Draft and policy-check Japanese e-commerce review replies with OpenAI

Quick overview

This workflow collects a Japanese e-commerce customer review via an n8n form, drafts a polite reply with OpenAI, checks and fixes it against your store policies, then logs the result to n8n Data Tables and displays the final reply with risk and change notes.

How it works

1. Receives product, rating, channel, and review text from an n8n Form trigger.
2. Loads all store policies from the n8n Data Table store_policies, combines them, and selects up to three relevant policies by matching policy keywords against the product name and review text.
3. Uses OpenAI (chat model) to draft a policy-based review reply in polite Japanese and labels it as praise, complaint, or question.
4. Uses OpenAI again to review the draft against the original review and the matched store policies, then fixes issues and assigns a risk level (low/medium/high).
5. Saves the review, final reply, category, risk, and a posting status (manager_review for high risk, otherwise ready_to_post) to the n8n Data Table review_replies.
6. Shows a formatted results page with the final reply, the fixes applied, the matched policy topics, and the saved-to-table confirmation.

Setup

1. Add an OpenAI credential and select a chat model in the OpenAI Chat Model node (for example, the configured gpt-5-mini).
2. Create two n8n Data Tables named store_policies (topic, keywords, policy) and review_replies (product, channel, rating, review, category, reply, risk, status, created_at).
3. Populate store_policies with your real rules (optionally run the “Load Sample Policies” branch once, then edit the inserted rows to avoid duplicates).
4. Publish the n8n form and use the generated form URL where your team submits reviews for drafting.