A lead generation chatbot answers a visitor's question, then asks for a way to follow up and sends that lead to the person or system that acts on it. The setup that works: answer first, ask for three to five fields, show a short privacy notice, and route every lead to an inbox and your CRM right away.
This guide is the lead-capture part of our guide to adding an AI chatbot to your website. It covers what to ask, how to handle consent, where each lead should go, what happens after hours, and five scripts you can copy.
Key takeaways
- Answer the visitor's question before you ask for anything. A bot that opens with a form is just a form.
- Ask for one contact detail plus the two or three facts that change what you do next. Each extra field is one more reason for a visitor to stop, so test how many you need.
- Tell people why you want their details and how long you keep them, at the moment you ask (GDPR Article 13).
- Route each lead three ways: a leads list you work from, an email you can reply to, and a webhook into your CRM.
- Plan for the moments the bot cannot answer, including when it runs out of AI replies. Those are leads too.
What a lead generation chatbot does: four jobs in order
A lead bot has four jobs, in this order:
- Answer the question the visitor actually asked, from your own content.
- Ask for a way to follow up, one question at a time, only when it makes sense.
- Tell the visitor what happens to their details.
- Route the lead to a person and a system as soon as it is captured, because a lead that waits goes cold. Harvard Business Review audited how quickly companies answered web inquiries in "The Short Life of Online Sales Leads" (March 2011) and concluded that most did not respond nearly fast enough. The data is more than a decade old, so treat it as direction, not a benchmark. Rylvo has no response-time benchmark of its own to offer.
Answer first, ask second
The most common lead-bot mistake is to gate the answer: "Before I can help, what's your email?" A visitor who came with a question and gets a demand may leave, or type a throwaway address, so answer first and measure the drop-off on your own traffic (see the measurement section below).
The pattern that works:
- Reply one answers the question completely, with prices or hours if they are on your site.
- The same reply ends with one short, relevant question: "Would you like me to have someone confirm a time for you? What's the best number to reach you?"
- If they ignore the question, the next reply just helps. It does not ask again straight away.
Rylvo builds this into the engine rather than leaving it to prompt wording: open the bot's Data Collection tab, switch Data Collection on, add the fields, and for a website chat also switch on "Remember anonymous visitors" (website visitors are anonymous until they leave details, and that switch is off by default). Then each turn the engine decides whether to ask and for which field, answering first and holding back on the reply after it asked for a contact detail. A stricter option, "Gate answers until required fields are saved", is off by default. Leave it off unless the details are a real precondition, such as a claim number before a claim lookup.
What to ask: a qualification table by business type
Ask only for what changes your next action. A dentist needs to know whether it is an emergency; a roofer needs the ZIP code; a real estate agent needs to know whether the person is buying or selling.
| Business | Contact field | Qualifying fields (pick two or three) | Why each one matters |
|---|---|---|---|
| Dental practice | Phone | New or existing patient, reason for visit, pain today (yes/no) | Pain today routes to the same-day list |
| Home services | Phone | ZIP or postal code, job type, how soon | Out-of-area jobs are declined politely; urgent jobs go first |
| Real estate | Email or phone | Buying, selling or renting, area, timeline | Sellers and buyers go to different agents |
| Salon or spa | Phone | Service, preferred day, stylist (optional) | Lets the front desk offer a real slot |
| B2B services | Work email | Company, team size, what they want to solve | Size decides who takes the call |
Rylvo's industry pages, such as dental practices, show how this looks for one business type.
Field mapping: from the chat to your CRM
Decide the field keys first, because the keys are what your CRM receives. In Rylvo each field has a stable snake_case key, a type (text, email, phone, number, URL, date, choice, yes/no or free text) and a required flag, and is either asked or inferred (worked out from the conversation when the session ends).
