A lead messages your WhatsApp number at 9:40pm: "hi, how much for a 3-bed kitchen refit?" That's WhatsApp lead capture at its best and its worst: their phone number is already in the chat. They have not filled in a form, and they won't. They want a number, or at least a reason to think you're the right call.
That is the whole opportunity of WhatsApp lead capture. The contact detail you'd normally fight for arrives with the first message. What you're missing is everything else: budget, timing, location, whether they're serious.
So the job is not "capture" at all. It's qualification, done in a conversation, without sounding like a survey.
This is part three of our series on WhatsApp automation. Part one covered the setup and part two covered grounded auto replies. Here we deal with the commercial end: turning inbound chats into leads a person can act on, and following up without getting your number restricted.
Written by the ChatterMate team. We build an open-source AI support agent, so weigh our product claims accordingly. We've kept them to what we could verify, and we say plainly where we couldn't.
WhatsApp lead capture in ChatterMate: what we can and can't promise
Start with the awkward bit.
ChatterMate has a lead capture page. It describes an agent that answers first, then asks qualifying questions at a sensible moment through an inline card, one field at a time, and routes the result to a rep, a CRM and Slack. It suggests three to five focused questions. It also describes modes: automatic, always ask, or off.
Everything on that page is written about the website chat widget. WhatsApp is not mentioned on it. And our README lists WhatsApp as a Business Cloud API channel with templates, media and 24-hour session handling, but doesn't say the inline capture card works there.
We haven't verified that it does, so we won't tell you it does. A card is a web interface idea anyway. WhatsApp is a thread of plain messages.
What we can stand behind: the README says WhatsApp shares one inbox with the other channels, with the same AI answers, handoff and workflows as the web widget. The workflow builder has Start, AI Response, Human Transfer, Condition and End nodes, with branching. Outbound ticket webhooks exist, and the README lists HubSpot and Pipedrive as OAuth-connected CRMs, though the lead capture page names HubSpot and Salesforce. Those two lists don't match, so check which your version actually ships before you plan around either.
So the recipe below is built from those parts: a workflow that asks questions conversationally, a human who takes over when the lead is warm, and a CRM or webhook on the end. It is a design, not a one-click feature, and we'll mark the places where you need to test.
Why a form is the wrong shape for WhatsApp
Forms ask everything at once. A chat asks one thing, waits, and reacts to the answer. That difference is the entire reason people open WhatsApp rather than your contact page.
Two practical consequences.
You already have the phone number. Don't ask for it. The number is the identity. Asking "what's your number?" in a chat that is literally from their number makes you look like a script.
Every extra question costs you leads. On a form that cost is invisible. In a chat it is visible: the person stops replying. We'd rather ask two questions and have a human finish the job than ask six and lose half the room.
Pick three questions, not eight
The biggest mistake in qualification is writing the questions for your CRM instead of for the customer. A CRM wants twelve fields. A customer will give you about three before they feel interrogated.
Choose by asking one thing: what decides whether a human should call this person today?
For the kitchen refit above, that's probably:
- What is the job, in a sentence? (The agent often has this from the first message.)
- Where is the property? (Decides whether you can even serve them.)
- When do they want to start? (Decides urgency.)
Budget is tempting and usually wrong as question four. People lie about it or dodge it, and a price conversation is what the human is for. For a SaaS trial the three might be team size, current tool, and what triggered the search. A dental practice might need only "new patient or existing" and "which treatment."
The point is that the list is short and each answer changes what happens next. If an answer doesn't change anything, delete the question.
Build it as a workflow, not a prompt
You could write "ask the user three qualifying questions" into the agent's instructions and hope. Sometimes that works. It also drifts: the model skips a question, asks one twice, or decides the lead is qualified because it liked their tone.
A workflow is more boring and more reliable. In ChatterMate's visual builder you can branch on a condition, and that is where qualification belongs. The shape:
- Answer first. If the opening message is a question your docs cover, the AI answers it from your knowledge base. Nobody qualifies a person who hasn't been helped.
- Ask, one at a time. After the answer, a single natural question. Wait. Never two in one message.
- Branch on the answers. In your area and starting soon goes one way. Outside your area goes another, with a polite "we don't cover that postcode yet" and no handoff, so nobody's time is wasted.
- Transfer when warm. Qualified leads go to a Human Transfer node with the answers attached to the conversation, so the person picking up doesn't start from zero.
We covered what a good handoff looks like in human handoff that doesn't frustrate customers. The one line to repeat here: the human should see the whole chat. A salesperson who asks "so what are you looking for?" to someone who just answered three questions has thrown the whole exercise away.
