Leads12 min read

AI Chatbot for Website Visitors: How to Capture, Qualify and Book Them

By Anton Schwartz, co-founder of Konverto

In short

  • A website chatbot has four jobs: answer the first question, qualify with three to five questions, book a call or visit into a real calendar slot, and hand the conversation to a person with a summary.
  • Every qualifying question should change whether you can help, who handles the job or which slot you book. Ask for contact details early, with a reason, so the conversation survives a closed tab.
  • In the EU, Article 50 of the AI Act requires people to be told they are dealing with an AI at the latest at the first interaction, from 2 August 2026. Put it in the greeting.
  • The AI model behind a chat costs around a cent per conversation on published API rates, so compare tools on what one booked call or visit costs you, not on the sticker price.

A chatbot for website visitors earns its place when it does four things well. It answers the visitor's first question at once. It asks the few questions that tell you whether and how you can help. It books a call or visit into a real slot in your calendar. And it hands the conversation to a person, with a clean summary, when a person is the better choice. Everything else (avatars, animations, a clever bot name) is decoration.

Below: what to ask, how booking works and how chats reach your team, then integration, design, price and a two-week test.

What should a chatbot for website visitors actually do?

Visitors who open the chat usually want a quick answer, want to book, or are existing customers with a question. A good website chatbot works out which within a message or two.

For a new inquiry, the job looks like this:

  1. Greet and disclose. Say it is an AI assistant and that a person is available.
  2. Answer the first question from information you control: services, service area, hours, how pricing works.
  3. Qualify with three to five questions.
  4. Capture contact details, so the conversation survives if the visitor closes the tab.
  5. Book a call, consultation or visit.
  6. Hand over to a person when the case needs one.

What it should not do: invent prices, promise dates it cannot see, give medical, legal or financial advice, or keep someone in a loop after they ask for a human.

A rule-based bot follows the buttons you scripted. An AI chatbot understands free text like "the heating keeps cutting out and we have a newborn" and still follows your rules about what to ask and when to hand over.

The disclosure is not just politeness. In the EU, Article 50 of the AI Act applies from 2 August 2026. It requires AI systems intended to interact directly with people to be designed so those people are informed they are dealing with an AI, unless that is obvious, in a clear and distinguishable manner at the latest at the first interaction. The duty sits with the provider, but it shows up in your greeting, so check it is there. Elsewhere it is simply good practice.

What should a website chatbot ask? Questions that qualify without scaring people off

One rule keeps the questions short: every question must change whether you can help, who should handle the job, or which slot to book. If the answer changes none of those, collect it later.

A good order:

  1. The need, in the visitor's own words. One open question.
  2. Fit: which service, and where (ZIP code or area).
  3. Timing: how urgent, or when they want to start.
  4. Contact details, with a reason.
  5. One preparation detail, such as a photo, a budget range or an insurer.

Ask one question per message. Use buttons for fixed choices (new or existing customer, morning or afternoon) and free text for the need itself.

Examples by business type:

BusinessQuestions worth asking
Home services (plumber, electrician, roofer)What needs doing? ZIP code? How urgent? Can you send a photo?
Dental or physio clinicNew or existing patient? What is the concern? Which location and time of day?
Salon or med spaWhich treatment? First visit? Preferred day or staff member?
Real estate agencyBuying, selling or renting? Which area? What is your timeline?
Driving or language schoolWhich course? Current level? When do you want to start?
Car dealershipWhich car or stock number? Trade-in? Cash or finance?
Insurance brokerWhich cover? Current provider? Renewal date?

Two templates you can adapt. The opening message:

"Hi, I'm the AI assistant for [business name]. I can answer questions about our services and book a call or visit with the team. You can ask for a person at any time. What can I help you with?"

And the contact request, asked right after the visitor has described their need:

"Thanks, that helps. What's the best mobile number or email for you? If we get cut off, the team can pick up from here."

Asking second, not first, keeps the chat from feeling like a form. Asking before booking means a half-finished chat is still a lead you can reach.

How an AI chatbot books a call or visit

First decide what the chatbot may book:

  • A callback or phone consultation: short, and almost anyone on the team can take it.
  • A site visit or assessment: needs an address, travel time and the right person.
  • An appointment at your premises: length depends on the service, often with a specific staff member.

Then the flow:

  1. Check real availability in the calendar your team uses, across everyone who could take the job.
  2. Offer two concrete options instead of "When suits you?"
  3. Read the details back: date, time, address for visits, anything the visitor should have ready.
  4. Write the booking into the calendar with the chat summary attached.
  5. Send a confirmation by email or SMS with what happens next.

