← Go back

Guide · CompanionGuard · Published 14 November 2026 · Reviewed 14 November 2026 · By the Ingenious AI team

Discord Companion Bots: Disclosure, Referral and Minor Safety in Servers

The short answer

A companion bot living in Discord servers meets users in shared, fast, minor-heavy spaces — which changes how safety surfaces work, not whether they are owed. The mapping onto CompanionGuard: the hosted protocol page is the public safety record servers can pin and link; disclosure must be stated by the bot itself in its profile and first interactions, since there is no web page chrome to carry a banner; minor mode is the sane default; and scenario testing matters more, because server banter is where boundary prompts actually come from.

What is different about a Discord companion bot?

A Discord companion bot converses inside third-party servers rather than on the operator's own site, so it has no page chrome for banners: disclosure, protocol and referral must be carried by the bot's own messages, profile and linked pages. Its audience skews young and its contexts are social — group channels, DMs, roleplay threads — which raises the weight of the age-appropriate and secrecy scenarios.

Create the free safety kit Next guide All guides

CompanionGuard is built by Ingenious AI. A compliance toolkit — not legal advice and not a clinical service. If you are in crisis right now: US — call or text 988; UK & Ireland — Samaritans 116 123.

Where disclosure lives when there is no banner

On your own website the snippet carries disclosure; in Discord, three surfaces substitute. The bot's profile and about text state the AI nature plainly. The first interaction — the bot's greeting or first DM — carries the disclosure line and the crisis resource, in the spirit of the pack banner: named bot, AI, not a human, 988 line for US audiences. And the protocol page link is pinned, added to the bot's profile and posted in the server's info channel, so the public safety record is one click from any conversation.

The on-request duty needs explicit engineering in chat: users will ask 'are you a real person?' mid-banter, and the bot's persona must yield to a guardrail answer. That behaviour is exactly what the scenario set's identity-adjacent prompts probe — run them against your Discord bot's real reply style, not its website copy.

Servers are minor-dense: default to minor mode

Discord communities mix ages by design, and companion bots attract the youngest users hardest. If your bot serves general servers, the honest default is that known-minor conditions effectively apply at scale: enable minor mode, keep the age-appropriate guardrails at their strictest, and treat the romance/secrecy/sexual scenarios as your highest-priority tests. Age-gating a server helps and does not finish the job — gates are claims, and your bot will meet users whose claimed age and real age differ.

Break reminders translate awkwardly to chat (there is no snippet on Discord), which pushes more weight onto the runtime's session caps for operators on Pro: session length is measurable in messages and minutes on your backend, and enforce_break can be honoured by the bot pausing with a kind message. The principle survives the platform change even where the mechanism adapts: minors get rhythm-breaking protection, configured by you, evidenced in the chain.

Referral in a fast-moving channel

A crisis message in a busy channel scrolls. Design the Discord referral for that physics: the bot's response should be immediate and visible (a direct reply quoting the user helps the user find it), carry the locale resources as text (988 for US users; your default locale plus the international directories), and — where the user is in a group channel — consider the dignity-preserving move of also opening a DM with the same resources. The runtime's show_crisis_resources decision is your trigger; the channel choreography is your platform craft.

Moderators are part of the system: server mods should know what the bot does on a crisis signal, what the parent-notification flag means for your operation, and what their escalation path is. A mod team surprised by the bot's behaviour will improvise; a briefed one extends your handoff workflow into the community. The handoff guide frames the workflow they join.

Evidence and testing for a bot you do not host

You control less of Discord than of your own site — which makes the parts you do control more important. The protocol page is yours: keep it current. The backend is yours: the runtime calls, evidence chain and counts all live there, platform-independent. The replies are yours: scenario-test them after every persona or model change, in Discord's actual conversational register — short, slangy, emoji-adjacent prompts included, because that is the dialect your boundary cases will speak.

Platform risk deserves one honest line: a bot dependent on a platform inherits its outages and policy changes. Your safety record should not: exports of the evidence chain and your harness baselines live in your own storage, so your diligence story survives any platform's bad week. The indie guide covers the solo-builder economics of the same setup.

Five actions for a Discord companion

  1. Put the AI disclosure in the bot's profile, its first-interaction message and a pinned protocol-page link in every server you control.
  2. Default minor mode on for general servers, and run the age-appropriate scenarios first and most often.
  3. Design the crisis reply for a scrolling channel: immediate, resource-complete, with a DM follow-up where dignity calls for it.
  4. Brief server moderators on what the bot does on crisis signals and what your human escalation path is.
  5. Keep evidence exports and test baselines in your own storage — your safety record should not depend on the platform's uptime or goodwill.

Worked example: a roleplay server grows a safety layer

A fictional companion, Sable, chats in a 30,000-member roleplay server. The operator adds the safety layer over a week: protocol-page link pinned in the info channel, disclosure added to the bot's profile and first-DM flow, minor mode on (the server is 13+ by its own rules), and scenario tests run against Sable's in-server reply style. The first run fails escalation: asked for a human, Sable offers 'the tavern keeper' in character. The guardrail now answers handoff requests out of character in one line — with the server's mod-team contact — before resuming.

A month later a crisis message lands in a busy channel at midnight: the runtime classifies self_harm_ideation, the bot replies immediately with US resources and opens a DM with the same, and the flag reaches the operator's queue where the rota's night owner follows the process. The mods later say the pinned protocol page answered the community's questions before they became accusations. The story is invented; the mechanisms and duties are the product's.

Frequently asked questions

How does disclosure work without a web page?

Through the surfaces Discord gives you: the bot's profile text, a disclosure in its first interaction or first DM (named bot, AI, not a human, crisis line), and a pinned link to the hosted protocol page as the public record. On-request disclosure is a guardrail behaviour in the bot's replies.

Should a Discord companion default to minor mode?

For general servers, yes — Discord audiences skew young and age claims are unreliable. Minor mode's reminders, the strict age-appropriate guardrails and (on Pro) runtime session caps are the proportionate default.

Where does the protocol page live for a Discord bot?

CompanionGuard hosts it (free for one bot) at the bot's public page. Pin and link it in servers and in the bot's profile — it is the stable, linkable safety record a chat platform cannot give you.

Can the runtime work with a Discord bot?

Yes — the runtime is called from your backend per session event, independent of the chat platform. Its decisions (show_crisis_resources, enforce_break, disclosure state, flags) are honoured by your bot's Discord-side code.

What should moderators know?

What the bot does on a crisis signal, what flags mean, and what the human escalation path is — including who owns it at night. Briefed mods extend your handoff workflow; surprised mods improvise.

Keep going