How to build a Discord moderation team

Moderation teams fail in the same three ways, and none of them is a shortage of capable people. They fail on inconsistent standards, on permissions granted too broadly, and on having no route for a decision nobody wants to make alone.

Before you start: you need Manage Roles. Allow an hour, and do the writing before the recruiting.

Recruit from behaviour, not applications

The best predictor of a good moderator is somebody already doing part of the job without the title.

Watch for the members who answer questions from newcomers, who de-escalate an argument by changing the subject, and who report rather than engage when something is wrong. All three are the actual work, performed voluntarily, in public, where you can see it.

Applications are worth running as a second step rather than a first — they tell you who wants the role, which is a different question. If you use one, ask for judgement rather than intent: describe a real situation from your server and ask what they would do. “Why do you want to be a moderator” produces answers that are indistinguishable from each other.

Two people is the minimum that works. One moderator is a single point of failure and has nobody to check a decision with; the pair is what turns a judgement call into a conversation.

Grant permissions by role, never individually

Create a Moderator role and grant it there. Individual member-level grants are invisible in the role list, survive the person leaving, and are the reason a server discovers two years later that somebody it forgot about can still ban.

Do not grant Administrator. It overrides every channel override you have set — including the private staff channel and the appeals channel where you discuss the people holding it. The specific permissions take ten minutes longer once.

A workable moderator set:

Manage Roles is the one to think about. It lets a moderator hand out any role positioned below their own — which includes the Moderator role itself if the ordering is wrong. Check the hierarchy before granting it.

Consider a two-tier structure: a Moderator role without Ban and a Senior Moderator role with it. Bans are the irreversible action and the one most improved by a second opinion.

Write down the decisions before anyone makes one

This is the step that separates a consistent team from an inconsistent one, and it takes twenty minutes.

List the three or four situations that actually recur in your server. Not every possibility — the ones you have already seen twice. For most communities that is: an argument that will not stop, an advertisement, a first-time rule break, and a member who is technically within the rules and clearly making the room worse.

Write down what happens for each, with a default. Not a legal code — a default and a note about when to depart from it:

Argument that will not stop. Ask once publicly to move on. If it continues, timeout both for 10 minutes. If it resumes afterwards, 24 hours and a note in the case history.

An unwritten standard is interpreted differently by every moderator, which members experience as unfairness rather than as variation — and it is the single most common reason a moderation team loses the room’s confidence.

Full treatment in the rules channel guide.

Give them somewhere to escalate

A team with no escalation route resolves everything itself, including the things it should not.

Name the cases that go up, explicitly: anything involving a threat, anything involving a minor, anything where a moderator is personally involved, and anything a moderator simply does not want to decide alone. That last one matters — permission to escalate on discomfort rather than on category prevents a great deal of over-correction.

Give it a channel, not a DM. A staff channel means the decision is visible to the team afterwards and becomes precedent. A DM to the owner means the same question gets asked again next month.

The tooling that makes consistency possible

Three things, and they are what turn a written standard into an enforced one.

A shared case history. Without it, each moderator starts at zero and a second warning means nothing — the fourth offence gets handled like a first because nobody remembers the previous three. Vetox’s case history holds 100 warnings per user across 1,000 warned users, free on Starter, which is what makes “this is your third time” a fact rather than a recollection.

Automated escalation. A rolling-window counter applies the rising response automatically, which removes the most common inconsistency: whether a repeat offence is treated as one depends on which moderator is online. Vetox’s Automated Actions covers this — 1 free, 50 on every paid tier.

A record of who did what. Not for suspicion — for answering questions weeks later, and for protecting moderators from accusations. 148 log event types and a 90-day trail of who changed which setting, both free on every tier.

Onboarding a new moderator

A short sequence beats a long document.

Week one — shadow. They read the staff channel and the case history and act on nothing. Most of what a moderator needs to know is contextual, and a week of watching transfers it faster than any guide.

Week two — reversible actions only. Deleting and timeouts, not kicks or bans. Every action they take gets a sentence in the staff channel explaining why, which is how the standard becomes shared rather than assumed.

Week three — full set, with escalation encouraged. The goal is not independence; it is calibrated judgement about when to ask.

What burns a team out

Recognising these early is most of retaining people.

Being the only one online. Coverage gaps mean one person absorbs everything. Automod is the partial answer — it works at 3am and does not need anyone present.

Decisions with no backing. A moderator whose call is reversed publicly without discussion stops making calls. Reverse in private, support in public.

Unbounded scope. “Moderator” quietly becoming event host, support agent and community manager is how people leave. Name the job.

No end. A rota with defined off-hours, and permission to be off during them, matters more than any tool. A team that is always on is a team that is briefly on.

Removing someone

It happens, and doing it badly costs more than the original problem.

Remove the role, not the person, unless they broke a rule. Someone stepping back from moderation is often still a good member.

Do it in one action. Removing permissions gradually is visible and humiliating.

Check the audit log afterwards — not from suspicion, but because a departing moderator’s last week is the period nobody was reviewing.

And revoke actual access, not just the label. A member-level permission grant survives the role being removed, which is the argument for granting by role in the first place.

Frequently asked questions

How many moderators does a server need? Two minimum, and enough to cover your busiest hours without anyone being permanently on.

Should moderators have Administrator? No. It overrides every channel override, including the private staff and appeals channels.

How do I keep decisions consistent? Write down the three or four recurring situations with a default response, and keep a shared case history so repeat offences are visible.

Should I use an application form? As a second step. Recruit from members already de-escalating voluntarily — the behaviour predicts the role better than the application.

What causes moderator burnout? Coverage gaps, unbacked decisions, unbounded scope and no defined off-hours.