WhatsApp AI Chatbot for Business: Complete Setup Guide 2026

WhatsApp is where Indian customers actually are. 530 million users, 98% open rate, opened 23 times a day. An AI chatbot on WhatsApp that answers queries, qualifies leads, and books calls is not a nice-to-have in 2026 — it is the fastest leverage point for any Indian business under ₹10 crore revenue. This guide gives you a working AI chatbot on WhatsApp by the end. Every step, every tool, every prompt — all included.

WhatsApp Business App vs. API — Which One You Need

The free WhatsApp Business App is for manual conversations. You cannot automate it, you cannot connect AI to it, and you cannot handle volume with it. This guide uses the WhatsApp Business API — the programmable version that lets you connect AI, automation, and your CRM. You access it through a BSP (Business Solution Provider) rather than Meta directly.

Step 1: Get WhatsApp API Access via Interakt (30 Minutes)

1. Go to interakt.shop → click "Start Free Trial" → sign up with your business email
2. In the Interakt dashboard: Settings → WhatsApp Account → click "Connect Number"
3. You need: a phone number not on any WhatsApp account (buy a new SIM or use a landline), your Facebook Business Manager account ID, your GST certificate or MSME registration
4. Follow the embedded signup flow — it opens Facebook directly inside Interakt. Create/connect your Facebook Business Manager → verify your business → add the phone number
5. Meta typically approves within 2–24 hours. You will get an email confirmation. While waiting, complete Steps 2 and 3.

Step 2: Design Your Chatbot Flow (Do This on Paper First)

Before building anything, answer these 5 questions. Every minute spent here saves an hour of rework later.

1. What is the ONE primary job of this chatbot? (Lead qualification / Customer support / Appointment booking / Order tracking) — pick one, not all four
2. What are the top 8 questions customers actually ask you? (Check your WhatsApp chat history right now and list them)
3. What information do you need to collect from every user? (Name, phone, budget, city, product interest — be specific)
4. When should the bot hand off to a human? (Rule of thumb: any complaint, any purchase above ₹10,000, any question the bot answers incorrectly twice)
5. What should happen after the bot collects a lead? (Add to Google Sheet, notify sales rep, book Calendly call, send payment link)

Step 3: Build the AI-Powered Chatbot in n8n (60 Minutes)

There are two ways to build: (A) use Interakt's own no-code flow builder — fast but limited to rigid decision trees, cannot handle free-text intelligently. (B) Connect WhatsApp to an AI backend via n8n — handles any message naturally. This guide covers Option B because it is the one that actually works at scale.

Connect Interakt to n8n

1. In n8n: New Workflow → add a "Webhook" node → Method: POST → Path: "whatsapp-in" → click "Listen for Test Event"
2. Copy the webhook URL from n8n
3. In Interakt: Settings → Developer API → Webhooks → paste the n8n URL → enable "Incoming Message" events → Save
4. Send a WhatsApp message to your business number from your personal phone — n8n should receive the webhook. Check the incoming JSON structure — you need: body.messages[0].text.body (the message text) and body.messages[0].from (the sender's phone number)

Add the AI Response Node

1. Add an "OpenAI" node after the Webhook → Model: gpt-4o-mini → Resource: Chat → Operation: Message
2. In the OpenAI node, set the System Prompt to the following (customise the bracketed sections):

You are a friendly, helpful assistant for [YOUR BUSINESS NAME], an Indian [TYPE OF BUSINESS] based in [CITY].

Your job: answer customer questions, collect lead information, and book discovery calls.

About the business:
[Paste your services, pricing, and USP here — 5-10 lines]

Your personality: warm, professional, concise. Use simple English. If the customer writes in Hindi or Hinglish, respond in Hinglish.

Collect this information naturally over the conversation:
1. Their name
2. What they need specifically
3. Their budget range
4. Their timeline (when do they need this?)
5. Their city

Once you have all 5 pieces of information, say exactly this:
"Thank you [Name]! I have noted your requirements. Our team will call you within 2 hours (10am–7pm IST, Mon–Sat). Your enquiry reference is #[random 4-digit number]."

If asked about pricing, give a range and say exact quotes need a 15-minute call.
If asked to speak to a human, say: "Absolutely! Type HUMAN and I will connect you to our team right away."
Never make promises about delivery times or pricing without saying "approximately".
Never discuss competitors.

