MB SuiteMB Suite

The AI Community Manager: drafts or automatic replies in the Inbox

Updated 2026-10-12

In two lines: the AI Community Manager is the position that replies in the Inbox — and you can have more than one, each with its own territory. In Suggest drafts mode it writes the reply and your team sends it; in Auto-reply mode it answers on its own, only in the sources you enable.

Where do I configure it?

  • Go to Social Media > Inbox and open the Settings tab.
  • Open the AI Team section. The table starts empty: use Add position to hire a Community Manager AI —you can create several— or the Sentiment analyst, which classifies sentiment and spam and is one per account.
  • Click the row to open its configuration, organized in tabs: General (status, name and agent), Behavior (working mode, escalations and instructions), Knowledge, Territory, Schedule and Test. Each position keeps its own switch to turn it on or pause it.
  • If you create more than one Community Manager, Territory is what tells them apart: the sources and types each one covers, plus specific accounts and posts. When two of them cover the same thing, the table order breaks the tie — drag the rows to set the priority.

You need the Inbox running first: at least one connected network and the tray activated with Turn on inbox. Until then there's nothing for it to reply to.

Who replies: choosing the agent

In the Agent section you pick which agent from your AI team holds the seat. The selector only offers agents that reply to your audience on social, so you don't put a lead-facing salesperson in front of your comments. If you don't pick one, it runs on the Default Social agent and shows the DEFAULT badge; the Edit agent link opens that agent's file in a new tab, without losing the configuration you're editing.

In the Knowledge tab, the Inherits from {agent} block shows the knowledge it's taking from the agent that holds the seat. Whatever you write in Instructions belongs to this seat and takes precedence if the two contradict each other. The whole model is explained in Agents and seats: who's who on your AI team.

The two working modes

  • Suggest drafts — the AI writes a suggested reply; your team reviews it and sends it. Nothing goes out without a person clicking send. It's the mode to start with.
  • Auto-reply — the AI replies on its own. Two controls define how far it goes: Where it can reply (a matrix where you enable network by network and type by type: comments, messages, mentions) and the escalations in Behavior, led by Escalate complaints to a human — which only works while the Analyst position is active (see the warning below).
Auto-reply publishes without review, and a comment reply is public. Before turning it on: run a few weeks in Suggest drafts to see how it writes, and leave Escalate complaints to a human on — note: this safety net only works while the Analyst position is active; without it sentiment isn't classified and the Community Manager replies anyway, complaints included. Create it from AI Team > Add position > Sentiment analyst before switching to Auto-reply. With the Analyst on, when a complaint or negative sentiment is detected the AI doesn't reply: it marks that commenter's conversation as Needs attention and alerts the team. In a direct message, asking to talk to a person is the exception: it answers with the contact channel it has loaded or, if there's none, alerts the team (details below). Enable only the sources you're sure about. Keep in mind that a busy post is many conversations, one per commenter: the AI decides comment thread by comment thread, so on the same post it can reply to some people and escalate others. If something goes off, pausing the position with its switch stops it from taking new conversations right away.
If its territory includes Instagram mentions, keep in mind those mentions live on other accounts' posts. When someone names your brand with @ in the text of their post or in a comment, the Community Manager replies right there, as a guest —brief, with no promotions or links, and without speaking as if the post were your brand's—; and if the mention is in a comment, since Instagram doesn't say who wrote it, it doesn't make up a name. Posts where you were only tagged in the photo can't be answered from MB Suite: the Community Manager neither replies to them nor prepares a draft. More in The Inbox: every comment, message and mention in one tray.
The Schedule tab decides when it works. With Working hours off, the position works around the clock. When you turn it on, you set the time zone and the slots, and choose what happens Outside working hours: Do nothing (messages wait for a human), Draft only (it prepares the reply for you to review, without sending it) or Reply anyway. Draft only is the sweet spot for going to sleep with the inbox covered but nothing published without a human reading it first.

How does it answer direct messages?

In direct messages people often send several messages in a row: "Hi", "do you ship?", "it's for Friday". That's why the Community Manager doesn't answer each message the moment it arrives: it waits for the person to stop writing and replies once to everything pending, in a single message that follows the order in which they wrote it. It usually replies around one or two minutes after their last message. If the person writes again while it's drafting, it throws that draft away and waits again, so the reply that finally goes out already takes the new message into account. This applies to Instagram, Facebook (Messenger) and TikTok direct messages; comments and mentions work as before: it answers each one separately.

To write, it reads the conversation from both sides: what the person wrote and what your brand answered, from MB Suite or from the network's app. That's how it avoids repeating what was already said. It only greets when the conversation starts, or when your brand hasn't written in it for more than a day; the rest of the time it gets straight to the point. And when it calls the person by name, it uses only the first name on their profile and never their username on the network; that goes for comments too. If MB Suite doesn't have their name (TikTok direct messages, for instance, only bring the username) or the profile looks like a company's (a name in capital letters or in decorative letters), it simply doesn't use one; and if the profile starts with a title, such as Arq. or Dr., it skips it.

