How to Automate Customer Review Responses with n8n

A three-star review lands at 11 PM. By morning, the reply window has narrowed and the customer has already told two colleagues about the experience. n8n review response automation changes that dynamic: the workflow fires when a review arrives, scores its sentiment, drafts a reply, and pages your team if the rating demands immediate action.

Most businesses handle reviews manually. Someone checks the inbox, decides whether to respond, drafts something, routes it for approval, and posts it. That's four steps, two or three people, and usually 48 hours between the review and the reply. The n8n workflow cuts that to one human decision: approve the pre-drafted, pre-routed response.

Running on self-hosted n8n, the cost is GPT-4o-mini API spend — roughly $0.001 per review classified and drafted. At 300 reviews per month, that's 30 cents in AI costs. No per-seat fees, no review management platform subscription.

What n8n Can Automate in Your Review Workflow

The workflow handles more than just drafting. Here's what you can automate end-to-end:

• Classify every incoming review as positive, neutral, or negative with an urgency score (1–10)
• Generate a professional draft reply tailored to the specific review text and platform tone
• Route negative reviews (urgencyScore >= 6) to a Slack alert in a dedicated channel
• Email the draft and full review context to the business owner for approval before posting
• Log every review, its sentiment, key themes, and draft reply to Google Sheets for trend tracking
• Escalate repeat negatives from the same customer to a manager DM after two incidents in 30 days

The Review Response Pipeline

The full flow from review arrival to approved draft:

Webhook → AI Classifier → Code (JSON parse) → Switch (Sentiment) → Draft Generator → Slack Alert + Email → Google Sheets Log

Every step is auditable in the n8n execution log. If a draft comes out wrong, you can replay the execution with modified inputs — no re-submitting the review through an external platform.

Building the Review Automation Step by Step

1. Ingest the review via webhook

The pipeline starts with a Webhook node set to POST mode. Any platform that can send an HTTP request feeds it: Trustpilot's outbound webhook, a bridge from Google My Business, or a custom scraper. The payload needs at minimum: reviewText, platform, rating, and customerName.

The Webhook node's default timeout is 120 seconds. For synchronous review submission flows that block on a response, that's not a constraint. If you're aggregating from a slow external API, move the classification logic to an async sub-workflow triggered from a second webhook call after the ingestion step resolves.

2. Classify sentiment with the AI node

A single call to the @n8n/n8n-nodes-langchain.openAi v2.1 node handles classification. The prompt requests JSON output with three fields: sentiment (positive/neutral/negative), urgencyScore (1–10), and keyThemes (string array). That response lands at $json.output[0].content[0].text — don't use the older n8n-nodes-base.openAi response path, which fails silently when copied from tutorials written before n8n's v1.0 upgrade.

Parse the JSON immediately after. Add a Code node (v2, jsCode parameter) with something like return [{ json: JSON.parse($input.first().json.output[0].content[0].text) }]. Skip this step and the downstream Switch node tries to evaluate a string. The execution log makes the failure obvious; the error message itself doesn't tell you where it breaks.

3. Generate the draft response

A second AI call takes the classified sentiment and the original review text, then returns a draft reply. The prompt should instruct the model to acknowledge the specific detail mentioned in the review, match the platform's expected tone (warmer for Yelp, more formal for G2 or Capterra), and stay under 150 words. Drafts longer than 150 words rarely get posted verbatim — they force the approver to edit, which defeats the point.

It's worth setting the model to gpt-4o-mini here. GPT-4o at ten times the cost doesn't produce meaningfully better short review replies. The classification call is where nuance matters more; the draft is usually straightforward once sentiment is known.

4. Route negatives to Slack, positives to email

A Switch node routes on the sentiment field. Negative reviews (urgencyScore >= 6) go to a Slack node that posts in a #review-alerts channel with the customer name, star rating, urgency score, a one-sentence summary, and the draft reply. Neutral and positive reviews skip Slack entirely.

The Slack message should include everything the on-call person needs without opening the original review: customer name, rating, key themes, and the draft. A bare "new negative review" ping with a link forces someone to context-switch twice. That's the kind of friction that kills response time.

5. Log every review to Google Sheets

The final step in both branches writes a row: timestamp, platform, rating, sentiment, urgencyScore, keyThemes, draftResponse, and a status column that starts as pending. When the business owner approves and posts the reply, they update the row to posted — which can trigger a separate workflow to mark the review handled in your CRM.

There's no reason to skip the logging step even if you don't analyze the data immediately. Six months of review sentiment data per platform is genuinely useful for spotting product issues early.

Implementation Patterns Worth Knowing

Pattern 1: Scheduled polling for platforms without webhooks

Not every review platform offers outbound webhooks. Google My Business requires polling via the Business Profile API. Set a Schedule trigger (n8n-nodes-base.scheduleTrigger v1.2) to run every 30 minutes. The workflow fetches reviews published since a last-checked timestamp stored in a dedicated Google Sheet cell updated after each run.

Google's Business Profile API allows 10 requests per second — won't get close at 30-minute polling intervals. Trustpilot's Business API caps at 60 requests per minute. In practice, for most review volumes, you'll never hit either limit.

Pattern 2: Multi-platform tone switching

Different platforms expect different reply styles. Google skews formal, Yelp runs warmer, G2 reviewers expect product-specific detail. Add a second Switch node after classification that routes on the platform field. Each branch hits a slightly different AI prompt with platform-appropriate instructions. The approver sees the same draft queue — the tone difference is invisible to them but meaningful to the customer reading the public reply.

Pattern 3: Escalation on repeat negatives

If the same customer submits two or more negative reviews within 30 days, auto-escalate to a manager DM rather than the standard Slack channel. Store a running 30-day negative count per customer email in Google Sheets. A Code node checks the count before the Slack step and overrides the destination channel if it hits 2. This catches the customers who are genuinely unhappy and need a phone call, not just a polished reply.

And no, the deprecated Function node shouldn't still be in your escalation logic — use Code (v2) with the jsCode parameter.

n8n Nodes You'll Use Most

Getting Started with Review Automation

1. Deploy n8n (self-hosted on a $10/month VPS or n8n Cloud) and configure OpenAI, Slack, and Google Sheets credentials under Settings > Credentials.
2. Create a Google Sheet with columns: timestamp, platform, rating, reviewText, sentiment, urgencyScore, keyThemes, draft, status.
3. Import the workflow or build from scratch starting with a Webhook node in POST mode.
4. Set the classification prompt in the first OpenAI node. Include an explicit instruction to return JSON matching the three-field schema — without it, the model sometimes returns prose instead of structured output.
5. Test with a sample payload using n8n's manual execution. Use rating: 1 and urgencyScore: 9 to confirm the Slack branch fires.
6. Enable the live Webhook trigger, paste the URL into your review platform's outbound settings, and post a test review.
7. Once you're past the test phase, the Review Response Engine ships pre-wired: credentials, channel, and Google Sheets ID are the only config you touch.

For related patterns, the post on n8n customer support automation covers ticket routing and SLA alert workflows that pair naturally with a review pipeline. If you're adding AI classification across multiple parts of your stack, n8n AI-powered automation workflows covers structuring multi-step LLM calls and parsing their output reliably.

Review automation isn't complicated. It's four nodes doing a job that currently takes four people 48 hours. Start with the webhook trigger and the classification call. Get those working. Then add the routing and logging branches one at a time.