You write the rules: appointment lengths, buffers and travel time, shortest notice, how far ahead people can book, and which jobs need a person before anything is booked.

An illustrative example, not a real customer: a heating company gets a chat at 9:40 pm about a boiler that keeps cutting out. The chatbot asks for the ZIP code and the safety question the company wrote for this case (with a fixed emergency instruction if the answer is yes). It is not urgent, so it takes a mobile number and offers "tomorrow between 8 and 10, or between 1 and 3". The booking lands in the technician's calendar with the summary, and a confirmation goes out by SMS.

When no slot fits, the chatbot offers a callback request, says when the team will be in touch and flags the chat for a person.

How chat conversations reach your team

A chat that ends in a personal inbox, or in a chat app nobody opens, is a lost inquiry. Route every conversation into one shared inbox, and make each one arrive with:

  • name and contact details
  • the need in one line
  • the answers to your qualifying questions
  • booking status (booked, callback requested, needs a person)
  • the full transcript and the page the visitor was on

The handover rules decide whether the chatbot helps or hurts. Write them down before you switch it on:

Situation in the chatThe chatbotA person
Visitor asks for a personSays who will reply and when, takes contact detailsReplies in the chat or calls back
Complaint or unhappy customerAcknowledges and takes details, never arguesFollows up the same day
Existing customer with an account questionAnswers general questions, routes the restHandles anything account specific
Large or complicated job, or ready for a big purchaseCaptures the details, offers a callTakes over
Urgent safety issueGives your fixed emergency instructionAlerted at once
Chatbot still unsure after two triesStops guessingPicks it up

Urgent handovers should alert the person on duty at once; the rest can wait in the inbox. Only promise a response time your team can keep: "Someone from the team will reply by 9 tomorrow morning" beats "an agent will be with you shortly" followed by silence.

If a visitor leaves halfway but gave a phone number or email, reply there: a short message soon after, perhaps one more the next day, then stop.

Using an AI chatbot for website customer service as well as lead capture works if you keep the line clear. Repetitive questions (hours, parking, what to bring, which areas you cover) can be answered from a knowledge base you control. Anything that depends on a specific customer's job, booking or invoice goes to a person. The same split applies on the phone, which is covered in what an AI receptionist does.

Chatbot for website integration: WordPress, calendar and CRM

Your site and WordPress. Most tools give you a short script to paste before the closing body tag, or a plugin. On a self-hosted WordPress site, install the vendor's plugin or paste the script with a header and footer code plugin. On WordPress.com, installing plugins is available on all paid plans (Personal, Premium, Business and Commerce), and a free site has to upgrade before it can add one.

Speed. Chat widgets load a lot of third-party code. Google's Lighthouse documentation lists live chat widgets among embeds that can use a facade: a lightweight placeholder that preconnects on hover and is swapped for the real widget on click. The trade-off: features such as an unread-message badge only appear once the real widget loads. Ask your vendor whether lazy loading is supported.

Calendar. The chatbot should read busy times from every relevant calendar and write bookings back. Try to double-book a slot during your test.

CRM. If you use a CRM, check that each chat creates or updates a contact, matches on phone number or email instead of creating duplicates, and attaches the transcript.

The best chatbot for website integration also puts chats next to your other inquiries (web forms, lead ads, social messages), so your team works from one list, not four.

Chatbot for website design: placement, mobile and accessibility

  • Placement. A bottom corner, clear of your main call or booking button and your cookie banner. On mobile, check that the chat launcher does not cover a sticky "Call now" bar.
  • No full-screen ambush. A chat window that opens over the page on a phone frustrates people. A small greeting bubble after a short delay is enough.
  • Keyboard and screen readers. WCAG 2.2 Success Criterion 2.4.11 Focus Not Obscured (Minimum), level AA, says that when a component receives keyboard focus, it must not be entirely hidden by content the site added. W3C's explanation uses a chat window the user closes with the Escape key as its example. So: Escape closes the chat, the launcher has a clear label, and an open chat never fully hides the element in focus.
  • Inside the window. Your business name, the AI disclosure, a visible way to reach a person and a link to your privacy notice. Quick-reply buttons with large tap targets.
  • Honest naming. "The [business name] assistant" works better than a made-up human name with a stock photo, and it keeps you on the right side of the disclosure rule.

