Click-to-Messenger Ads + an AI Chatbot That Answers Every Lead

A click-to-Messenger ad does something a normal ad can't. Instead of sending people to a landing page they'll skim and close, it drops them straight into a conversation with your business. They tap "Send message", a chat opens, and they ask the thing they actually want to know: "How much is it?", "Do you deliver to Pune?", "Is this available in size 9?"

Then nobody answers for four hours.

That gap is where most message ads lose their money. Below: why speed matters so much in a DM, how to train a bot on your real questions, how to connect it, launch the ad, and hand off to a person.

Why message ads die without instant replies

When someone taps a message ad, you've paid for their attention right then. A landing-page visitor can come back later; a person who opened a chat and got silence usually doesn't. They've scrolled on, possibly to a competitor's ad.

The research on lead response time points the same way. A 2011 Harvard Business Review study of more than 2,000 US companies found that firms that responded to online leads within an hour were nearly seven times more likely to qualify the lead than firms that waited even one hour longer. The same study found that the average response time was 42 hours. That was web forms. In a chat, where the person is staring at the screen, you have minutes.

There's also a mechanical reason. Messenger gives businesses a 24-hour window to reply freely after a person messages them. Reply after that and you're limited to a narrow set of message types. A lead that arrives at 11pm on a Friday and gets a reply on Monday morning is, for practical purposes, gone.

So the economics of a message ad are simple. You pay per click or per conversation started. If half those conversations never get a timely answer, your real cost per lead is double what Ads Manager shows you.

What the chatbot actually needs to handle

Before you build anything, pull up the last 50 to 100 chats your Page received. You'll notice most of them are the same eight or ten questions. For a typical small business it looks something like:

• Price, and whether there's a discount
• Delivery area and delivery time
• Availability (stock, sizes, appointment slots)
• What's included, or how it compares to the cheaper option
• Payment methods, cash on delivery, EMI
• Returns and refunds
• Location and opening hours
• "Can I talk to someone?"

Write those down with the exact wording people use. That list is your training set and your test set.

The goal is not a bot that can talk about anything. It's a bot that answers those eight or ten questions instantly and correctly, and knows when to stop.

Step 1: Train the bot on your FAQs

A chatbot is only as good as the content it reads. If your website's FAQ page says "Contact us for pricing", the bot will say the same thing, and the lead will leave.

1. Write a plain FAQ document. One question per heading, one direct answer underneath. Include real prices, real delivery zones, real timelines. Phrases like "starting at ₹1,499, delivered in 3 to 5 days across Maharashtra" are what you want. Avoid marketing copy.
2. Add it as a source in MagicChat. Paste your website URL, upload the FAQ as a PDF, or both. MagicChat indexes it and answers only from what you've given it.
3. Add the awkward answers too. Leads ask about refunds, delays, and "why is yours more expensive". If you don't write an answer, the bot will either dodge or guess. Write the honest version.
4. Test with the real wording. Take your list of questions and ask them exactly as customers typed them, typos included. "pune delivery?" is a different query from "Do you deliver to Pune?" and both need to work.
5. Ask three questions it shouldn't know. Note how it responds when your content doesn't cover something. You'll use that phrasing later to trigger a handoff.

If you've already set up a support bot on your site, you can reuse it. The same knowledge base works in Messenger; you're just adding a new place for questions to come from. Our guide to building an AI customer support agent goes deeper on choosing sources and setting tone.

Step 2: Connect the bot to Messenger

MagicChat answers through a simple public endpoint: you send it your bot_id and the customer's question, and it sends back an answer. Anything that can make an HTTP request can use it. For Messenger, that means you need something in the middle that receives the incoming message and makes that request.

The most common setups:

• ManyChat: if your Page already runs on ManyChat, add an External Request action (available on the Pro plan) to your default reply flow. POST the user's last message to https://www.magicchat.ai/api/public/ask-ai-query with your bot_id, map the answer field from the response to a custom field, and send that field back as the reply.
• n8n or Make: if you already receive Page messages through a webhook, add an HTTP request step between the incoming message and the reply. Same request, same answer field.
• Your own server: a small webhook that listens for Messenger events, calls the endpoint, and posts the reply through the Send API. This is maybe 40 lines of code.

