Portfolio·Bespoke lane

Google Reviews
Auto-Response Pipeline

A fully automated review management workflow built for a local business. Polls Google for new reviews, routes by rating, generates AI responses, handles first-run initialisation automatically, and queues responses for human approval before anything goes live.

The problem

Responding to Google reviews promptly matters — Google's algorithm notices, and so do potential customers. Reputation management platforms exist, but for a small operation they bring unnecessary cost and complexity. The client needed something purpose-built: a pipeline that fits their workflow, not a subscription product designed for someone else's.

The additional constraint: Google provides no native auto-reply feature, so both review ingestion and reply publishing require direct integration with the Google Business Profile API. There is no button to press — the integration had to be built from scratch.

Approach

Diagnosed the workflow requirements, assessed available tooling, and designed an architecture that automates the routine while keeping a human in the loop for anything that needs judgement.

The key design decision was a three-tier rating system rather than a binary positive/negative split. Four and five star reviews get an AI-generated response queued for approval. Three star reviews — the ambiguous ones worth personal attention but not urgent — follow the same approval path. One and two star reviews skip AI entirely and go straight to a human alert. Nothing gets published without approval.

Architecture decisions

Polling, not webhooks

Google Business Profile doesn't push review events — there's no webhook to listen to. The workflow runs on a schedule, deduplicating against a Google Sheet. On first run, the sheet is seeded automatically with existing review IDs rather than processing them retroactively — no manual setup required.

Three-tier routing

Positive and borderline reviews are handled differently in tone but follow the same approval flow. Negative reviews always go to a human — an AI-generated response to a one-star review is a risk no client should take.

Approval before publish

Every AI-generated response goes through a Telegram approval queue. The owner sees the review, the proposed reply, and approves, edits, or rejects before anything is posted to Google.

Workflow
The complete n8n workflow with annotations. Click to zoom.
Tools used
n8n
Workflow orchestration
Google Business Profile API
Review ingestion and reply publishing
Claude API
AI response generation
Google Sheets
Deduplication and audit log
Telegram Bot API
Human approval queue
Deliverables
← Back to portfolioBespoke services →