Recruiting Admins Who Don't Quit After a Month

Published 15 September 2026 · JoinGG

Admin teams collapse for four predictable reasons: the job was never defined, the permissions were all-or-nothing, the only feedback was a complaint, and the reward for good work was more work. Fix those four and recruit from your regulars, and a volunteer team holds together for years instead of weeks.

The pattern is familiar to anyone who has run a server for more than a season. You promote three people in March. By May, one has stopped logging in, one is arguing with players in public, and one is doing everything, which means they will be gone by July. So you promote three more, and the cycle repeats, and each round leaves you slightly more convinced that nobody is reliable.

Nobody being reliable is not the explanation. Volunteer moderation is unpaid work with visible costs and invisible rewards, and it only survives when the structure around it is deliberate. That structure is not complicated, but almost none of it is the recruiting step everybody focuses on. It is worth reading this alongside the guide to why nobody joins your server, because a thin staff team and a thin population usually have the same root cause: nothing about the server was ever written down.

Why admin teams fall apart in the first month

Run a post-mortem on any collapsed staff team and you will find the same four causes, usually all of them at once.

1. The job was never defined. "Help out when you can" is not a job. The new admin cannot tell whether they are doing well, so they either do nothing and feel guilty or do everything and burn out. Both look like quitting.

1. Permissions were all-or-nothing. Either they got no tools, in which case being staff means being blamed for things they cannot fix, or they got everything on day one, and one confused evening later somebody has changed the map rotation and banned a regular by mistake.

1. The only feedback was negative. The messages an admin receives are complaints, ban appeals and "why did you kick my friend". If the owner only speaks up when something went wrong, the job feels like being on trial permanently.

1. The reward for good work was more work. The reliable admin gets asked to cover the extra shift, handle the difficult appeal, and be online for the event. That is a promotion into exhaustion.

You can see this in your own action log. On most teams, a couple of people generate most of the recorded actions. That imbalance is not laziness on one side, it is an unmanaged rota on the other, and the two overloaded people are the ones you are about to lose.

Workload also depends on what you run, which owners routinely underestimate. Minecraft is the clearest example: JoinGG tracks roughly 56,400 concurrent players across 4,230 listed Minecraft servers, an average of about thirteen per server, and a 60-slot survival server with land claims, an economy and a build queue produces steady administrative work that a deathmatch rotation never does. Grief reports need investigation, claims need adjudicating, and none of it is instant. Check the shape of your own game against the Minecraft player statistics before you decide how many people you need.

Write the role description before you post the opening

Write the document first. Not because it is professional, but because writing it forces you to answer questions you have been avoiding: how many hours you actually expect, what an admin is allowed to decide alone, and who they go to when they are unsure.

A usable admin role description fits on one screen and contains seven things.

The seven fields

• Scope. Which servers and which Discord channels. Not "the community".

• Time. A number: about four hours a week, at least two of them inside your peak window.

• Powers. The exact commands and tier, linked to the permission table below.

• Escalation. The cases they must hand up rather than decide: permanent bans, disputes involving another admin, anything involving a minor.

• Reporting line. One named person. Not "the admin team".

• Term. Six months, renewable by a two-minute conversation.

• Review date. The calendar date of the first check-in, set on day one.

The term limit is the field people skip and the one that does the most work. A six-month renewable term makes leaving normal. Without it, the only way to stop being an admin is to quit, and quitting a volunteer role feels like letting people down, so instead admins go quiet and stay on the permissions list for a year. A term that simply ends removes all of that.

Write the negative list too, because it is what protects the person. An admin is not required to play when they do not feel like it. An admin is not required to answer direct messages about server decisions. An admin does not set the rules, they apply them. Making those explicit stops the slow expansion of the role that turns a two-hour commitment into a second job.

Then publish the description where players can read it. When the community knows exactly what a moderator may and may not do, half the accusations of favoritism disappear on their own, because the boundary is checkable rather than assumed. This is the same mechanism that makes a whitelist people actually apply to work: stated criteria beat vibes, on both sides of the table.

Recruit from your regulars, not from an application form

Open applications have one structural flaw that no form design fixes: they select for people who want the powers. The trait you actually need is somebody who already behaves like staff without the tag, and those people rarely apply, because they are busy playing.

So invert it. Watch your regulars for four weeks and look for specific, observable behavior.

• They play at consistent hours. This is not a personality trait, it is a rota solving itself.

• They answer other players' questions unprompted, including boring ones about how to connect.

• They report with evidence, and they report people they get along with.

• They lose an argument in chat without escalating it.

• They are still around three months after they first joined.

The counter-signals are just as reliable. Somebody who asks for admin more than once, somebody whose first conversation about the role is a list of people they would ban, somebody who only appears when there is drama, and somebody whose only connections are to existing staff. None of these are character judgments. They are predictors, and they hold up across games and communities.