The request itself is identical to the one in our walkthrough on automating email replies with the MagicChat API. That post shows the exact request and response, including what an error looks like.

Before going live, message your own Page from a personal account and run through your full question list. Check that answers arrive in a few seconds and that they read well on a phone screen. Shorten any answer that runs past three or four lines; a website paragraph is a wall in a chat bubble.

Step 3: Launch the ad

With the bot answering, the ad is the easy part. A few setup choices matter:

• Objective: choose Engagement or Leads, and set the conversion location to Messenger (or Messenger plus Instagram, if your bot is connected to both).
• Message template: Meta lets you set a greeting and a few suggested questions that appear when the chat opens. Use your top three real questions from the list above. People tap the suggestion instead of typing, and your bot already knows exactly how to answer those.
• Creative: make the ad ask a question your bot can answer. "Ask us if we deliver to your area" works better than a generic "Message us", because it sets up the first message.
• Budget: start small. ₹500 to ₹1,000 a day, or $10 to $20, is enough to see whether conversations turn into sales before you scale.

You can build this in Ads Manager the usual way. If you'd rather not click through it, you can also set the campaign up from Claude by describing the audience, budget and objective in plain English. The campaign is created paused, so you can check it in Ads Manager before anything spends.

Whichever route you take, set a spend cap for the first week. Testing should cost a small, known amount.

Step 4: Hand off to a human at the right moment

The bot's job is to answer the routine questions so your team doesn't have to. It is not meant to close every sale on its own. In my experience the handoff is where sales actually close, so give it clear triggers.

Hand the chat to a person when:

• The lead asks for one. "Can I talk to someone" should never get another bot answer. In MagicChat you can add a Talk with human button that opens a WhatsApp chat with your team, with a pre-filled message so they know the context.
• The bot doesn't know. Use the phrasing you noted in Step 1. If the answer contains it, route the conversation to a human inbox instead of sending it. In ManyChat that's a condition on the custom field plus a live-chat assignment.
• The lead is clearly ready to buy. Questions about payment, booking a specific slot, or bulk orders are buying signals. Tag them and notify a person straight away.
• It's a complaint. A refund dispute or an angry message needs a human, even if the bot technically has an answer.

For high-value leads, a reply in chat isn't always enough. If someone shares their phone number and asks about a big order, a call within minutes will often close it faster than any message thread. Tools like CallSimple can place that first follow-up call automatically, so the hottest leads hear from you while they're still holding their phone.

What to measure after the first week

Ads Manager will show you cost per conversation started. That number on its own doesn't tell you much. Track these as well:

• Bot answer rate. Out of all incoming questions, how many did the bot answer without a handoff? Under 60% usually means your FAQ content has gaps.
• Handoff response time. When the bot passes a chat to a human, how long until a person replies? This is the new bottleneck, so keep it under 15 minutes during business hours.
• Conversations to sales. Count them by hand if you have to. Ten conversations that led to two sales tell you far more than a dashboard full of click-through rates.
• Unanswered questions. Read every chat where the bot punted. Each one is a missing FAQ entry. Add it, re-index, and the bot covers it from then on.

You will get questions in week one that you never thought of. Expect it. Filling those gaps is most of the job.

A realistic starting setup

If you want the short version:

1. Pull your last 50 chats and list the top ten questions.
2. Write direct answers with real numbers and add them to a MagicChat bot.
3. Connect the bot to Messenger through ManyChat, n8n or a small webhook.
4. Launch a small message ad with those top questions as the suggested replies.
5. Route "talk to a human", unknown answers and buying signals to your team.
6. Read every handed-off chat at the end of each day and fill the gaps.

Most message ads that fail have decent creative. They fail because the reply came too late. A bot trained on your own answers fixes the reply, at 2pm or 2am, and leaves your team the conversations worth having.