Step-by-step guide to building an AI-powered customer support bot using n8n, ChatGPT, and your knowledge base.
customer support is expensive. you hire agents, train them on the same 40 questions, and watch them answer "where's my order?" every single day. response times creep up, customers churn, and your best people burn out on repetitive tickets.
an AI customer support bot built on n8n handles the boring 70-80% automatically, escalates the rest to humans, and pulls answers from your actual data — not generic chatgpt waffle. here's exactly how to build one.
the bot sits between your customers and your team. when a ticket arrives — via email, chat widget, whatsapp, or a form — n8n catches it, searches your knowledge base for relevant context, asks chatgpt to draft a reply, and either sends it or queues it for a human to review.
the difference from a basic chatbot: it uses your actual data. your faq docs, product pages, past resolved tickets, internal runbooks. not random internet knowledge. the answers are accurate, on-brand, and specific to your business.
we covered the building blocks for this in our AI agent workflows guide — the support bot is one of the most practical use cases.
four components, all connected through n8n:
this approach handles questions it's never seen before, as long as the answer exists somewhere in your knowledge base. for more on how small businesses are using AI agents, check our breakdown of AI agents that actually work.
start with the trigger. easiest entry point is a webhook — most chat platforms (tawk.to, crisp, intercom) can post to a webhook url. add a webhook node in n8n and paste the url into your chat platform's settings.
add a set node to normalise incoming data. extract the message, customer email/user id, and channel. keeps the rest of your workflow clean regardless of source.
next: knowledge base lookup. for simple setups, use google sheets or notion to search your faq. for something more powerful, connect to a vector database via http request — embed the question, search for similar entries, return the top 3-5 matches as context.
the openai node takes over. your system prompt should include: company name and tone, instructions to only answer from provided context, a directive to say "i don't know" when context doesn't cover it, and any formatting rules.
after the ai generates a response, add an if node to check confidence. if the response contains phrases like "i'm not sure" or "i don't have that information," route to escalation. otherwise, send automatically via the original channel's api.
the bot is only as good as what you feed it:
update the knowledge base weekly. n8n can automate this — schedule a workflow that pulls new articles from your cms or helpdesk and adds them to the database. stale data means wrong answers.
the bot should never pretend to be human. make it clear the customer is talking to an ai, and give them a path to a real person at any point.
escalation triggers: customer explicitly asks for a human, the ai can't find a confident answer, the topic involves refunds or complaints (high-stakes), or the conversation goes back and forth more than 3 times without resolution.
when escalation fires, create a ticket with: the full conversation, the ai's draft response (so the agent doesn't start from scratch), customer context, and a priority level based on topic.
this is where n8n's flexibility shines compared to locked-in chatbot platforms. you decide exactly what triggers escalation and what the human sees. we break down these differences further in our n8n vs zapier comparison.
before going live, test with real tickets. take 50 recent support conversations, run them through the workflow, compare the ai's responses to what your agents actually sent. you'll find gaps fast — missing knowledge entries, unclear prompts, edge cases.
once live, monitor three things: resolution rate (what percentage the bot handles end-to-end), accuracy (spot-check 10 responses per day for the first month), and escalation rate (if it's above 40%, your knowledge base needs work).
add error handling to every node. if chatgpt is down, the knowledge base times out, or the response channel fails — n8n should catch it, log it, and route to a human automatically. silent failures in support mean ignored customers.
Need this implemented? we build these workflows for clients →