Learn how to qualify leads in website chat, WhatsApp and Messenger with an AI agent: choose criteria, ask well, route hot leads to people, and tune.
Most chat enquiries are not ready to buy, and your sales team cannot tell which ones are until someone asks the right questions. An AI agent can ask those questions at any hour, in the visitor's language, and pass only the promising conversations to a person. This article shows how to define a qualified lead, design the conversation, route each outcome, and improve the process over time.
What "qualifying" means in a chat context
Qualification is the step between "someone said hello" and "someone on the team should talk to this person now." In a form-based world, that step happens after submission, when a salesperson reads the answers. In chat, you can do it during the conversation.
A good qualification flow does three things:
- It collects the few facts your team needs to decide whether to follow up.
- It lets the visitor get useful answers while they provide those facts.
- It ends with a clear next step: a booked call, a human handoff, a self-serve link, or a polite "here is where to start."
Notice what it does not do: it does not interrogate. The aim is a conversation that feels like talking to a well-informed receptionist, not a ten-field form broken into ten messages.
Decide your criteria before you write a single prompt
AI will qualify exactly the way you describe it. If your team has never written down what "qualified" means, the agent will improvise, and so will your salespeople.
Start with a short list of signals. Most businesses can work with four categories:
| Signal | What it tells you | Example question |
|---|---|---|
| Need | Is there a real problem you solve? | "What are you hoping to get done?" |
| Fit | Are they the right kind of customer? | "Roughly how big is your team?" |
| Timing | Is this a project or a browse? | "When are you hoping to get started?" |
| Authority or budget | Can they decide or influence? | "Are you choosing this yourself or with others?" |
You do not need all four for every business. A local clinic may care only about need and timing. A software vendor may care about fit and authority. Pick two or three and write a plain-language definition of each outcome.
Write three outcomes, not a score
A numeric score looks precise but is hard to defend. Three outcomes are easier for everyone to understand and test:
- Ready now: clear need, near-term timing, right fit. A person should reply quickly.
- Interested, not yet: real need but vague timing, or missing fit details. Nurture with information and a booking link.
- Not a fit: outside your service area, product range or customer type. Point them somewhere helpful.
Write each outcome as a rule in one or two sentences, for example: "Ready now means the visitor has described a specific problem and expects to start within the next two months." You will paste language like this into the agent's instructions later.
Ask questions the way a person would
The biggest quality difference between a good and a bad qualification bot is tone and sequencing. A few principles carry across website chat, WhatsApp and Messenger.
Answer first, then ask
If a visitor asks about pricing or delivery, answer from your knowledge base before asking for anything. Someone who gets value in the first message will usually keep talking. Someone who hits a question wall will close the tab.
One question at a time
On a phone, a message with four questions gets one reply, usually to the last one. Ask a single question, wait, then move on. Let the answer shape the next question rather than running through a fixed script.
Make questions earn their place
For each question, ask yourself what decision it changes. If the answer does not alter the routing, drop it. "What is your company name?" matters for a B2B lead you will email; it does not matter for a visitor asking about opening hours.
Adapt to the channel
- Website chat: visitors are often anonymous and browsing. Keep early questions light and ask for contact details only once they show intent.
- WhatsApp: you already have a phone number, so asking for it again is a mistake. Spend your questions on need and timing instead.
- Messenger: the person's name is usually known, but not their email or phone. Ask for a preferred contact method only when you are about to hand off.
Route every outcome somewhere useful
Qualification only matters if the result changes what happens next. Map each outcome to an action before you go live.
- 01Visitor asks a question
- 02Agent answers and asks 2–3 qualifying questions
- 03Outcome decided: ready, not yet, or not a fit
- 04Handoff, booking link, or helpful pointer
Ready now
Hand the conversation to a human quickly, with a summary of what the visitor said so the salesperson does not re-ask. If your team works in shifts, route to whoever is on duty. Tell the visitor honestly what will happen next, and never promise a response time you cannot keep.
Interested, not yet
Offer something low-pressure: a booking link for a later call, a relevant guide from your knowledge base, or an invitation to come back. Capture contact details so that your team can follow up when timing changes.
Not a fit
Be kind and specific. If you do not serve their region, say so and, if you can, suggest an alternative. Do not capture a lead you will never contact; it clutters your records.
Set guardrails so the agent stays honest
An AI agent that qualifies leads is also your first impression. A few rules prevent most problems.
- Do not invent offers. Tell the agent not to quote prices, discounts or delivery dates unless they appear in your knowledge base.
