The AI Analytics Center builds a behavior profile, toxicity score and risk score for each member from the messages, voice activity and interactions on your server. You see the results on the dashboard and with the /analyze command; when something crosses a threshold, the bot sends an alert to your staff.
The first part of the window above resembles the hourly activity chart you see on the dashboard: the busiest and quietest hours of the server are marked. Below it is the heat meter alert posted to the moderator channel: the argument between Kai and Leo in #general is heating up, and a staff member turns on a 30-second slowmode with one click.
Who can see it?
Analysis runs in the background on every server, regardless of plan. A plan is only needed to view the results.
- To see the results, the server owner or one of the admins added on the dashboard must have the Max plan (or higher). The plan of the person opening the page is not enough on its own.
- The dashboard page can be opened by the server owner, dashboard admins, and anyone with the Administrator or Manage Server permission on Discord.
/analyzerequires the Manage Messages or Administrator permission. - If a plan is purchased later, past records become visible too; data keeps being collected in the meantime.
Critical safety is free
Detection of self-harm, grooming, scams, doxxing, threats and under-13 age disclosures doesn't use this plan gate; it runs on every server and every plan. It is configured with /critical-safety.
What's on the dashboard?
Dashboard → My Servers → the Analytics shortcut on the server card takes you to the AI Analytics Center page. The AI Analytics Center card on the server page opens the same page; there is no separate setup.
| Section | What it shows |
|---|---|
| Server Summary | Analyzed users, total messages, toxic incidents, high-risk users, messages today, members active in the last 30 days and the busiest hour. |
| Top Users | Ranks members over the last 7 days by toxicity, overall risk, ban risk, alt account suspicion, raid/troll risk, messages or toxic incidents. |
| User Detail | The selected member's behavior profile, risk assessment, 30-day risk and toxicity trend, activity heatmap, banter/toxic ratio, the people they interact with most and their recent toxic incidents. |
| Server-wide Statistics | The toxic incident trend over the last 30 days, the most frequent categories and the distribution of behavior types. |
All Servers view
In User Detail you can switch between This Server and All Servers. All Servers only combines data from servers where you are an admin; other servers are shown only as a count.
/analyze step by step
Risk alerts and feedback
Risk scores are recalculated every night. If a member crosses the threshold, the bot sends an alert to the Analytics & Moderator Feedback channel you chose in the Logs settings. The decision your moderators make tunes the system to your server.
Safety analysis
| Command | What it does | Who uses it |
|---|---|---|
/heat-meter | Argument heat meter: when an argument heats up in a channel, it sends an alert and slowmode buttons to the log channel. Subcommands: status, toggle, dry-run, cooldown (1–180 min), banter-bonus, exclude-channel. | Administrator |
/fingerprint | Reports possible alt accounts using writing-style fingerprints. Subcommands: status, toggle, dry-run, pool-limit (50–2000), scan. | Administrator |
/critical-safety | Configures the six critical safety classes, alert channels, the role to mention, the quarantine duration for scams and the self-harm support message. | Administrator |
/calibration | Threshold suggestions based on moderator feedback. | Administrator |
The heat meter and fingerprinting start off
Both are off by default, and when turned on they start in trial (dry-run) mode: they send no real alerts or reports, they only keep records. For real alerts, turn it on with toggle, then turn off trial mode with dry-run. Alerts go to the Analytics & Moderator Feedback log channel; if that channel isn't set, they reach no one.
Critical safety classes
| Class | What the bot does |
|---|---|
| Self-harm / Suicide | No punishment for the user. A quiet notification goes to the separate self-harm alert channel, and the user is sent a support message. |
| Grooming | Alert in the general alert channel; the notification role is mentioned and the message is preserved. |
| Scam/Phishing | The message is deleted and the user is quarantined (timed out). The duration is 1440 minutes (24 hours) by default. |
| Age Disclosure (<13) | Notification as required by Discord's rules; no punishment. |
| Doxxing | A message containing personal information is deleted immediately and an alert goes to the general alert channel. |
| Threat | For a concrete threat of violence the role is mentioned and the message is preserved as evidence. |
Be sure to set the critical safety channels
All six classes are on by default, but without an alert channel the notifications reach no moderator. Use /critical-safety channel to pick the general alert channel and a separate, private channel for self-harm.
Nightly Maintenance
Risk scores, clusters, channel report cards and churn predictions are refreshed in order every night. /nightly-report shows a step-by-step summary of that night's work; since it processes all servers, only the bot owner can use it.
Setup
Step 1
On the dashboard, pick your server on the My Servers page and open the Analytics shortcut. Data is already being collected; there is nothing you need to turn on.
Step 2
For the channel where alerts are posted, choose the Analytics & Moderator Feedback channel in the Logs settings.
Step 3
Set the critical safety alert channels with /critical-safety channel and the role to mention with /critical-safety role.
Step 4
If you like, turn on the heat meter and alt account detection with /heat-meter toggle and /fingerprint toggle; watch them in trial mode for a while, then switch to real alerts with dry-run.
On every server
Data collection and critical safety (self-harm, grooming, scams, doxxing, threats, age disclosure) work on every plan.
Viewing the results
The dashboard page, /analyze reports, heat meter alerts, alt account reports and the Community Panel require the Max plan on the owner or a dashboard admin.
View PlansScores are signals, not verdicts
Risk and toxicity scores are based on a sample of messages; with little data confidence is low and the report says "uncertain". Never punish anyone based on a score alone; look at the messages in the alert yourself and record your decision with the feedback buttons.
Frequently asked questions
Do I need to turn analysis on?
No. The AI Analytics Center is always on for every server; it can't be turned off. What you turn on and off are the heat meter, fingerprinting and the critical safety classes.
Can the member see the /analyze report?
No. /analyze replies are only visible to the staff member who used the command.
I bought a plan but the page still says "Subscription Required"
The plan must be on the server owner or on an admin added to that server on the dashboard. Being an Administrator on Discord, or the plan of the person opening the page, is not enough on its own.
Why is a new member's report empty?
For accounts with few messages, behavior and risk can't be determined yet. The report fills up as the member writes.
Where are channel health and churn risk?
These are covered in the Community Panel guide.
Risk alerts never arrive
In the Logs settings, check that the Analytics & Moderator Feedback channel is selected and that the bot can post in it. If no member crosses the threshold, no alert arrives either.
Open the analytics page
On the dashboard, click your server's Analytics shortcut and look at the server summary and member details.
Go to dashboard
Sign in to comment
You can sign in with your Discord account in seconds.