1. Set the User Message to: {{ $json.body.messages[0].text.body }}

Send the AI Reply Back via WhatsApp

1. Add an "HTTP Request" node after OpenAI → Method: POST
2. URL: https://api.interakt.ai/v1/public/message/
3. Headers: Authorization: Basic [your Interakt API key from Settings → Developer API] · Content-Type: application/json
4. Body (JSON):

{
  "countryCode": "91",
  "phoneNumber": "{{ $json.body.messages[0].from.replace('91','') }}",
  "callbackData": "lead_chat",
  "type": "Text",
  "data": {
    "message": "{{ $node['OpenAI'].json.message.content }}"
  }
}

1. Test the full flow: send a WhatsApp message to your number → within 5 seconds, the AI reply should arrive. If it does, your chatbot is working.

Step 4: Write a System Prompt That Actually Works

The system prompt is everything. A bad prompt makes your bot useless. Here is the structure that works for Indian SMB chatbots:

• Identity block: Who the bot is, what business it represents, city and language defaults
• Knowledge block: Your services, pricing ranges, what you do and do not do (paste this from your website)
• Collection goal: The exact 4–5 pieces of information to gather from every conversation
• Escalation rule: The exact trigger phrase that routes to a human (HUMAN, AGENT, HELP ME)
• Tone rule: Formal / casual / Hinglish — be explicit. GPT defaults to formal American English
• Boundary rules: What the bot should never say (pricing guarantees, competitor comparisons, false promises)

Step 5: Connect to Google Sheets + Notify Your Team

The bot is worthless if leads disappear into chat history. Log every completed lead to Google Sheets and alert your team.

1. Add an IF node after OpenAI → Condition: {{ $json.message.content }} contains "enquiry reference" (this fires when the bot has collected all lead info)
2. True branch → add a second OpenAI call with this prompt: "Extract the following from this conversation in JSON: name, phone, service_needed, budget, city, timeline. Return only valid JSON." Pass the full conversation as context.
3. Add a Google Sheets node → Append Row → log: Timestamp (IST), Name, Phone, Service, Budget, City, Timeline, Source: WhatsApp
4. Add a Gmail or Slack node → notify your sales rep: "New WhatsApp lead: [Name] — [Service] — [Budget] — [City]. Call within 2 hours."

Real Cost Breakdown for Indian Businesses

• Interakt Starter: ₹999/month — includes 1,000 free conversations, ₹0.50–₹1.20 per conversation after (Meta charges this, not Interakt)
• OpenAI API (gpt-4o-mini): ~₹0.50–₹2 per conversation depending on length
• n8n Cloud: ₹1,500/month or free if self-hosted on a ₹500/month VPS
• Total for 500 conversations/month: ~₹3,000–₹5,000
• A human support executive handling the same volume: ₹18,000–₹25,000/month plus management overhead

Mistakes That Get WhatsApp Numbers Banned

• Sending bulk broadcasts to people who never messaged you first — WhatsApp bans numbers for this. Only send proactive messages using pre-approved Message Templates.
• No opt-out mechanism — every broadcast must include a way to stop messages. Failure to do this leads to spam reports and number suspension.
• Responding after the 24-hour window with free-form text — after 24 hours of inactivity, you must use a Template Message (pre-approved by Meta). Free-form messages fail silently.
• Ignoring quality rating — Interakt and WATI both show your number's quality rating. If it drops to "Low", proactively pause outreach and fix the messaging quality.

Frequently Asked Questions

Can my bot speak Hindi or regional languages?

Yes. GPT-4o handles Hindi, Hinglish, Tamil, and Bengali well. Add this to your system prompt: "If the customer writes in Hindi, respond in Hindi. If they write in Hinglish (mixed Hindi-English), match their style." Test with actual Hindi messages before going live.

What is a Message Template and when do I need one?

A Message Template is a pre-approved message format required when you initiate a conversation (reach out first) or reply after the 24-hour window. Submit templates in Interakt under Templates → Create Template. Meta approves them in 1–2 hours. Always have 3–4 templates ready: appointment confirmation, payment reminder, follow-up after enquiry.

How do I handle a customer asking for a human agent?

Add this to your system prompt: "If the customer types HUMAN or asks to speak to a person, respond: 'Connecting you to our team now. A team member will reply within 10 minutes.'" Then in n8n, add an IF node that detects "HUMAN" in the message and sends a Slack/email alert to your team with the customer's phone number and conversation summary.