| What the bot asks | Field key (yours to choose) | Type | Where it arrives in the webhook | Typical CRM field |
|---|---|---|---|---|
| "What's your name?" | name | text | data.name, data.fields.name | Contact: first and last name |
| "Best email for the details?" | email | data.email | Contact: email | |
| "Best number to reach you?" | phone | phone | data.phone | Contact: phone |
| "What can we help with?" | need | free | data.fields.need | Deal or ticket description |
| "Which area are you in?" | zip_code | text | data.fields.zip_code | Contact: postal code |
| "How soon do you need this?" | timeline | choice | data.fields.timeline | Custom field: timeline |
| (not asked) | (automatic) | data.first_message | Note: first message | |
| (not asked) | (automatic) | data.channel, data.bot_name | Lead source | |
| (not asked) | (automatic) | data.captured_at, data.lead_url | Created date, link back |
If your site already knows the visitor (a logged-in customer), the widget's RylvoWidget.identify({ name, email, phone, company, fields }) call tells the bot what not to ask. Use the keys name, email, phone and company for those four, and pass any other field by its key under fields. Those details are used only to decide what to ask and are not saved to the Audience.
What do I tell people before I take their email?
This section is practical guidance, not legal advice, and reflects the sources linked here as of November 2026. If you collect details from people in the EU or UK, the GDPR applies, and a few rules shape the conversation:
- A lawful basis. Article 6 of the GDPR lists six lawful bases. For "please call me back", replying to the request is the obvious purpose. Adding the person to a marketing list is a separate purpose, so treat it separately.
- Tell them at the point of collection. Article 13 requires you to say who you are, why you want the details and how long you keep them, when you collect them. The UK regulator's guide, What privacy information should we provide?, lists the contents (checked November 2026).
- Say it is an AI. The EU AI Act requires AI systems that talk directly to people to tell them they are dealing with an AI, unless that is obvious (Regulation (EU) 2024/1689, Article 50(1); in force from 2 August 2026, checked November 2026). Even where the law puts the duty on the software provider, say it in your greeting.
- Texting or calling the number. If you will text or call the number for marketing, check the rules for your country first. Replying to a request for a call back is different from marketing.
A notice that covers the basics in one line, placed in the same reply that asks for the detail:
I'm the [Business] assistant, an AI. If you'd like a call back, I'll pass your
name and number to our team. We use it only to reply and keep it for [period].
Privacy policy: [link]
What Rylvo gives you to back that notice up:
- A consent question. Each bot has an "Ask new users for consent" setting: on first contact the bot asks permission to remember the chat, with wording you can rewrite (up to 500 characters each for the question, the yes and the no).
- Memory levels. Each person has one: no memory, 7 days, 365 days (one year) or 730 days (two years, with recall). A scheduled sweep deletes people when their time is up.
- Retention by plan. On Starter, leads are kept while you are subscribed, even after the rest of a person's data expires. On other plans a lead goes when its person does: their memory level, capped by the plan's audience window (7 days on Free, 14 on Lite, 90 on Pro). If you need leads for longer, the CRM is where they should live.
- Erasure and export. Forget or Delete in Audience removes a person and the leads their details produced; the Leads page exports CSV.
For the full checklist, see Is your AI chatbot GDPR compliant?.
Where the lead goes
A lead that sits in a chat transcript is not a lead. In Rylvo, the first time a live conversation yields an email address or a phone number, four things happen:
- The Leads page gets a row: contact, what they said, which assistant, which channel, when. Each lead has a status (new, contacted, won, lost), and the list has search, filters and CSV export.
- An email goes to the workspace owners and admins with Reply-To set to the lead, so answering is one press of Reply. On paid plans you can add other addresses, such as the front desk; each address confirms before it receives anything.
- An in-app notification reaches every member.
- A
lead.capturedwebhook goes to every event connector the bot subscribes, which is how the lead reaches Zapier, Make, n8n, GoHighLevel or your own endpoint.
One conversation gives one lead per contact: a visitor who confirms the same email twice does not trigger a second alert.
Sending leads to a CRM
On the Leads page, "Send leads to your CRM" creates the connector: paste a Zapier Catch Hook, Make custom webhook, n8n Webhook node or GoHighLevel inbound trigger URL, press Send test lead, and map the fields. The body:
{
"event_id": "evt_lead_…",
"event_type": "lead.captured",
"bot_id": "…",
"data": {
"lead_id": "…",
"email": "casey@example.com",
"phone": null,
"name": "Casey Morgan",
"contact_hidden": false,
"fields": { "email": "…", "name": "…", "need": "…" },
"first_message": "Do you do implants?",
"bot_name": "Acme Dental Assistant",
"channel": "widget",
"channel_label": "your website",
"conversation_id": "…",
"lead_url": "https://rylvo.com/dashboard/leads",
"captured_at": "2026-09-23T10:00:00.000Z"
}
}
Four details save you trouble later:
- Dedupe on
event_id. It is stable per lead, so a retried delivery does not create a second contact. - Check
contact_hidden. When it is true the email and phone are withheld from the event; open the lead in Rylvo instead of importing an empty contact. - Deliveries are signed (Standard Webhooks) and retried, and events that still fail after the retries are kept so you can replay them from the connector once your endpoint is fixed.
