How to Set Up Auto-Reply Google & TripAdvisor Reviews — GPT-4o Writes Personalized Responses That Reference Specific Details in n8n
If you run a restaurant, hotel, clinic, or any local business, reviews are your storefront. Google and TripAdvisor ratings decide whether a stranger books you or scrolls past. But responding to every
If you run a restaurant, hotel, clinic, or any local business, reviews are your storefront. Google and TripAdvisor ratings decide whether a stranger books you or scrolls past. But responding to every review — thoughtfully, quickly, without sounding like a robot — is a job nobody has time for. Most owners either ignore reviews entirely or paste the same "Thank you for your feedback!" under every one. Both hurt you. This article shows you how to build an n8n workflow where GPT-4o reads each review, understands what the customer actually said, and writes a personalized reply that references specific details — automatically.
The Problem: Generic Replies Are Worse Than No Reply
Review platforms reward engagement. Google's local ranking factors explicitly favor businesses that respond to reviews, and prospective customers read your responses as closely as they read the reviews themselves. A reply that says "Thanks for the 5 stars!" under a review that praised your sommelier by name tells everyone you didn't actually read it. A copy-paste apology under a detailed complaint tells the angry customer — and every future reader — that you don't care.
The math gets worse at volume. A business pulling 30–50 reviews a month across Google and TripAdvisor is looking at hours of writing per week to do it properly. So it doesn't get done properly. Replies pile up, negatives sit unanswered for weeks (the exact reviews that need fast, human attention), and the ones that do get answered are indistinguishable boilerplate. The bottleneck isn't caring — it's the time cost of writing something specific, every single time.
What you actually need is a reply for every review that (1) mentions something concrete from that specific review, (2) matches the emotional register — warm for praise, empathetic and solution-oriented for complaints — and (3) goes out within hours, not weeks. That's exactly the kind of judgment task GPT-4o is good at, and exactly the kind of orchestration n8n is built for.
The Solution: GPT-4o as Your Review Reader, n8n as the Plumbing
The architecture is simple. n8n polls for new reviews on a schedule, passes each one to GPT-4o with a prompt that instructs it to reference a specific detail and adapt tone to sentiment, and then either publishes the reply automatically or routes it to Slack for a one-click human approval first. GPT-4o doesn't fill in a template — it reads the review the way a thoughtful owner would and writes an original response.
The difference is in the output. For a 5-star review that says "The tasting menu was incredible, especially the scallop course," GPT-4o writes back thanking them and naming the scallop course specifically. For a 2-star review complaining about a 40-minute wait for a table they'd booked, it acknowledges the wait, apologizes without excuses, and offers a concrete next step — rather than a defensive wall of text. Same workflow, radically different replies, because the model is reading, not matching.
Step-by-Step: Building It in n8n
Here's the node-by-node build. The whole thing runs on a schedule and needs no server code.
1. Schedule Trigger. Start with a Schedule Trigger node set to run every few hours (a cron of 0 */4 * * * is a sensible starting point). Frequent enough that negatives get a fast reply, not so frequent you hammer the APIs.
2. Fetch reviews. Use an HTTP Request node for each source. For Google, call the Google Business Profile API endpoint accounts/{account}/locations/{location}/reviews with OAuth2 credentials configured in n8n's credential store. For TripAdvisor, use the Content API reviews endpoint with your API key. Set both to return only reviews since the last run. If you run both sources, follow them with a Merge node in "append" mode so everything flows down one pipe.
3. Filter out what's already handled. Add an If node (or a Filter node) that drops any review that already has a reply, or store processed review IDs. The cleanest approach is an n8n Data Table (or an external store like Postgres/Airtable via node) holding every review_id you've answered — look it up and skip matches. This is your idempotency guard; without it you'll re-reply on every run.
4. Split into individual items. Use a Loop Over Items (Split in Batches) node so each review is processed one at a time. This keeps GPT-4o focused on a single review per call and makes rate-limiting trivial.
5. Generate the reply with GPT-4o. Drop in the OpenAI node (or a generic HTTP Request to the Chat Completions endpoint), model gpt-4o. Your system prompt is where the magic lives. Something like:
"You are the owner of {{business_name}} replying to a customer review. Write a short, warm, personal reply in the reviewer's language. You MUST reference at least one specific detail they mentioned. If the review is positive, be warm and add a small invitation to return. If it's negative, be genuinely empathetic, apologize without excuses, and offer a concrete way to make it right — never defensive, never generic. Sign off as the owner. Max 80 words."
Pass the review text, star rating, and reviewer name in the user message via expressions like {{ $json.comment }}. Set temperature around 0.7 for natural variation and cap max tokens near 200. To make sentiment routing explicit, ask the model to return JSON with sentiment and reply fields, then parse it.
6. Optional Slack approval. This is the safety valve. Add a Slack node that posts the drafted reply to a channel and, using Send and Wait for Response, pauses the workflow with Approve/Edit buttons. Nothing publishes until someone clicks. For negative reviews especially, this human-in-the-loop step is worth it — you can branch with an If node so 4–5 star replies auto-publish and 1–3 star replies always route to Slack first.
7. Publish the reply. A final HTTP Request node writes the response back — for Google, a PUT to the review's /reply endpoint; for TripAdvisor, the management response endpoint. Then write the review_id to your processed-IDs store so it never gets touched again.
Why This Pays for Itself Fast
The immediate win is time: what took hours a week now takes zero, and the replies are better than what you were writing when rushed. But the compounding wins matter more.
Ranking and visibility. Google rewards response rate and speed. A business that replies to 100% of reviews within hours signals an active, engaged owner — and that feeds local search ranking.
Negative-review damage control. The single highest-leverage moment in reputation management is a fast, human reply to a bad review. It shows every future reader you handle problems gracefully, and it often turns an angry reviewer into someone who edits their star rating up. This workflow guarantees no negative sits unanswered.
Consistency of voice. Every reply sounds like you, in your language, at your quality bar — even at 3 a.m. when a review lands and even when you're slammed during service.
Common Pitfalls (and How to Avoid Them)
Skipping the idempotency guard. The number-one failure mode: forgetting the processed-IDs check and re-replying to the same review every run. Build the Data Table lookup before you build anything else, and test it.
Auto-publishing negatives on day one. Don't. Route every 1–3 star reply through Slack approval for the first few weeks until you trust the tone. AI empathy is good, but a legal or safety-sensitive complaint deserves human eyes.
A prompt that's too loose. If you don't explicitly demand a specific detail and cap the length, GPT-4o drifts toward generic, over-long replies. The "you MUST reference one specific detail" instruction and the word cap are what make output feel human. Test with a batch of real reviews before going live.
Ignoring language. International reviewers write in their own language. Instruct GPT-4o to reply in the reviewer's language — a mismatched-language reply is worse than boilerplate.
API rate limits and auth expiry. Google OAuth tokens expire; set up n8n's OAuth2 credential to auto-refresh, and add an error branch that alerts you if a fetch fails rather than silently doing nothing. Wrap publish calls so a single failure doesn't halt the whole batch.
Start with approval-mode on for everything, watch the drafts for a week, then loosen the leash for positive reviews once you trust it. Within a month you'll have a review response system that's faster, more personal, and more consistent than anything you could do by hand — running entirely on autopilot.
Ja construimos isso pra voce
Nao comece do zero. O Auto-Reply Google & TripAdvisor Reviews — GPT-4o Writes Personalized Responses That Reference Specific Details e um workflow n8n pronto para instalar — conecta suas ferramentas em minutos, sem codigo.
Instalar por $49 →
