Every question looks the same in the feed
A billing dispute, a feature request, a how-to question, and spam all arrive as identical messages. Without classification, the team either reads everything or misses the urgent ones. Both fail at scale.
How to Build a Discord Support Bot using AI Agent
A production-ready workflow that routes every message through a 6-bucket triage, gates tool calls so only real tickets hit Linear or Notion, and resolves 70% of questions instantly — so your team wakes up to a clean queue instead of a 3 a.m. backlog.
01The Problem
Most Discord support breaks at scale
Communities grow faster than support teams. The inbox fills with questions that already have answers, tickets that belong in Linear, and 3 a.m. pings no human should answer.
Answers live in docs, not in Discord
The knowledge base has the answer. The FAQ channel has the answer. The pinned message has the answer. The user asks anyway because searching is slower than pinging @support. The bot must bring the answer to the conversation, not expect the user to leave it.
Ticket creation is manual and inconsistent
When a real issue surfaces, someone copies the message, opens Linear or Notion, fills the fields, assigns a label, and pastes the link back. That takes minutes per ticket. At 50 messages a day, it becomes a part-time job. The agent must create the ticket, populate the fields, and post the link — automatically.
02The How-To
The Discord AI support bot prompt, step by step, step by step
Copy it once, paste it into Zaira, connect the Discord gateway and your ticket tracker, and the agent handles triage, answers, and escalation — end to end.
Automatic Discord support agent
Step 0: Find the support channels and the answers before you touch the queue
Access to @Discord. Act as the support agent for this server. Do not answer anything yet. First work out two things by reading the server rather than by asking me.
Support channels: find the ones that take questions, such as help, support, how-to-ask or troubleshooting. Say which you chose and what made each one look like a support channel.
Answer source: find where the answers already live - an FAQ or rules channel, a readme or docs channel, pinned messages, and the topic or pinned post in each support channel itself. List every source with its channel and roughly how many questions it answers. An answer you cannot point at is an answer you do not give.
Step 1: Put every new message in exactly one bucket
For each message since the last run - on the first run, since [WINDOW] - use exactly one bucket: answerable, bug report, feature request, account or billing, spam, or needs a human. Give the bucket, a confidence from 0 to 100, and one sentence of why. Anything you are genuinely torn between goes to needs a human. Do not guess your way past a coin flip.
Step 2: Answer only what the source actually supports
For answerable messages at or above your threshold, search the answer sources you found in Step 0 first. If a source contains the answer, reply in-thread with the answer and a link to where you found it, and keep it under four sentences. If no source contains it, say plainly that you do not have a documented answer yet, then continue to Step 3. Never assemble an answer from general knowledge - a confident wrong answer in a support thread costs more than an honest one.
Step 3: Open a ticket in the review channel
For bug reports, feature requests, billing, and every answerable question you could not answer, post one structured message to the review channel the rules nominate: the reporter, the original message linked, the bucket, the confidence, one line of context, and what is needed to resolve it. Then reply in-thread to the member pointing at that ticket so they know it was not swallowed.
Step 4: Escalate the ones a person must decide
For needs a human, ping the role the server's own rules name for this, in the review channel, with the message linked and your reasoning. If no role is named, post in the review channel and say plainly that no escalation role is defined. Do not attempt an answer.
Step 5: Report the run
Close with counts per bucket, every message you answered, every ticket opened, and every time you chose not to answer. The last list is the one I read.
Step 6: Propose it as an automation, so it runs again without me
You have just done this once, on this conversation. Now offer to repeat it. Call create_automation with:
name: three or four words for the task. frequency: every_15_min, because a support queue moves faster than a day. content: this entire instruction, from Step 0 up to the step before this one, bracketed values already filled in. run_at: the first scheduled run, with today's date and my timezone. timezone: mine.
Two things. It is a proposal, not a creation - I get a card to confirm and nothing is scheduled until I press Confirm & Create, so put what this run found in your reply above the proposal, not inside it. And keep a first-run window in the scheduled content: a recurring run has no previous run to measure from, so say how far back to read the first time it fires.
Discord gateway connection 03Why People're Using
Here's what it does once it's running
Three stages, one goal. Zaira runs the support flow continuously, so the only thing left for your team is the complex issues that actually need human judgment.
Six buckets, one per message
Answerable, bug, feature, billing, spam, needs a human. Each gets a confidence and a reason, so a wrong call reads as a visible misclassification rather than a silent wrong answer.
Answers cite a source or do not ship
The agent searches your docs first. If the answer is there it replies in-thread with the link. If not, it says it has no documented answer and opens a ticket instead of improvising.
Members learn what happened to their message
Every ticket gets an in-thread reply pointing at it. Nobody sends a support question into silence and concludes the server does not read them.
Are you ready?
Launch your automatic Discord support agent in seconds
All without you getting tasks done on manual, missing out the opportunity you want, or putting your hard work in vain.
Try it with Zaira nowReady to runImmediate startNo code required
04FAQ
Frequently asked questions
The practical questions people ask before setting this up.
Support teams running Discord communities where volume exceeds what humans can triage manually — open-source projects, SaaS companies, gaming communities, DAOs. If you have a knowledge base and a ticket tracker, this replaces the manual bridge between them.
It separates answerable questions (docs exist) from bug reports (reproducible, needs code), feature requests (product input), billing issues (account changes), spam (noise), and ambiguous cases (needs human judgment). The six buckets are adapted for Discord's real-time, threaded format.
Every action is tied to the classification. The agent won't search your knowledge base unless the message is confidently answerable, and won't open a Linear issue unless it's a bug report, feature request, or billing change. Everything else escalates to a human. The flow stays auditable and you don't get hallucinated tickets or answers.
Yes. The prompt prefers Linear (structured issues, labels, assignees) and falls back to Notion (database pages) if Linear is unavailable. You configure the target in the Environment block. Both get the same payload: title, description, labels, and a backlink to the Discord thread.
Two safety nets: (1) The auto-resolve threshold defaults to 85% confidence — below that, it escalates to human. (2) Weekly review of auto-resolved conversations catches false positives and feeds corrections back into the knowledge base. The threshold is tunable per channel.
The 6 buckets are fixed, but the routing rules (which bucket → which Linear team, which Notion database, which escalation role) are configured in the Environment block. The weekly review loop is where you adjust: if 'bug_report' keeps landing in 'needs_human', add reproduction-step keywords to the classification prompt.