One inbox, never two replies
A workflow answers a customer, then a colleague answers the same customer five minutes later, and two different answers land in one inbox. That's worse than no automation. Below is the workflow I built: it checks the whole thread for a human reply before it does anything else, and it only ever answers from a knowledge base you control.
The question
this build answers
How do I auto-answer repeat support questions on Gmail without a bot talking over a colleague who already replied?
From the first trigger
to the finished job.
The working version, step by step.
Check for a human first
Before anything else, the workflow fetches the whole thread and checks it for a message from anyone but the customer. A person already replying stops the workflow right there and logs it, no exceptions.
Classify at temperature 0
If the thread is still quiet, I have a model call read the message at temperature 0 and return a label for the kind of question it is, plus the customer's actual question in one plain sentence. A failed or empty call is marked unscored and goes straight to a person.
Answer only from the knowledge base
A known question only gets answered when it scores above your confidence threshold against your own support_kb table. The workflow never lets the model answer from its own memory, and every reply quotes the exact entry it used.
Draft first, send only if you've turned it on
Every reply becomes a Gmail draft before anything else. Sending automatically stays switched off by default, behind one setting, and a failed draft or a failed send gets escalated the same way a low-confidence match does.
Keep urgent messages away from auto-reply entirely
Anything flagged urgent gets a louder Slack alert with no path back to the draft or send nodes. An urgent message never gets an automatic reply from this workflow, whatever the model decided.
Going through n8n's review
Answer repeat support questions and hand off to a human without losing the thread
This one is going through n8n's review. My creator page lists every template of mine that's already live, so you can see what else is available while this one clears.
See my templates on n8nWho this is for
Anyone running a small support inbox on Gmail who gets the same five or six questions on repeat, and wants those handled without a bot talking over a real reply.
Why urgent never auto-replies
Air Canada was held liable for a fare its own chatbot invented (Moffatt v Air Canada, 2024 BCCRT 149). I built no path from the urgent branch back to the draft or send nodes, on purpose, so that specific failure mode isn't available to this workflow.
What it costs to run
I count one model call per new email, at temperature 0 for a consistent read, on top of whatever your Gmail and Slack accounts already cost. The knowledge base is a plain n8n Data Table, so there's no extra service running alongside it.
Before you spend anything
The fit matters.
Who should not build this
If your support inbox gets genuinely varied questions rather than the same five or six on repeat, there's no knowledge base worth building yet, and this is overhead for a problem you don't have.
If you want every reply sent automatically from day one, know that the default here is a draft, and turning that switch on before you've read a week of drafts skips the whole point of shipping it off.
Yours to keep
What lands
on your desk.
- The workflow set up in your own n8n, watching your own Gmail inbox
- Three Data Tables for the knowledge base, the thread state and the log, columned as the setup note specifies
- Your confidence threshold and urgent keyword list written into one Settings node
- Every escalation landing in Slack with the question, the reason and a suggested reply
See the work
Open the proof.
Published case studies of running systems. Each links to the thing itself.
A few useful
answers.
What stops it answering over a colleague?
It reads the whole thread before doing anything else, and a message from anyone but the customer stops the workflow right there and logs it. No exceptions.
Can it answer something that isn't in the knowledge base?
A question below your confidence threshold is treated as not a known FAQ, so it goes straight to a person to answer.
Does it send replies automatically?
Every reply is a Gmail draft first by default, and sending automatically stays switched off behind one setting until you turn it on.
What happens to an urgent message?
It gets a louder Slack alert and never an automatic reply. There's no path from that branch back to the draft or send nodes.
A little further down
the same path.
Related builds for the rest of your business.
Make this work for you
Your tools.
Your working version.
Want this built and handed over working? Tell me what your version needs to do. One line is enough to start.
Start a brief