Make the ask privately and make it easy to decline. Name the specific thing you noticed, send the role description, and give them a week. Say plainly that saying no costs them nothing and you will not ask differently in future. Out of a base of roughly forty weekly regulars you should expect to find three or four genuine candidates, and expect at least one of them to say no, which is a healthy sign that they understood the commitment.

Your Discord is where most of this observation happens, so keep the invite on your listing current in the JoinGG communities directory. It also helps to have low-stakes mechanisms that show who participates constructively before you ever hand out powers: map votes and rock-the-vote thresholds are a surprisingly good signal, and the explainer on how server voting works covers how to set them up without letting a loud minority run the rotation.

The trial period that actually filters

Thirty days, with the end date in the calendar from day one, and a real possibility of ending in a no. A trial that always converts is not a trial, it is an onboarding delay that teaches the candidate their judgment is not being assessed.

Structure it in two halves. In week one the trial admin has no powers beyond a spectator tag and one instruction: when something happens, write down what you would have done and send it that evening. You compare their answer to yours. This is the highest-signal week of the whole process, it costs the server nothing, and it surfaces the two failure types early, the person who would ban for backchat and the person who would let anything go rather than be disliked.

For weeks two to four they act with the trial tier from the permission table: kick, gag and mute up to 24 hours, nothing that is hard to undo. Four things are worth measuring, and all four are countable.

1. Attendance against their own stated hours. Not against yours. They set the number in the role description; the only question is whether it was real.

1. Logging rate. What share of their actions have a written reason attached. Under about 80 percent is a habit that will not improve on its own.

1. Escalation accuracy. Did they hand up the cases the role description said to hand up, and did they avoid escalating routine calls they were allowed to make?

1. Behavior in the one bad case. Every trial contains one incident that goes wrong. How they handled being wrong tells you more than the other 29 days.

Set expectations openly at the start: roughly one trial in three does not convert, that is normal, and a candidate who does not convert keeps their standing as a regular. Say this on day one and a no becomes survivable for both of you. Scale matters here too. A small co-op server, the kind that fills most of the Left 4 Dead 2 server list at eight slots a time, may only generate two or three incidents a month, in which case extend the trial to 60 days rather than promoting on no evidence.

Permission tiers: what each role can and cannot do

Every permission you grant has a blast radius, which is the amount of damage the holder can do in one bad decision, one compromised account, or one confused evening. Design tiers around blast radius rather than around seniority or trust, because trust is not a technical control and it does not survive a stolen session token.

The single worst habit in server administration is handing out RCON. RCON is one shared password, it carries no per-person audit trail, and it has no scope: whoever holds it can change any cvar, execute any config, empty the server, or take it offline. When two admins share it, nobody can prove who did what. The same problem applies to /op on Minecraft, which grants permission level 4 in one step, and to Pterodactyl subusers given file manager or SFTP access, where a single wrong click deletes a world folder that your backup schedule may or may not cover.

The fix is named accounts at every tier. SourceMod admin entries keyed to a SteamID, LuckPerms groups with inherited nodes, txAdmin accounts with two-factor enabled, and Pterodactyl subusers with narrow permission sets. The comparison of SourceMod, AMX Mod X and Oxide is a good starting point for which framework your game supports and how granular its flags actually get.

Two rules make the table hold. Grant the lowest tier that lets somebody do their actual job, and review tiers on the same date as the term renewal so promotion is a decision rather than a drift. Anchoring this to a live example helps: a competitive Counter-Strike 2 server running practice nights needs moderators who can change maps at 21:00, which is a senior tier decision on paper, so either promote deliberately or add a scheduled rotation and remove the need. CS2 currently sits at roughly 4,020 players across 859 listed servers on the CS2 statistics page, so most of those servers are small enough that one or two senior moderators is the whole answer.

Covering the clock without burning anyone out

Start from the arithmetic, because it settles most arguments. A server that is meaningfully populated sixteen hours a day needs 112 hours of coverage a week. Six volunteers at four hours each gives you 24. You are not covering the clock, and any plan that assumes otherwise will quietly convert into two people covering everything.

So cover the peak and let systems hold the rest. Across everything JoinGG polls, the busiest hour measured was 242,719 players connected at 18:00 UTC on a Saturday, against 73,158 at 06:00 UTC on a Wednesday. That is a spread of more than three to one, and your own curve will be sharper still because your player base is regional rather than global. Two moderators overlapping in your busiest three hours are worth more than eight names on a flat rota, and the network-wide statistics give you a reasonable shape to start from.

Rota rules that survive contact with volunteers

• Name a person per block, not a group. "Someone will be around" reliably means nobody was.

• Use two to three hour blocks. Longer shifts get skipped, and a skipped shift is worse than an unstaffed one because it was counted.

• Spread timezones, not headcount. Three admins at UTC+1, UTC-5 and UTC+8 cover more real hours than six in one country.

• Playing is not covering. An admin in a competitive match is not moderating. If the block requires attention, it requires somebody who is not mid-round.

• Publish the gaps. Players are far more tolerant of "no staff on before 15:00 UTC" than of an unanswered report.

