The most useful thing I can tell you about an AI chatbot for website visitors is that the good ones are boring. They do not have a personality. They do not greet you with a wave. They answer “do you cover my area?” at 10pm on a Sunday, ask two questions to see whether you are a fit, and put a slot in the calendar. Then they stop talking.
I build these for clients on their WordPress sites and inside GoHighLevel, and this article covers what a chatbot has to do to earn its place on a page, how to train one on your own business, and the specific ways they go wrong. If you want the broader picture of what to automate first, start with AI automation for small business; this is the deep dive on one piece of it.
Why visitors leave without asking
Look at any service website’s analytics and you will see the same pattern: a good share of visitors reach the pricing or service page, spend a minute or two, and leave. Some were not a fit. Many had one question — “do you do X?”, “roughly how much?”, “how soon?” — and could not find the answer fast enough to bother with a form. A form feels like a commitment. A chat window feels like asking.
That is the whole case for a chatbot. It converts the visitors who had a question into conversations, and the conversations that fit into bookings. It does not create demand; it stops leaking it.
What a good chatbot does, in order
- Answers the ten questions. Every business has them. Pricing ranges, areas covered, availability, what is included, how long it takes, whether you handle their specific situation. The bot answers from your own material, in your words, and links to the page with more detail.
- Asks two or three qualifying questions. Not a survey. “Is this for a home or a business?” “Roughly when do you need it done?” “What’s the postcode?” — the questions your best salesperson asks in the first minute of a call.
- Offers the next step that fits. Good fit: “Want to pick a time for a quick call? Here are the next available slots.” Not a fit: a polite pointer elsewhere. Unsure: collect details for a human.
- Writes everything to the CRM. Name, contact details, the answers, the transcript, a source tag. A conversation nobody can find later did not happen.
- Hands off cleanly. When it cannot answer, it says so and gets a way to reach the person. It never invents pricing, never promises a date, never argues.
Training it on your business (the part most people skip)
The model is not the hard part; the material is. A chatbot trained on nothing but your website will be vague, because websites are vague. Give it:
- Your website pages — the service descriptions and FAQ, so it can cite them.
- A plain document of real questions and real answers. Go through your inbox and call notes and write down the thirty questions you actually get, with the answers you actually give, including the awkward ones (“we don’t do that”, “not before March”).
- Pricing ranges and what changes them. If you will not publish prices, give the bot the range and the rule (“from £X, depending on size — the call is where we confirm”).
- Boundaries. What it must not do: quote firm prices, promise availability, discuss competitors, give medical or legal advice.
- The booking link and the qualifying questions, with the rule for when to offer the booking.
Then the loop that makes the difference: read the first hundred conversations. Every wrong or waffly answer goes back into the document as a corrected one. After two or three rounds, the bot sounds like your best staff member on a good day. Without the loop, it sounds like a search engine.
Tone Write the answers the way you would say them on the phone. Short sentences, no exclamation marks, no “Great question!”. Visitors trust a bot that sounds like a competent person more than one that sounds like a mascot.
Where it lives: WordPress site, GoHighLevel, or both
- On a WordPress site. A lightweight chat widget that loads after the page (so it does not hurt speed), talks to an AI backend with your knowledge base, and posts every conversation and lead to your CRM through a webhook. This is the setup I build most often, because the website is where the visitors are. The widget must not add a second of load time — see the speed checklist on deferring third-party scripts.
- Inside GoHighLevel. The platform’s Conversation AI can run on its web chat widget, SMS, Facebook and Instagram DMs, with direct access to the calendar and CRM. Best when you already run the CRM there and want one bot across every channel.
- Both. Web chat on the site handled by the same AI that answers texts, so a visitor who starts on the website and continues by SMS gets one continuous conversation. This is the most useful arrangement and the most work to set up.
What it costs
Usage is cheap: AI messages are metered at fractions of a cent, so even a busy site spends a modest amount a month. The cost is the build — knowledge base, qualification logic, CRM connection, testing — and the review loop in the first month. Priced as a project, it is comparable to a good landing page. Priced as a monthly “managed AI” retainer, be sure you know what is being managed.
Measuring whether it works
Three numbers, compared before and after:
- Conversations started per 100 visitors. If this is tiny, the launcher is invisible or the greeting is wrong.
- Bookings or leads per 100 conversations. If this is low, the qualification or the offer to book comes too late, or too early.
- Handoffs per 100 conversations. How often the bot gave up. High at first is normal; it should fall as the knowledge base improves.
Tie the bookings back to revenue in the CRM by tagging the source source-webchat; the CRM setup guide covers the tag structure.
How chatbots go wrong
| Failure | What the visitor experiences | Fix |
|---|---|---|
| Pop-up on page load | Interrupted before reading anything; closes it, or the tab | Quiet launcher; open on click or after genuine dwell time |
| Invented answers | Quoted a price or a date you cannot honour | Hard boundaries in the instructions; “I’m not sure — a human will confirm” |
| Survey before service | Five questions before it will answer one | Answer first, qualify second |
| No handoff | Bot loops; visitor gives up | Collect contact details after two failed answers; notify a human |
| No CRM record | Leads vanish into transcripts nobody reads | Webhook every conversation to the CRM with a tag |
| Slow page | Widget script loads before the content | Defer the widget until after the page is interactive |
Is it right for your site?
Yes, if you get enquiries that are mostly the same questions, if you lose leads outside office hours, or if your booking calendar has gaps that a faster response would fill. Probably not yet, if you have no CRM to write to (fix that first), if your traffic is a handful of visits a day (a good FAQ section does the job), or if your service genuinely needs a human conversation from the first sentence — in which case a chatbot that just collects details and promises a fast call-back is the honest version.
Further reading: GoHighLevel Help Center.
FAQ
Will an AI chatbot annoy my website visitors?
A pop-up that interrupts within two seconds will. A quiet launcher in the corner that opens when clicked, answers real questions and offers a booking when it makes sense does not. Timing and restraint matter more than the AI model.
How do I train a chatbot on my business?
Give it your website pages, a document of the questions you are asked most with your actual answers, your pricing ranges and your booking link. Then read the first hundred conversations and correct what it got wrong. That review loop is the training.
Can the chatbot book appointments directly?
Yes, when it is connected to your calendar system. In GoHighLevel the Conversation AI can offer available slots and book them; on a WordPress site the bot can hand off to an embedded calendar or send the visitor a booking link.
What happens when the chatbot cannot answer?
It should say so and collect a name and phone or email so a human can reply — never invent an answer. A good setup routes those conversations to your inbox with the transcript attached.
If you would like a chatbot that answers like you and books into your calendar, book a free call. Bring your ten most common questions; that is most of the training material already.
