n8n + n8n: Auto-Reply Google & TripAdvisor Reviews — GPT-4o Writes Personalized Responses That Reference Specific Details
Every unanswered review is a signal to future customers that you're not paying attention. Google's own data shows businesses that respond to reviews are perceived as 1.7x more trustworthy, and review
Every unanswered review is a signal to future customers that you're not paying attention. Google's own data shows businesses that respond to reviews are perceived as 1.7x more trustworthy, and review response rate is a ranking factor in local search. Yet for any business past a handful of locations, keeping up is brutal: reviews trickle in across Google Business Profile and TripAdvisor at all hours, in multiple languages, each one demanding a reply that actually reads like a human wrote it. The manual reality is that most reviews go unanswered, and the ones that do get answered get the same three canned lines everyone recognizes as filler.
The problem: generic replies are worse than no reply
The trap most teams fall into is templating. Someone writes five "response templates" — one for 5-star, one for 1-star, and so on — and pastes them in rotation. Customers spot this instantly. A reviewer who wrote three specific paragraphs about a cold room and a rude front-desk interaction does not want "Thank you for your feedback, we value your input." That reply tells them a machine (or a bored employee) skimmed a star rating and moved on. It converts a recoverable complaint into a permanent grudge, and it's public.
The opposite extreme — genuinely personalized, detail-referencing replies written by a human — works, but it doesn't scale. A person reading each review, recalling the property or product context, and drafting an empathetic solution-oriented response takes 5–10 minutes per review. At 40 reviews a week across locations, that's an entire part-time job whose output is invisible until it lapses. The job gets deprioritized, the backlog grows, and response rate craters exactly when volume (and therefore visibility) is highest.
What you actually need is the personalization of a human with the throughput of automation. That's the gap GPT-4o closes: it can read the full text of each review and generate a response that references the specific things the reviewer mentioned — the cold room, the front-desk name, the dish they loved — without a human in the drafting loop.
The solution: GPT-4o reads each review, not just its rating
This workflow treats every review as unique input. GPT-4o receives the full review body, the star rating, the reviewer's name, and business context, then writes a response tuned to sentiment and content. Positive reviews get warm replies that reinforce what the customer loved and gently add value — a mention of a feature they might have missed, an invitation to return. Negative reviews get empathetic, solution-oriented responses: acknowledge the specific failure, take ownership, offer a concrete next step and an off-platform contact so the conversation moves out of public view.
The key is prompt design. You instruct the model to quote or paraphrase at least one specific detail from the review, to never use generic filler, to match the business's tone, and to stay under a character limit that fits Google's and TripAdvisor's display constraints. Because GPT-4o handles multilingual input natively, a French review gets a French reply and an English review gets English — no separate branch needed. An optional Slack approval step lets a manager green-light replies to sensitive 1- and 2-star reviews before they publish, while 4- and 5-star replies post automatically.
Step-by-step: building it in n8n
The workflow has five logical stages. Here's how each maps to n8n nodes.
1. Trigger and fetch new reviews. Use a Schedule Trigger node set to run every 15–30 minutes. Follow it with two HTTP Request nodes: one hitting the Google Business Profile API (accounts.locations.reviews endpoint) and one hitting the TripAdvisor Content API. Store the timestamp of the last processed review in a Set node or an external store so each run only pulls reviews newer than the last cursor — this is what keeps you from re-replying to old reviews.
2. Normalize and filter. Merge both sources with a Merge node, then use a Code node to normalize the two API shapes into one common object: { id, source, rating, text, author, language, created }. Add an IF node to skip reviews that have no text body (star-only ratings) or that you've already answered — check a reply field or your own processed-IDs list.
3. Generate the response with GPT-4o. Add the OpenAI node (Chat model, gpt-4o). In the system prompt, hardcode the business persona, tone rules, and hard constraints: "Reference at least one specific detail the reviewer mentioned. Never use generic phrases like 'we value your feedback.' For ratings ≤3, acknowledge the problem, apologize sincerely, offer a concrete resolution, and provide this contact: support@business.com. Keep under 700 characters. Reply in the same language as the review." Pass the normalized review fields into the user message. Set temperature to around 0.7 — high enough to avoid robotic sameness, low enough to stay on-brand.
4. Optional Slack approval. Add an IF node branching on rating. For ratings ≤3, route to a Slack node that posts the review and the drafted reply to a review-ops channel, then use n8n's Wait node in "on webhook call" mode so a manager can approve or edit before the flow continues. High ratings skip straight to publishing. This gives you human oversight exactly where reputation risk is highest, without bottlenecking the routine positive replies.
5. Publish the reply. Route back to two HTTP Request nodes that PUT the approved text to the correct platform — the Google reviews/reply endpoint or the TripAdvisor management endpoint — keyed off the source field. Finally, write the review ID and the published reply to a Google Sheets or database node so you have an audit log and a dedup source for the next run. Wrap the publish nodes with n8n's error handling (an Error Trigger workflow) so an API failure alerts you instead of silently dropping a reply.
Why this pays off
The immediate win is response rate. A workflow running every 30 minutes means no review sits unanswered for more than half an hour, including overnight and weekends — the windows where manual processes always fail. That consistency compounds into local SEO: Google rewards active, responsive profiles, and fresh replies keep your listings visibly maintained.
The second win is quality at scale. Because every reply references specific details, prospective customers reading your reviews see a business that genuinely listens. Negative reviews, handled with a fast empathetic response and an off-platform resolution path, often get updated or amended by the original reviewer — turning a liability into social proof. And the time cost collapses: a task that consumed hours per week drops to occasional Slack approvals for the handful of sensitive cases.
There's also a defensibility angle. Every reply is logged with its review ID, timestamp, and text. When a franchise owner or a legal team asks "how did we respond to this complaint," you have a complete, timestamped record instead of scattered memory.
Common pitfalls to avoid
Skipping the dedup cursor. The single most common failure is re-processing reviews on every run because you didn't persist the last-seen timestamp or an answered-IDs list. This spams reviewers with duplicate replies and burns API quota. Build the dedup check before you build anything else.
Letting the model hallucinate specifics. If your prompt pushes too hard for "specific details," GPT-4o can invent things the reviewer never said. Constrain it: instruct the model to only reference details explicitly present in the review text, and to stay general if the review is vague. Test with real edge cases — one-word reviews, sarcasm, mixed sentiment.
Auto-publishing negative replies. Resist the urge to fully automate 1-star responses. A legally sensitive complaint, an allegation, or a viral-risk review needs a human glance. The Slack + Wait approval branch costs you nothing on volume and saves you from a public misfire.
Ignoring rate limits and character caps. Both Google and TripAdvisor enforce reply length limits and API quotas. Cap GPT-4o output in the prompt, and add a small delay or batching if you process large backlogs so you don't trip rate limits mid-run.
Set-and-forget tone drift. Review your published replies weekly for the first month. Tone rules that read well in the prompt can produce subtly off-brand output at scale. Adjust the system prompt as you see real examples — treat it as a living config, not a one-time setup.
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 →