AI chatbot for website price: free plans, pricing models and real cost

The AI itself costs less than most people expect. On OpenAI's published API rates, gpt-5-mini costs $0.25 per million input tokens and $2.00 per million output tokens. For example, if a chat runs to eight replies and each request sends about 4,000 tokens of instructions, knowledge and history and gets 150 tokens back, that is 32,000 input tokens (about $0.008) and 1,200 output tokens (about $0.0024). Roughly one cent per conversation, less when repeated input is cached.

So most of what you pay buys the product around the model: booking, the inbox, integrations, handover logic, setup and support. The common pricing models:

ModelHow it is billedWatch out for
Free tierCapped conversations or featuresWhether AI replies and calendar booking are included
Per seatPer team member per monthCost grows with your team, not with your inquiries
Per conversation or outcomePer chat the AI handlesWhat counts, and any monthly minimum
Flat monthlyFixed fee with a usage capThe overage price once you pass the cap
Do it yourselfModel usage plus hostingYour time, upkeep and no team inbox

One published example of outcome pricing: Intercom lists its Fin AI agent at $0.99 per outcome, charged once per conversation. An outcome includes a customer confirming the issue is resolved, a customer who does not ask for more help after Fin responds, or Fin completing a workflow, including handoffs. Used alongside another help desk, Intercom lists a minimum monthly commitment, for example 50 outcomes. That model comes from support. For lead capture the goal is a booked call or visit, so work out what each booking costs you.

Searching for an AI chatbot for website free of charge usually turns up a free trial, a capped free tier or a do-it-yourself build. All are fine for testing. Before real visitors use one, check that it understands free text, books into your calendar and says what happens at the cap.

How to choose the best AI chatbot for website inquiries: a two-week test

Feature lists look alike; two weeks of real use does not. Run this before you sign anything long term:

  1. Write the knowledge: services, service area, hours, how pricing works and what you do not do.
  2. Set three to five questions per type of inquiry, in order.
  3. Write the booking and handover rules from the sections above.
  4. Run ten test chats yourself: a clear booking, a vague visitor, a price question, someone out of area, an existing customer, a complaint, someone asking for a person, an emergency, a visitor who disappears after one message, and one typed fast on a phone with typos.
  5. Go live and read every transcript for the first week. Every wrong answer points to a missing line in your knowledge.
  6. Track four numbers: the share of chats that leave contact details, the share that book, how fast handovers get a human reply, and how many bookings turned out to be a poor fit.

Tools such as Konverto put the website chat widget into the same shared inbox as SMS and email leads. Konverto replies in your tone, asks the qualifying questions you chose, books a call or site visit into your calendar and hands the conversation to a person when a lead asks for one or a call-back, or raises a complaint.

Whichever tool you choose, read ten transcripts and you will know whether your chatbot is winning inquiries or quietly losing them.

Frequently asked questions

What questions should a chatbot for a website ask?

Start with the need in the visitor's own words, then fit (which service, which area), then timing, then contact details with a reason. Three to five questions is enough. Drop any question whose answer does not change whether you can help, who takes the job or which slot you book.

How much does an AI chatbot for a website cost?

Vendors charge per seat, per conversation or outcome, or a flat monthly fee. Intercom, for example, lists its Fin AI agent at $0.99 per outcome. The AI model itself is cheap: on OpenAI's published rates for gpt-5-mini, a typical eight message chat costs around one cent, so most of the price pays for booking, the inbox, integrations and setup.

Is there a free AI chatbot for website use?

Usually as a free trial, a free tier with a cap on conversations or features, or a do-it-yourself build where you pay only for model usage. Before you rely on one, check that it understands free text, books into your calendar, routes chats somewhere your team will see them and tells you what happens at the cap.

How do I add a chatbot to my WordPress website?

On a self-hosted WordPress site, install the vendor's plugin or paste its script with a header and footer code plugin. On WordPress.com, installing plugins is available on all paid plans, and a free site has to upgrade first.

Does a website chatbot have to say it is an AI?

In the EU, yes in practice: Article 50 of the AI Act, applicable from 2 August 2026, requires AI systems that interact directly with people to be designed so those people are informed they are dealing with an AI, unless it is obvious, in a clear way at the latest at the first interaction. Elsewhere it is good practice.

Can a website chatbot book appointments?

Yes, if it is connected to the calendar your team actually uses. It should read real availability across everyone who could take the job, offer two concrete options, write the booking into the calendar with the chat summary and send the visitor a confirmation.

Sources

Want every enquiry answered within a few minutes?