AI FAQ Auto-Responder for Incoming WhatsApp Messages
An n8n workflow that receives inbound WhatsApp messages through the Cloud API, answers common questions with Claude using your FAQ as context, and replies instantly — while quietly handing anything sensitive or unknown to a human agent, so customers get 24/7 answers without a bot embarrassing your brand.
- 1
Expose the WhatsApp webhook
Add a
Webhooknode (POST). In Meta for Developers, set the WhatsApp webhook callback URL to it and subscribe tomessages. Meta sends a GET verification first — return thehub.challenge(a second GET webhook handles this, per the notes). - 2
Extract the message
A
Codenode reads the payload atentry[0].changes[0].value.messages[0], pulling the sender's phone number and the text. It ignores status callbacks (delivered/read) so only real inbound messages continue. - 3
Draft a reply with Claude
An
HTTP Requestto Claude sends the message plus your FAQ and brand voice. Ask for strict JSON:replyandneeds_human. Instruct it to setneeds_humanfor refunds, complaints or anything off-FAQ, with a polite holding message as the reply. - 4
Branch: auto-send or escalate
An
IFnode onneeds_humanroutes the flow. False → send the AI reply. True → aSlacknode posts the conversation and customer number to your support channel so an agent takes over inside the 24-hour window. - 5
Send the WhatsApp reply
An
HTTP RequestPOSTs tohttps://graph.facebook.com/v19.0/YOUR_PHONE_NUMBER_ID/messageswith the recipient number and a text body, using your access token. The customer gets an instant, accurate answer any time of day.
Frequently asked questions
What do I need to use the WhatsApp Cloud API?
A Meta Business account, a WhatsApp Business phone number, and an app in Meta for Developers with the WhatsApp product added. You'll get a phone number ID and an access token, and you subscribe a webhook for incoming messages. It's free to receive and reply within the 24-hour customer service window.
How does it avoid answering things it shouldn't?
The Claude prompt includes your FAQ and a rule: if the question involves refunds, complaints, account changes or anything not in the FAQ, return a short holding reply and set `needs_human`. Those conversations route to Slack for an agent instead of auto-sending, so the bot never improvises on sensitive topics.
Can it reply outside the 24-hour window?
Free-form replies are only allowed within 24 hours of the customer's last message. Outside that you must use a pre-approved message template. This workflow handles the in-window case (the common one); for proactive or late replies, swap the send step for a template message.
Let Customers Check Order Status by Texting WhatsApp
An n8n workflow that lets a customer text their order number to your WhatsApp, looks the order up in Shopify (or a Google Sheet), and instantly replies with the current status and tracking link — deflecting the single most common support question without an agent lifting a finger.
Free recipe
Feedback Analyzer: every survey response tagged, scored, and trended — bad ones escalated live
An n8n workflow that catches every survey or feedback-form response via webhook, has Claude tag the themes and sentiment, appends it to a running analysis sheet, and pings Slack immediately when a detractor writes in — plus a monthly themes report nobody has to compile.
Free recipe
FAQ Auto-Responder: draft replies from your own answers, waiting in the thread before you open it
An n8n workflow that matches every incoming support question against your FAQ knowledge sheet, has Claude compose a reply using only your approved answers (with a strict 'say so if you don't know' rule), and saves it as a Gmail draft in the thread — you review, tweak, send.