Answer Gmail Support Emails with Gemini & Slack in n8n

Answer Gmail support emails with Gemini—backed by your docs—and escalate gaps to Slack

This n8n workflow watches your Gmail Support label, retrieves the most relevant passages from your documentation (indexed via Google Gemini embeddings), and drafts a grounded Gmail reply. If the docs don’t cover the question, it sends the missing context to Slack and logs it for documentation follow-up.

What this workflow does

• Indexes your documentation by running manually to fetch a docs sitemap (supports both sitemapindex and flat sitemap formats), collecting eligible page URLs.
• Builds a searchable passage library: downloads each page, extracts the main article content, splits it into overlapping text passages, and generates Google Generative Language (Gemini) embeddings for each passage.
• Stores and refreshes the index in an n8n Data Table: normalizes vectors and upserts passage records (text, URL, title, vector, and the run ID), then deletes older rows from previous indexing runs.
• Monitors Gmail every few minutes for new unread emails with the specified Support label, extracts the new question (removing quoted history), and embeds the question with Gemini.
• Finds top matching answers by loading the stored passage index from the Data Table, computing cosine similarity, and stopping early when the best match falls below a configurable minimum score.
• Drafts grounded replies and audits them using Gemini from the retrieved passages only, then runs a second Gemini pass to audit claims against the same passages.
• Responds or escalates: if the audit passes, it creates a Gmail draft reply in the original thread; if coverage is insufficient, it escalates to Slack and logs the gap for documentation follow-up.

Use cases

• SaaS teams reducing support response time by turning documentation into draft Gmail replies.
• Automation engineers maintaining a Gemini + n8n knowledge index that stays current via sitemap re-indexing.
• Operators routing out-of-coverage questions to Slack so docs can be updated quickly.

Technical details

• n8n nodes: if, set, code, gmail, merge, slack
• Core components: Gmail label trigger, Data Table passage index, Gemini embeddings, cosine similarity scoring, grounded JSON reply + audit