Then let automation hold the unstaffed hours: chat rate limits, vote-kick with a sensible threshold, and a report queue that captures evidence without a human present. Slot count drives all of this more than owners expect, and the guide to choosing how many slots your server needs is worth reading as a staffing document, since every ten extra slots adds roughly proportional moderation load. Uptime alerting belongs in the same bucket: monitoring your server uptime means the 04:00 crash is handled by a notification rather than by an admin who happened to be awake.

Keeping admins, removing admins, and the exit nobody writes down

Recruitment is the easy half. Whether the team exists in six months is decided by what happens between incidents, and by how you handle the two endings that every team eventually faces.

Pay in something other than more work

Volunteers are not paid in gratitude, they are paid in status, autonomy and reduced friction. Concrete things that work: a reserved slot so they never queue, a visible tag, a real say in the map rotation, hosting credit toward their own project, and public credit by name when a call was handled well. The most valuable one is the cheapest: let them reduce their hours when they ask, without an explanation and without it counting against them. An admin who knows they can step back to one shift a week is an admin who does not disappear.

Feedback that is not a complaint

Book fifteen minutes a month with each admin. Show them their own numbers, actions taken, reports handled, appeals overturned, and name one specific call they made well. This is a small ritual and it directly counters the third collapse cause, because it guarantees at least one conversation a month that is not somebody complaining. Skip it for two months running and you will notice the logging rate fall before you notice anything else.

When an admin abuses power

Separate the bad call from the pattern. A single wrong ban is a training issue and should be handled the way you would want yours handled: correct the ban, tell the player it was corrected, and speak to the admin privately. A pattern is different, and the tell is usually a run of actions with no logged reason, bans of people the admin was arguing with, or permissions being used outside their tier.

When you see a pattern, freeze the permissions before you open the conversation. Freezing is not an accusation, it is how you stop the situation getting worse while you read the logs, and it protects the admin too if the logs clear them. Then talk privately, decide, and if the answer is removal, announce it in one factual line with no details. Publicly litigating a former admin's conduct is how you lose the rest of the team, who are all now imagining the same thread about themselves.

The exit checklist

Run the same list for everybody, including friendly departures, and run it the same day. Remove the SourceMod entry or LuckPerms group, delete the txAdmin account, remove the panel subuser, drop the Discord role, rotate the RCON and panel passwords if they ever held them, take them off the rota, and move any documents they owned. Then thank them publicly by name.

Doing this identically for every exit is what makes it routine rather than an insult. The alternative is easy to end up with: a permissions list containing people who have not logged in for a year, any one of whom might be reusing that password elsewhere. If you run several servers across different titles, the same discipline applies to each of them, and the game hubs in the JoinGG game catalogue are a practical way to keep an inventory of what you are actually responsible for.

Frequently asked questions

How many admins does a game server actually need?

Enough to cover your three busiest hours twice over, which for most community servers is four to six people. Below three, one holiday leaves the server unstaffed. Above eight, most of the team never acts, the rota stops meaning anything, and standards drift because nobody handles enough cases to build judgment.

Should I let players apply to be an admin?

Open applications select for people who want the powers, which is the wrong filter. Use them only as a signal box: let people register interest, then ignore the form and pick from your regulars. If you must run applications, treat a repeated application as a reason to wait, not a reason to promote.

How long should an admin trial period be?

Thirty days, with the end date written in the calendar on day one. Shorter than that and you see too few real incidents to judge anyone. Longer and the trial becomes the job, the limited permissions start to feel like an insult, and ending it reads as a demotion rather than a decision.

Should admins have RCON access?

No. RCON is a single shared password with no per-person audit trail and no scope limits, so one leaked credential or one bad night can wipe your config. Give admins named accounts in SourceMod, LuckPerms or txAdmin instead, and keep RCON with the one or two people who handle the server process itself.

What should I do when an admin goes inactive?

Ask once, privately, after two missed weeks. If life got busy, move them to a reserve tier that keeps the Discord role and drops the rota slot. If there is no reply after a second week, remove the permissions and say why in one line. Inactive admins with live permissions are a security problem, not a courtesy.

Should I pay my server admins?

Usually not, because money changes a volunteer relationship into a job you are managing badly. Pay in things that cost you little and matter to them: a reserved slot, a say in the map rotation, hosting credit, a public thank-you, and the ability to drop hours without being made to feel guilty.

Do the cheap parts first. Write the role description this week, publish it, and put your existing admins into tiers even if the tiers only exist on paper for now. Those two steps take an evening and they remove most of the ambiguity that makes volunteer teams fall apart, long before you need to recruit anybody new.

After that, staff against real numbers instead of impressions. JoinGG polls 11,561 listed servers hourly across 147 games and keeps seven days of history, which is enough to show you exactly when your server fills and where your coverage gaps are. If you are not tracking that yet, add your server to JoinGG, watch a fortnight of data, and build the rota around what the graph shows rather than what you assume the evenings look like.

JoinGG lists 11,561 game servers across 147 games with hourly polling, live player counts and seven days of history.