Website chatbots that actually capture leads: setup, scripts and handoffs

What makes a chatbot capture leads instead of just answering questions?

A lead-capturing chatbot is built around an outcome: a qualified visitor who leaves contact details, books a time or reaches a person. An FAQ bot is built around answers. Both can use the same AI, but they are designed, scripted and measured differently.

Most website chat widgets are FAQ bots in practice. They answer "What are your hours?" well, then let the visitor leave. The table shows what changes when you build for leads instead.

The underlying AI chatbot technology matters less than this design. We would back a well-scripted bot on a modest model over a clever model with no goal.

What should a lead capture chatbot ask?

Ask the three to five questions your best salesperson asks first, and nothing else. Each question should change what happens next: who gets the lead, how fast, or what the bot offers. If an answer would not change anything, cut the question.

Most businesses need some version of these:

1. Need. What do you want help with? Offer buttons for your main services.
2. Timing. When do you need it? "This week", "This month", "Just researching."
3. Fit. One question that tells you whether you can serve them: location, size, type of case, budget range.
4. Decision. Are you the person deciding, or choosing for someone else?
5. Contact preference. Call, text or email, and the best time.

Here is what the "fit" question looks like across different businesses:

Three rules keep qualification from feeling like an interrogation:

• One question at a time. Never stack three questions in one message.
• Buttons where answers are predictable. Free text where they are not.
• Never ask what you already know. If the visitor came from your "Emergency repairs" page or a "Family law" ad, the bot should say so, not ask.

Qualification rules belong in writing, not in the bot's head. Decide which answers make a lead "qualified", which make it "nurture" and which make it "not a fit", and how each is routed.

When should the chatbot ask for contact details?

Ask once the bot has been useful and the visitor has shown intent: they asked about price, availability, a quote or next steps. Asking in the first message, before helping, turns the chat into a form with extra steps.

There are two exceptions. If the visitor's first message is already high intent ("Can someone call me about a leak?"), ask straight away. And if the visitor is on a page with one clear purpose, such as "Book a consultation", collect details as part of booking.

How to ask:

• Give a reason. "So [Name] can send your quote, what's the best email?" works better than "Please enter your email."
• Ask for one channel first. Email or phone, whichever your team follows up on fastest. Ask for the second only if you need it.
• Never hold the answer hostage. If someone asks your opening hours, answer, then offer the next step.
• Save partial leads. If a visitor answers your qualifying questions and gives an email but leaves before booking, that is still a lead. Save it and follow up.
• Get consent for automated follow-up. If you plan to text or call the lead using automated tools, collect that consent in the chat as carefully as you would on a form. Our guide to whether AI phone calls are legal explains why marketing calls need written consent in the US.

How do you book appointments inside the chat?

Connect the chatbot to the calendar or booking system your team already uses, and let it offer real slots. A qualified visitor who can commit to a time in the chat doesn't have to wait for a reply, or keep shopping around while they wait.

Good in-chat booking follows five rules:

1. Offer two or three specific times, not a whole calendar. "I have Tuesday at 10 AM or Wednesday at 2 PM. Which works?"
2. Book only qualified visitors. Visitors who do not meet your rules get a different next step, such as a resource, a referral or a callback.
3. Book the right thing. Some work needs a quote, a site visit or a short call before an appointment. Book that instead.
4. Confirm outside the chat. Send an email or text confirmation with the time, the address or video link, and a reschedule link.
5. Assign an owner. Every booking lands in the CRM with an owner and the conversation summary, so whoever takes the meeting is prepared.

If you can't connect a calendar yet, offer a callback window instead ("We'll call you between 2 and 4 PM today") and make sure someone actually calls. The appointment booking use case shows the full flow.

What scripts should a lead capture chatbot use?

Use short, plain scripts with a clear next step in every message. These templates cover the five moments that decide whether a conversation becomes a lead. Replace the bracketed parts, and keep each message under about 40 words.

1. Opening (page-specific)

"Hi, I'm [Business]'s AI assistant. Looking for help with [page topic: a leaking roof / a family law matter / a bulk order]? I can answer questions, check availability and book you in."

Opening (after hours)

"Hi, I'm [Business]'s AI assistant. Our team is offline until [time], but I can answer questions and book you a time now. What can I help with?"

2. Qualification

"Happy to help. To point you to the right person, a couple of quick questions. First: what do you need help with?" [buttons: Service A / Service B / Service C / Something else]

"Thanks. And when do you need this?" [buttons: This week / This month / Just researching]

"Got it. [Fit question from the table above]"

3. Contact capture

"Thanks, [first name if given]. [Answer to their question.] I can have [team or role] send you [a quote / available times / next steps]. What's the best email to send it to?"

"And is there a phone number we can reach you on if anything needs clarifying? It's optional."

4. Booking

"Based on what you've told me, the next step is a [15-minute call / free consultation / site visit]. I have [slot 1] or [slot 2]. Which suits you?"

"You're booked for [time]. I've sent a confirmation to [email]. If you need to change it, use the link in that email."

5. Handoff

"That's a good question for [role]. I'm passing this conversation to them now with everything you've told me. [Live: They'll join this chat in a few minutes.] [After hours: They'll reply by [time] at [email or phone].]"

Fallback (when the bot doesn't know)

"I don't want to guess on that. I'll pass it to [role], who can answer properly. What's the best way to reach you?"

