n8nOpenAIPinecone/QdrantWebsite widget

AI Customer Support FAQ Bot

A chat widget that answers 70% of customer questions from your own docs.

Difficulty: AdvancedTypical price: $800–$1,500

n8n nodes used

WebhookEmbeddingsVector StoreOpenAIHTTP Response

The problem it solves

60–70% of support tickets are the same handful of questions answered in the docs nobody reads. A retrieval bot deflects those and frees the team for real issues.

Tools you'll need

OpenAI Embeddings node — vectorizes docs + queries
Qdrant / Pinecone node — stores and searches vectors
Webhook node — receives chat messages from the site widget
OpenAI Chat node — answers grounded in retrieved chunks
Slack node — escalation when confidence is low

Build steps in n8n

Check each step as you complete it — your progress is saved automatically in this browser.

0 of 7 steps complete0%

1Build an ingestion workflow: read all help docs → split into chunks → OpenAI Embeddings → upsert to Qdrant.
2Schedule the ingestion to re-run nightly so doc updates flow through.
3Build a query workflow: Webhook receives {question, session_id}.
4Embed the question → Qdrant top-K search (k=5).
5OpenAI Chat node: system prompt 'answer only from context, otherwise say I'll connect you to a human', stuff retrieved chunks.
6If the model returns the escalation phrase: Slack-ping #support with the question and session.
7HTTP Response node returns the answer JSON to the widget on the site.

Premade n8n workflow

Skip the setup — download this workflow and import it into your n8n instance.

Download workflow.json

Import in n8n: Workflows → Import from file. Then add your own credentials to each node.

What to charge

$800 for a single-site bot on top of existing docs. $1,500 with multi-language, conversation memory, ticket creation on escalation, and an analytics dashboard.

How to pitch it

"Your customers get instant, accurate answers 24/7. Your team only sees the tickets that actually need a human. Most clients deflect 60–70% of L1 support."