- Test leads are marked as Test. By default a lead from the Test environment is not sent to your CRM webhook and is not emailed to client addresses, so rehearsals do not reach them. (Your own team still gets the in-app notification and the owner email, and an owner can verify a Test destination on a connector to let Test leads through.)
On Starter ($10 a month as an early price for the first 12 monthly billing cycles; new buyers pay $19 from 31 Jan 2027 or after the first 100 customers, whichever comes first), you get 3 connectors, and a lead webhook uses one. The step-by-step setup for each tool is in Chatbot webhooks: send leads to Zapier, Make, n8n or Slack, and the product page is Connectors. Prices checked on 2026-11-17; current plans are on Pricing.
After hours and at zero replies: keep the lead you can still get
The point of a lead bot is the visitor who writes at 11 pm on a Sunday. Three situations deserve a plan.
After hours. The bot answers at any hour and the lead email goes out as soon as the lead is captured. Tell the visitor when a person will follow up ("We open at 8 am Monday") rather than implying an instant call back.
The bot does not know. Instruct it to say so and offer a follow-up: "I don't have that detail, but [name] can confirm it. What's the best number?" That is a lead; a confident guess is a liability. For hot leads you may want a person to take over; live takeover is in Mission Control, a Pro feature that is also in the 7-day Pro trial.
The bot runs out of AI replies. Every paid Rylvo plan includes AI replies each billing cycle (about 650 on Starter, 1,300 on Lite and 2,650 on Pro, on the default model). At zero, the agent tells the visitor the business will get back to them, takes their details and emails the lead to the owner, who is warned at 20% left and again at zero. On the website widget the visitor fills a small form; on a WhatsApp or other messaging channel the sender is already known, so the lead is saved without one. A reply pack, an upgrade or your own provider key (optional) restores answers. Free has no AI after the trial, but it still takes messages and emails you the lead.
If you do not have a website, a shareable Beam link can collect the same details; see AI chatbot without a website.
Five conversation scripts
Use these as the shape of the bot's instructions, and rehearse them in Test before going live. Fill the brackets from your own content; never let the bot invent a price, a slot or a policy. Set the greeting once, for every script: "Hi, I'm the [Business] assistant, an AI. Ask me anything about [topic]." In each script, the reply that first asks for a contact detail also carries the one-line notice from the section above, and each reply asks for one thing.
1. Dental practice
Visitor: Do you do emergency appointments? My tooth is killing me.
Bot: [Your emergency-appointment policy, from your site], and the
first visit is [fee from your site]. I'm sorry you're in pain.
Can I take a number so the front desk can call you with the
first free slot? We use it only for that call and keep it
for [period].
Visitor: 555 0142
Bot: Thanks. And your name, so the front desk knows who to ask for?
For someone in pain, the call back comes before the qualifier.
2. Home services
Visitor: How much to replace a water heater?
Bot: Most standard replacements are [range from your site], including
removal of the old unit. The final price depends on the size and
access. What's your ZIP code? I'll check it against our service area.
Visitor: 30307
Bot: [Yes, we cover 30307 / Sorry, that is outside our area, from
your service-area list]. If you'd like a firm quote, what's the
best number for our technician to call? We use it only to
arrange that call and keep it for [period].
3. Real estate
Visitor: What's the market like in Riverside right now?
Bot: [Two or three facts from your latest market page.] Are you
thinking of buying or selling there?
Visitor: Selling, probably in spring.
Bot: Then [your seller next step, from your site] is a good first
step. What email should [agent] send it to? We use it only to
send that and keep it for [period].
4. Salon or spa
Visitor: Do you have anything Saturday for a cut and color?
