A website chatbot captures more useful leads when it helps before it asks. Answer the visitor’s question, request only the details you need to follow up, say plainly that the chat is AI, and offer a route to a person. Add accessible design, a short privacy notice and honest measurement, and the chatbot earns trust as well as contacts.
What should a lead capture chatbot ask, and when?
Think of the exchange from the visitor’s side. They came with a question. If the first thing they meet is a form in a chat bubble, they have no reason to fill it in.
Help first
Let the chatbot answer the question that brought the visitor to the chat. Once it has been useful, there is a natural moment to ask: “Want me to send that quote to your email?” or “Shall I book a time with our team?” The request now has a purpose the visitor can see.
Ask for less
Collect what the next step requires and nothing more. For most businesses that means:
- A name
- One way to reach the person (email or phone, their choice)
- A short description of what they need
The FTC’s Start with Security guide for business makes the case in one line: “No one can steal what you don’t have.” The same guide advises holding on to information only as long as you have a legitimate business need. Fewer fields are easier for the visitor and safer for you.
Ask one thing at a time
A conversation works best one question per message. Explain why you are asking (“so we can send your estimate”) and what happens next (“someone will reply within one business day”, if that is true for your team). Let people skip optional questions. If someone prefers not to share details, keep helping them anyway.
How should the chatbot hand off to a person?
Some visitors want a person, and some questions need one. Plan the handoff as carefully as the chat itself.
Nielsen Norman Group’s research on chatbot user experience is a small qualitative study from 2018, and its findings still read as common sense. Participants were annoyed when a bot repeated the same answers. They preferred a bot that admitted it did not understand over one that gave a confidently wrong answer. They viewed an escape hatch to a real human, such as a live agent or a phone number, favorably.
In practice:
- Make the exit visible. Offer “talk to a person” as an option from the start, and again whenever the chatbot cannot answer.
- Pass the context along. Send the conversation summary and contact details to the person taking over, so the visitor does not repeat themselves.
- Be honest about timing. If no one is available, say so, take a message and state when the visitor can expect a reply.
- Follow up fast. A captured lead only has value if someone responds. See speed to lead and AI follow-up for how to set that up.
Should the chatbot say it is AI?
Yes. The Nielsen Norman Group article calls it a mistake when businesses do not tell customers they are talking to a bot. Its participants were pleased when a business was transparent, and knowing they were dealing with a bot helped them adjust how they phrased things.
Put the disclosure where people will see it: in the chatbot’s name or greeting, before the visitor types anything. “Hi, I’m the virtual assistant for our team” is enough.
Disclosure can also be a legal matter, depending on where you and your visitors are. Our guide to AI disclosure laws for chatbots covers that side.
How do you make a chat widget accessible?
A chat widget is part of your website, so the Web Content Accessibility Guidelines (WCAG) 2.2 apply to it like any other page element. These five success criteria matter most for chat, and your developer or chatbot vendor should be able to confirm each one.
| WCAG 2.2 success criterion | Level | What it means for a chat widget |
|---|---|---|
| 2.1.1 Keyboard | A | Visitors can open the chat, type, send, choose options and close it using only a keyboard |
| 2.1.2 No Keyboard Trap | A | Keyboard focus can always move out of the chat window and back to the page |
| 1.4.3 Contrast (Minimum) | AA | Message text has a contrast ratio of at least 4.5 to 1 against its background |
| 2.5.8 Target Size (Minimum) | AA | Buttons such as send, close and quick replies are at least 24 by 24 CSS pixels, with limited exceptions |
| 4.1.3 Status Messages | AA | New messages and notices are announced by screen readers without moving the visitor’s focus |
The last row deserves a closer look, because chat is built on content that appears without a page reload. W3C’s explanation of status messages says the aim is to make users aware of important changes in content that are not given focus, in a way that does not interrupt their work. For a chatbot, that means a screen reader user hears each new reply as it arrives.
Keyboard access is the other common gap. W3C’s explanation of the keyboard criterion notes that it supports people who cannot use a mouse, including blind users and people who rely on speech input.
Also check that every input has a label or instruction (criterion 3.3.2) and that the chat launcher does not cover important page content on small screens.
What privacy notice does a chatbot need?
Tell people what you are collecting and why, at the moment you collect it. A short line near the first question works well: “We use your details to answer your request. See our privacy policy.” Link the policy from inside the chat.
Cover these basics:
- What information the chatbot collects
- What you use it for
- Whether the conversation is stored and who can see it
- Which outside providers process it, such as your chatbot vendor
- How a visitor can ask for their information to be deleted
Some laws make this a requirement. As one example, the California Attorney General’s CCPA page explains that businesses covered by that law must provide a notice at or before the point where they collect personal information, and the notice must list the categories of personal information collected. Other states and countries have their own rules. This is general information, so confirm what applies to your business with your own counsel.
How do you measure chatbot lead conversion?
Count the whole path, so you can see where visitors drop off.
| Metric | How to calculate it | What it tells you |
|---|---|---|
| Chat start rate | Chats started divided by visitors who saw the widget | Whether the greeting and placement invite a conversation |
| Lead capture rate | Chats that produced contact details divided by chats started | Whether the ask comes at the right moment |
| Handoff rate | Chats passed to a person divided by chats started | How often the chatbot reaches its limits |
| Qualified lead rate | Leads your team accepts divided by leads captured | Whether the chatbot attracts the right people |
| Lead to customer rate | New customers divided by qualified leads | Whether chat leads turn into revenue |
To track this in Google Analytics, send an event when a visitor submits contact details in the chat and mark it as a key event. Google’s Analytics Help defines a key event as an event that measures an action that is particularly important to the success of your business.
Review transcripts as well as numbers. Reading a sample of real conversations each week shows you which questions the chatbot missed and where visitors gave up. For a wider view of which type of tool fits your site, see AI chatbot vs AI agent.
The short version
Help first, then ask for only what you need. Say the chat is AI, keep a person within reach, and build the widget so everyone can use it. Explain what you collect, then measure the path from chat to customer and improve the weakest step.