- Do not pressure. Qualification questions should be skippable. If a visitor declines to share their budget, the agent should continue helping.
- Escalate on request. When someone asks for a person, hand over immediately. Hiding the human option destroys trust faster than any wrong answer.
- Respect consent. Say why you are collecting contact details and how you will use them. If you operate across several regions, check what your own privacy policy and local law require; that is a question for your legal adviser, not your chatbot.
Measure what happens, then tune
Do not judge the first version by feel. Read real conversations and check three things.
- Did the agent ask the right questions? Look for questions that visitors ignored or found confusing.
- Did the outcome match reality? Compare what the agent classified as ready with what your sales team thought after talking to the person.
- Where did visitors drop off? If many leave right after a particular question, rewrite or remove it.
The following chart is a hypothetical illustration. Suppose a shop receives 300 chats in a month and wants to understand where they end up.
In this made-up example, only 40 of 300 chats need a salesperson, and half the chats are plain support questions. That is the whole case for qualification: the team's time goes to the 40, while the other 260 still get a useful answer. Your real proportions will differ, so measure your own before setting targets.
Tune one thing at a time
Change a single question or rule, test it, and read the next batch of conversations. If you change five things at once, you will not know which one helped.
Common mistakes to avoid
- Too many questions up front. Three well-chosen questions beat eight.
- Qualification without answers. A bot that only asks questions feels like a gatekeeper.
- No human path. Every flow needs a clear way to reach a person.
- Criteria nobody agreed on. If sales and marketing disagree about "qualified," the bot cannot fix that.
- Never reviewing transcripts. The first week of real conversations always teaches you something your planning missed.
Doing this in AI SNS CHAT
AI SNS CHAT does not apply an automatic lead score. You describe your qualification rules to the agent, and the platform gives you the tools to capture, store and route the result.
1. Write the rules into the persona. Open your agent and set its persona and guardrails. Paste in your three outcomes, the questions that decide them, and the rules above: answer first, one question at a time, no invented offers. The Playground lets you test the conversation before it goes live, and you can compare models side by side with Playground A/B model compare.
2. Define what to capture. Under Contacts, use Custom attributes for the facts you care about, such as team size or timeline. The description of each field tells the agent when to ask for it, so a clear description like "Ask for this only after the visitor describes a project" shapes the flow directly.
3. Collect the lead. Add the "Collect leads" Action, or build a Custom form, from the agent's Actions. Each submission appears under Activity → Leads, and every lead also creates a contact, so your records stay in one place. Actions are available from the Hobby plan up, with 5, 8, 12 or 20 Actions per agent depending on the plan.
4. Hand off ready leads. Use the Escalations Action for visitors who are ready or who ask for a person. Under Helpdesk → Scheduling, set shifts per person and timezone, and handoffs are assigned to whoever is on duty. Alerts can go to Slack, Teams, LINE WORKS, WhatsApp, email or a webhook. The customer-facing "what happens next" line is based on your rota, and the system never invents a wait time. Helpdesk needs Standard or higher.
5. Offer a booking step. For leads that are interested but not ready, the Cal and Calendly Actions let the agent offer a booking link.
6. Review and improve. Open Activity → Chat logs to read conversations on chat channels, flag or escalate them, and rate replies with thumbs up or down. If you spot a better answer, edit a reply into the ideal response to train the agent. Analytics shows Chats, Topics, Question history and Sentiment; advanced analytics need Pro.
Your agent can be deployed to the website chat widget, WhatsApp, Messenger and many other channels. See all channels, the feature overview and plans and pricing, or start free and test the flow in the Playground first.
Frequently asked questions
Can an AI agent really tell a good lead from a poor one?
It can apply the rules you give it, such as need, timing and fit, to what a visitor says. It is not a replacement for human judgment, so review conversations regularly and compare the agent's outcomes with what your sales team finds after talking to the person.
Does AI SNS CHAT score leads automatically?
No. The agent follows the qualification rules you write in its persona, captures the answers through the Collect leads Action or a custom form, and stores them in Activity → Leads and Contacts. Routing to a human is done with Escalations and shift-based handoff.
How many qualifying questions should the agent ask?
Usually two to three, one at a time, and only those that change what happens next. Always answer the visitor's own question first, and let them skip anything they would rather not share.
Can I follow up with leads who said "not yet"?
You can offer a booking link during the chat, and your team can use the captured contact details. The Outbound Beta feature, available on Standard or higher, only reaches people who messaged first, carries a STOP opt-out on every message, and limits WhatsApp sends to the 24-hour window. It does not run SMS or email marketing campaigns.


