote a price you do not charge, promise a turnaround you cannot hit, or confirm you serve a postcode you do not cover. It says all of this in the same confident tone it uses for the answers that are correct, so a visitor has no way to tell the difference.
The second failure is scope creep in the conversation itself. A visitor asking about a refund policy is a different situation from a visitor asking a general question, and a bot that cannot tell the difference will cheerfully improvise a refund policy that your terms and conditions do not support. Once that happens in writing, on your website, it is a business problem, not a chatbot glitch.
The bot does not need to sound human. It needs to be right, and it needs to know when to hand off. Everything else is decoration.
Scripted versus AI-generated: the actual trade-off
| Scripted / decision-tree bot | AI-generated, scoped to your content | |
|---|---|---|
| Handles unexpected phrasing | Poorly, needs exact matches | Well, understands intent |
| Risk of wrong answers | Low, answers are pre-written | Low if scoped correctly, high if not |
| Setup effort | Low, map out common paths | Medium, needs your content organised first |
| Feels natural to use | Often feels robotic | Feels closer to a conversation |
| Best for | A handful of fixed, high-volume questions | A wider spread of real questions with defined boundaries |
Most businesses we build for end up wanting the second column, but the mistake is assuming "AI" means unrestricted. The versions that actually help are trained only on your own FAQ, pricing pages, service pages and policies, and instructed to say "I don't have that information, let me get someone to help" rather than guess. That single instruction is the difference between a useful tool and a liability.
How to scope one properly
- 1Write down the fifteen to thirty questions your team actually gets asked, pulled from real emails and calls, not guesses.
- 2Turn that into a clean knowledge base: current pricing, service area, turnaround times, policies. If it isn't accurate on your website, do not let the bot read it.
- 3Set a hard boundary: the bot answers from that content only, and hands off to a human or a booking form for anything outside it.
- 4Log every conversation for the first month and read it. This is where you find the questions you didn't know people were asking.
- 5Review monthly. Pricing and policies change; a bot trained on stale content starts quietly lying the day your prices do.
What it costs, and when it's worth building
A scoped chatbot is not a plugin you install and forget. It is closer to a small custom web application: a knowledge base that needs maintaining, a handoff flow into your booking or contact system, and monitoring so it doesn't drift. That's worth it once you have enough repeat-question volume to justify it, typically once a team member is spending real hours a week on the same handful of email replies. Below that volume, a well-written FAQ page and a fast contact form do the same job for a fraction of the cost.
If you're getting the volume but not the time to build this properly in-house, that's the conversation to have with us before you buy an off-the-shelf widget that answers from the open internet instead of your own business.
Questions people also ask
Will an AI chatbot replace my customer support team?
No, and treating it that way is how the wrong answers start reaching customers. It removes the repetitive first-line questions so your team spends time on the enquiries that actually need a person.
Can a chatbot make up information about my business?
Yes, if it has open access to general knowledge instead of being scoped to your own content. Scoped correctly, with an instruction to hand off rather than guess, this risk drops sharply.
How long does it take to set one up properly?
Building the knowledge base from your real support questions takes longer than the technical build itself. Most scoped chatbots we deliver take two to four weeks, most of it spent getting the content right.