One thing to test before you rely on it. Check that the answers a customer gives actually appear where your team will look, whether that's the conversation view, a ticket, or a pushed CRM record. Send yourself through the flow with a test number and look at what lands. If a field doesn't show up, you'll find out in ten minutes instead of after fifty leads.
Where the lead goes next
Three destinations, in rising order of effort.
The shared inbox. For a small business this is enough. A warm lead lands in front of a human, with the conversation intact, while the customer is still holding their phone. Speed matters more than plumbing. Do this first.
A CRM. Per the README, qualified leads can be pushed to HubSpot or Pipedrive over OAuth. Pushing to a CRM is where deduplication bites: the web lead capture page says it dedupes on email, and a WhatsApp lead might not give you an email at all. Decide now what your unique key is. For WhatsApp, the phone number is the natural one, and you should check how your CRM handles a contact with a phone and no email before you switch the sync on.
Your own automation. Outbound ticket webhooks let you fire a Zapier or n8n flow when a ticket changes. This is the escape hatch for anything the built-in integrations don't cover: a row in a spreadsheet, a message to a sales channel, a booking link. Note the direction. These are outbound, and we found no mention of inbound webhooks, so don't design a flow that needs to push data into ChatterMate this way.
Following up: the rule that decides everything
Here's where WhatsApp stops being like web chat.
When a customer messages you, a 24-hour customer service window opens. Inside it you can send free-form messages. Each new message from them resets the clock. Bird's documentation puts it plainly: once the window closes, you can't send free-form content, and you need an approved template plus permission to contact the person. A template to someone who hasn't messaged you recently starts a business-initiated conversation and needs opt-in.
Read that as a rule for lead follow-up. A lead who went quiet yesterday is reachable today for free-form chat. A lead who went quiet last Tuesday is not. For them you need a template, and you need to be able to show you were allowed to send it.
What counts as permission? Infobip's summary of the opt-in rules, which is a vendor's reading of Meta's policy and not Meta's own text, says businesses should only message people who have shared their number and given explicit consent. It says the consent should make clear that the person is opting into WhatsApp messages, name the business, and say what kind of messages to expect. It also advises keeping records of who consented, when and how, and honoring opt-outs promptly. For the exact current wording, read Meta's own policy, since it changes.
In practice, for lead capture:
- Ask for consent in the chat, as its own plain question. Something like: "Want us to message you here about your quote? Reply YES." A clear yes is a clean record. Burying it inside a long message is not.
- Log it. Who, when, which message. If you self-host, this is a record in your own database, which makes it easy to produce on request. Our note on chatbot data retention covers how long to keep it.
- Keep follow-ups useful. A template that says "your quote is ready" to someone who asked for a quote is service. A template that pitches an unrelated offer is the kind that gets numbers flagged.
- Make stopping easy. If someone replies STOP, stop. Immediately, across every flow.
Templates are approved by Meta before you can send them, and category matters (marketing, utility, authentication). Bird's page also notes that from 1 October 2026 billing for service messages and utility templates changes, with a free monthly allowance per number. That is one provider's description; check current Meta pricing for your country before you model costs.
What we'd skip
A few things look clever and aren't.
Scoring. The lead capture page doesn't say how a lead is judged qualified, and we'd be wary of any chat agent that claims a precise score from three short answers. Use rules you can read: in area, starting within a month, gave a clear job. A person can audit a rule.
Chasing. A sequence of five automatic nudges is how a good lead becomes a blocked number. One follow-up inside the window, one template after, then stop.
Pretending to be a person. If the agent is an AI, say so early. People forgive a bot. They don't forgive finding out.
Qualifying support questions. If the customer's message is "my order hasn't arrived," that's not a lead. Route it as support. A qualification flow that fires on every inbound chat will annoy your best customers.
A thirty-minute test
Before any real lead touches this:
- Write the three questions and what each answer changes.
- Build the workflow with one qualified path and one polite exit.
- Connect a test number and run it twenty times, with typos and with answers that don't fit ("not sure", "ASAP", a voice note).
- Check what arrives in the inbox and the CRM.
- Read how to test an AI chatbot for the longer version of this list.
The voice note case catches people out. Assume some leads will send audio, and decide what the agent does when it can't read the message: ask them to type it, or hand straight to a person.
The best WhatsApp lead capture is mostly restraint. Three questions, one human, one follow-up. If you want to try building it, ChatterMate is open source and free to start, with 50 free AI replies a month, and you can self-host it: chattermate.chat.