The fallback is the most important script. A bot that says "I don't know, here's a person" builds trust. A bot that guesses creates risk, because the business is accountable for what its chatbot says (see the mistakes section below).

When should a chatbot hand off to a human?

Hand off when a person will do better than the bot: when the visitor asks for one, when the stakes are high, when emotions are high, or when the bot is out of its depth. Write these rules before launch and give each one a response target. A human handoff should always carry the context with it.

Three details make handoffs work in practice:

• A summary, not just a ping. The person should see the visitor's name, answers, page, source and a two-line summary of the conversation before they reply.
• One owner. A handoff to a shared inbox with no owner is a handoff to nobody. Route to a named person or rotation.
• A backup. If nobody picks up within the target, the bot tells the visitor when to expect a reply and creates a follow-up task.

How do you sync chatbot leads to your CRM?

Every conversation that produces a lead should create or update a contact in your CRM, with the same fields your web form uses, so chat leads are followed up, reported and measured like any other lead.

Map these fields at minimum:

• Name, email, phone and contact preference
• Answers to each qualifying question, as separate fields rather than one text blob
• Qualification status: qualified, nurture or not a fit
• Source: page URL, referrer and campaign parameters
• Conversation summary and a link to the full transcript
• Booking time and owner, if booked
• Consent captured, with timestamp and wording, if you will send automated follow-ups

Three integration rules prevent the common failures. Deduplicate on email and phone, so a returning visitor updates their record instead of creating a second one. Start the follow-up the moment the lead is created: an unbooked qualified lead should trigger a task or a follow-up sequence, not wait for someone to notice. Our guide to automating lead follow-up covers what that sequence should say. Track the events in your analytics too. Google Analytics 4 has recommended events for this, including generate_lead ("when a lead has been generated") and qualify_lead ("when a user is marked as meeting the criteria to become a qualified lead"). Firing them from the chat lets you compare chat and form leads by source.

How do you measure whether your chatbot captures leads?

Measure the chatbot on the outcomes it was built for, and compare it against your web form on the same pages. These are the metrics that matter, with how to calculate each:

• Engagement rate = conversations where the visitor sent at least one message ÷ visits to pages with the chatbot.
• Conversation-to-lead rate = conversations that end with valid contact details ÷ engaged conversations.
• Lead-to-qualified rate = leads that meet your qualification rules ÷ all chat leads.
• Lead-to-booking rate = chat leads that book an appointment or call ÷ all chat leads. Track qualified leads separately.
• Handoff rate and time to human = share of conversations handed to a person, and the median time until a person replies.
• Show rate = booked appointments that take place ÷ appointments booked through the chat.

Report medians for time-based measures, not averages, so one slow reply over a weekend doesn't distort the picture. If speed matters in your market, our article on speed to lead explains what the research says about response times.

Do you have to tell visitors they are talking to a bot?

In some places you must, and it is good practice everywhere. California's bot law (Business and Professions Code 17941) makes it unlawful to use a bot online "with the intent to mislead the other person about its artificial identity" in order "to incentivize a purchase or sale of goods or services", and gives a safe harbor to anyone who discloses that it is a bot. The disclosure must be "clear, conspicuous, and reasonably designed to inform" the people it talks to.

In the EU, Article 50 of the AI Act requires AI systems that interact directly with people to be designed so that people are told they are interacting with an AI, unless that is obvious, "at the latest at the time of the first interaction." It applies from August 2, 2026; law firm Hogan Lovells reported that a 2026 agreement to delay part of Article 50 left these transparency duties on that date. The simplest way to comply everywhere is the word "AI assistant" in the greeting, as in the scripts above.

What are the most common chatbot lead capture mistakes?

Most failures come from design, not the AI. These are the ones to check for before launch:

1. No goal. The bot answers questions but never offers a next step. Every conversation should end with a booking, a captured lead, a handoff or a deliberate "not a fit."
2. Asking for details too early. A greeting that demands an email before helping turns the chat into a slower form.
3. Too many questions. Every extra question is another point where a visitor can leave. Keep three to five and cut any that don't change the routing.
4. Letting the bot guess. In Moffatt v. Air Canada (2024), a British Columbia tribunal held the airline liable after its website chatbot gave a customer wrong information about bereavement fares. Rejecting the argument that the chatbot was a separate entity, the tribunal said it "should be obvious to Air Canada that it is responsible for all the information on its website." Limit the bot to approved content, often through retrieval-augmented generation, and use the fallback script whenever it is unsure. Our AI hallucination entry explains why models invent answers.
5. Handoffs to nobody. "Someone will be in touch" with no owner, no target and no task is how chat leads go cold.
6. Leads that stay in the chat tool. If chat leads don't reach the CRM with the same fields as form leads, they won't be followed up or measured.
7. No disclosure. Pretending to be a person creates legal risk in some places and damages trust everywhere.
8. An inaccessible widget. Chat should work with a keyboard and a screen reader. The W3C's guidance on WCAG Success Criterion 4.1.3 explains how status messages, such as a new-message alert, should be announced to assistive technology without moving focus.
9. Never reading transcripts. Transcripts show exactly where visitors hesitate or leave. Read a sample every week and fix the script.

If you want a chatbot built to qualify, book and hand off, connected to your calendar and CRM, see how Concierge, our AI receptionist, handles chat, and how Cadenza, our lead follow-up system, follows up every lead it captures.