Bot: Cut and color takes about [duration] and starts at [price].
I can't see the live diary, but I can ask the team to check
Saturday for you. What's the best mobile number to reach you?
We use it only to reply and keep it for [period].
Visitor: 555 0177
Bot: Thanks. And your name?
The bot is honest that it cannot see the booking diary. If you connect a booking tool, change the script to offer real times.
5. B2B services
Visitor: Do you integrate with NetSuite?
Bot: Yes, [how, from your docs]. Most teams start with [first step].
How big is the team that would use it?
Visitor: About 40 people.
Bot: Thanks. If you share your work email, [name] can send a short
plan for a team that size. We use it only for that and keep
it for [period].
How do I measure a lead generation chatbot?
Count leads per 100 conversations, not leads alone. A bot that talks to twice as many people and gets twice as many leads has not improved; one that gets more leads from the same traffic has.
| Metric | What it tells you | Where to see it in Rylvo |
|---|---|---|
| Lead rate | Share of conversations that left a lead | Analytics, "Conversations & leads" room |
| Reachable | Share of leads with an email or phone | Analytics |
| Followed up | Share of leads moved past New | Leads page statuses, Analytics |
| Won | Share of leads marked Won | Leads page, Analytics |
| Knowledge gaps | Questions the bot could not answer | Analytics |
Analytics rooms are on Lite and up, and the 7-day Pro trial includes them; on Starter, the Leads page statuses and CSV export cover the basics. Two habits matter more than any dashboard: mark every lead's status, and read ten conversations a week that produced no lead. The ten metrics worth tracking are in Chatbot analytics: 10 metrics that show it is working.
An illustrative example (made-up numbers, not a benchmark): 400 conversations in a month produced 28 leads, 19 reachable, 12 followed up and 3 won. That is a lead rate of 7 percent, 68 percent of leads reachable and 11 percent of leads won. Next month, change one thing (the second question, say) and compare.
Hot, warm or cold
Sort each lead when it arrives, so the follow-up matches the urgency. Set your own thresholds; the table is a starting shape.
| Lead | What the fields say | Who follows up, and how fast |
|---|---|---|
| Hot | Reachable, a clear need, a near timeline (or pain today) | A person, the same hour you are open |
| Warm | Reachable, a clear need, a vague or later timeline | A person or a templated email, the same business day |
| Cold | Email only, no stated need, or outside your area | A short reply; add to a list only if the person agreed to it |
When a different tool fits better
If your leads already arrive through a booking system with its own chat, or your sales team lives in a CRM with a built-in chat widget, start there; a separate bot adds a hop. A lead bot earns its place when visitors ask real questions before they are ready to book.
Frequently asked questions
Do chatbots generate leads?
A chatbot does not create demand, but it converts the demand you already have. It answers a visitor's question at the moment they ask it, including at night and on weekends, and then asks for a way to follow up. The leads come from questions that would otherwise go unanswered until the visitor left.
What questions should a lead bot ask?
Ask for one contact detail (email or phone), the visitor's name, what they need, and one or two qualifying facts that change what you do next, such as location, timing or budget. Answer the visitor's question first, ask one thing at a time, and stop at three to five fields.
How do I connect a chatbot to my CRM?
Send each captured lead to a webhook. In Rylvo, create an event connector subscribed to lead.captured and paste the URL of a Zapier Catch Hook, a Make custom webhook, an n8n Webhook node, a GoHighLevel inbound trigger or your own HTTPS endpoint. Map the payload fields to your CRM fields once, and send a test lead to check the mapping.
Is collecting leads by chatbot GDPR compliant?
It can be. Under the GDPR you need a lawful basis for collecting the details, you must tell people at the point of collection who you are, why you want the details and how long you keep them, and you must respond to access and erasure requests (erasure applies on the grounds the GDPR lists). Use a short notice in the chat with a link to your privacy policy, keep only the fields you need, set a retention period, and get advice for your own case.
What if the chatbot can't answer?
Tell it to say so plainly and offer to have a person follow up, then take the visitor's details. That turns a gap in your content into a lead instead of a wrong answer. In Rylvo the same thing happens when a workspace runs out of AI replies: the agent tells the visitor the business will get back to them, takes their details and emails the lead to the owner.
