The AI Review Collection & Response System
A two-part n8n system that checks in with clients after every appointment, only asks for a review once they say yes, and drafts a personalized reply to every new Google review from a real knowledge base instead of a generic template.
Section One
The Problem
Most businesses never ask for reviews at the moment it would work best: right after a good experience, while it's still fresh. When they do ask, it's a generic blast to everyone, happy or not, which either gets ignored or occasionally embarrasses the business when an unhappy client gets asked to leave five stars.
On the other side, incoming Google reviews sit unanswered for weeks. Not because the business doesn't care, but because a good reply takes real thought: who was this client, what did they experience, what tone fits a 2-star complaint versus a 5-star rave. That's a research task, not a five-minute task, so it keeps getting pushed.
The pattern that fixes both
This system fixes both directions with the same underlying pattern: gather context first, then let an AI agent do the writing, with a human still deciding what actually gets said in public.
Section Two
How the System Works
Two connected n8n workflows, each triggered independently, sharing one Airtable base as the source of truth.
Fig. 1 — system flow
Trigger
Calendly
Booking captured, Airtable record created
Wait
Appointment complete
SMS + WhatsApp check-in sent
Collect
AI agent
Reads sentiment, asks permission, shares review + draft
Respond
AI agent
Drafts personalized replies to new Google reviews
Section Three
Part One: The Review Collection Agent
01
Step One
Capture the booking
A Calendly booking fires a webhook. The workflow pulls the client's name, email, phone, and any pre-visit notes, creates a contact in the business's CRM, and writes a record to Airtable with Appointment Complete set to false.
02
Step Two
Wait for the visit to actually happen
The business ticks a single checkbox in Airtable when the appointment is done. That's the only human input this half of the system needs. Everything downstream is triggered by that one field changing.
Why this matters
The system can't know an appointment happened unless something tells it. A calendar event firing isn't proof the client showed up. One checkbox, ticked by whoever's at the front desk, is the cleanest signal available.
03
Step Three
Send the check-in
A few hours after checkout, the workflow sends a short, warm check-in on both SMS and WhatsApp: "How was your visit today?" Nothing salesy, nothing asking for a review yet.
04
Step Four
Read the reply, gauge sentiment first
Every reply routes to an AI agent with full conversation memory. Its first and only job at this stage is judging sentiment: positive, neutral, or negative.
Why this matters
Asking for a review before you know how the visit went is how businesses end up with a 1-star review they asked for themselves. Sentiment-gating is the whole point of using an agent here instead of a static "please review us" text blast.
05
Step Five
If positive, ask two light questions, then ask permission
The agent asks at most two follow-up questions in one message, with an emoji, never a survey. Who took care of them, what stood out, what they'd tell a friend. Once it has an answer, it asks permission to leave a review. It does not assume yes.
06
Step Six
Draft the review, don't write it for them
Only after the client says yes does the agent share the review link, along with a short, SEO-friendly draft review built from the client's own words: treatment name, staff member, what they liked, a recommendation. It's offered as an edit-or-copy option, never a finished script.
07
Step Seven
If neutral or negative, no review ask
The agent thanks them, acknowledges specifically what they said, and asks if they'd like a team member to follow up. It never requests a review here, and it flags the conversation for a human.
08
Step Eight
Log everything, escalate what it can't handle
After every reply, the agent writes sentiment, a feedback summary, whether the review link was shared, and next steps back to Airtable. If a conversation turns into something it shouldn't handle alone (angry, high-risk, out of scope), it emails a human with a summary instead of guessing.
Where this lands
Every positive reply either ends in a shared review link or a logged reason it didn't happen yet. Nothing falls through silently, and nothing gets asked twice.
Section Four
Part Two: The Review Response Agent
01
Step One
Build the knowledge base once
Upload service menus, staff bios, and policies through a simple form. They get chunked and embedded into a Pinecone vector store. This is a one-time setup per business, updated whenever something changes.
02
Step Two
Catch every new review automatically
A Google Business Profile trigger polls for new reviews and fires the agent the moment one lands, no manual checking required.
03
Step Three
Draft with real details, not guesses
The agent pulls from three sources before writing a single word: the knowledge base (to confirm a real staff name or treatment), past client emails (if the reviewer is a known contact), and the business's own website. It never invents a detail it can't confirm.
Why this matters
A generic "thank you for your kind words!" reply is worse than no reply. A reply that gets a staff member's name wrong is worse than that. Grounding the draft in real data is what makes the reply feel like it came from someone who actually knows the business.
04
Step Four
Match tone to star rating
Five stars gets genuine enthusiasm and an invitation back. Three to four stars acknowledges the good and stays open to feedback. One to two stars empathizes once, addresses the specific issue, and moves the conversation to a private channel, never defensive.
05
Step Five
Log the draft, don't auto-post
Every drafted reply lands in a spreadsheet next to the original review and its star rating, marked not yet replied. A human still reviews and publishes it. This system drafts; it doesn't publish on its own.
Section Five
What You Need to Launch This
| Piece | Purpose | Status |
|---|---|---|
| Airtable base | Client records, sentiment log, do-not-contact list | Create before going live |
| Calendly | Appointment booking trigger | Connect account |
| GoHighLevel (or equivalent CRM) | Contact creation, SMS sending | Connect webhook |
| Unipile (or equivalent WhatsApp API) | WhatsApp check-in and reply | Connect account |
| OpenAI + Anthropic | The two language models behind both agents (fallback pair) | Connect API keys |
| Gmail | Escalation emails, past-client email lookup | Connect account |
| Pinecone | Knowledge base for the review response agent | Create index |
| Google Business Profile | New review detection | Connect account |
| Google Sheets | Review reply draft log | Create sheet |
This isn't a review-request bot. It's a system that waits for the right moment, asks permission before it acts, and never lets an AI publish something in public without a human's eyes on it first. That's the difference between an automation that helps and one that quietly makes a mess.
“This isn't a review-request bot. It's a system that waits for the right moment, asks permission before it acts, and never lets an AI publish something in public without a human's eyes on it first.”
— The AI Review Collection & Response System, p. 7
§ Fin.
Rather have it built for you?
This guide is one system from the Soulwire Studio workflow library. We build AI automations for businesses that run on process: intake, scheduling, proposals, compliance-heavy operations. If you'd rather spend your time closing deals than debugging automation, book an audit and we'll find the automation hiding in your business.