In Suggest drafts mode it works the same way, but instead of sending it leaves a single suggestion for everything the person wrote: you'll find it in the reply box as AI Community Manager suggestion, ready for Use suggestion. If the person writes again before anyone answers, the new suggestion covers everything that's still unanswered. Escalations are decided in one go too: if any of those messages is a complaint (with the Analyst active and Escalate complaints to a human on), it doesn't reply and alerts the team just once.

In a direct message, asking to talk to a person doesn't count as a complaint, even if it comes with some frustration at not finding one and the Analyst marks it negative: it answers with the contact channel it has in the agent's knowledge (you see it in the Knowledge tab) or in Instructions (a WhatsApp, a phone number, an email or a link). With Escalate complaints to a human on, if it has no channel, or if the request comes with a complaint about something specific, an accusation or an insult, it alerts the team instead. The Analyst is the one that recognizes the request: without it active, nothing changes.

Your team always comes first. If, before the Community Manager replies, someone on your team answers the person —from the Inbox or from the network's app— or the conversation is assigned to a person, the Community Manager leaves those messages to your team. And if it has already replied many times in the same conversation on the same day —for instance, because another bot is answering on the other side—, it stops replying and marks the conversation as Needs attention for your team.

Answering doesn't stop the Community Manager for good: if the person writes again, it picks the conversation back up. If you want to carry on yourself, assign it to yourself with Assign > Assign to me (or use Take control if the Community Manager is already handling it).
To see what it would answer without anything going out, open the position's Test tab and choose Real conversation: it puts the reply together just like in the tray, with everything the person wrote since your brand's last reply. It runs on the real agent, so it consumes AI credits. With Manual text, a test direct message counts as the first message of a new conversation, which is why that reply opens with a greeting. With Manual text the Analyst doesn't classify the message either, so complaints and requests to talk to a person don't behave as they do in the tray: test those with Real conversation.

How do I use it day to day?

  • The list has two views, with the Conversations | Posts switch at the top. In Conversations each row is one person: a commenter's thread on a post (their comment plus the replies below it) or a direct message conversation. That thread — not the whole post — is the unit the AI drafts for and replies to.
  • Open a conversation in the Inbox and, in the reply box, press AI draft. Pick a Tone if you want it warmer, briefer or more formal. In a direct message, the draft answers in a single message everything the person wrote since your brand's last reply.
  • The proposal appears as AI Community Manager suggestion: read it, use Use suggestion to drop it into the box, edit whatever you want and send it.
  • To moderate a piece of content as a whole, switch to Posts: each row shows how many comments are Unanswered, and opening it lists that post's commenter threads. Reply opens the same box — AI draft included — without leaving the post, Open in Conversations takes you to the full thread, and Mark all as resolved closes the pending ones on that post in one go.
  • If it escalated a conversation, you'll find it under Needs attention; you can also get an in-app notification by turning on AI escalation in the Inbox's notification settings.
  • If a conversation is already being handled by a position, the reply box appears covered with Conversation handled by {name} — click to take control. That's on purpose: it prevents two replies going out to the same person. Click it, confirm Take control, and the conversation becomes yours — the AI stops replying in that thread and an internal note records the handover.
  • You can also hand a conversation over the other way round: in Assign, under AI Team, pick the Community Manager you want to take it.
Reply with the content in front of you: on a comment or a mention, the context card above the thread shows the post it belongs to, and View post opens it inside MB Suite with its preview and, if the post is yours, its metrics — the same button sits in the header of the post panel in the Posts view. No need to jump to the network. The full published-post view is in Published tab: see how your posts are performing.
The Community Manager doesn't see images or listen to audio, but it knows what arrived —a photo, an audio message, a story, a shared post— and replies with that in mind. When a message brings nothing to answer —it arrives empty or with content the network doesn't deliver—, the Community Manager neither answers it nor prepares a draft: it stays in the tray for someone on your team to check in the network's app.
Everything these positions do consumes AI credits —the Community Managers when they write, the Analyst on every message it classifies— and it shows up with the Social Inbox origin in AI Agents > Usage, attributed to each position. Handy for telling apart what your inbox spends from what the rest of the team does, and for narrowing a territory if the cost surprises you. See See your AI usage and control your credits.

To make its writing sound like your brand, define the tone once in Brand Voice and load in the agent what it should always know about you. The Community Manager uses your Communication tones, Tone notes, Words and phrases to use, Words and phrases to avoid, Emoji usage, Audience persona and each network's Tone note. How it addresses people in Spanish (tú, vos or usted) is your call too: if you write it there —even by country, such as «usted in Mexico»— it follows it; if you say nothing, it uses tú.

Still have questions?

Ask MIA, the MB Suite AI assistant: open it with the MB AI button (⌘J) in the top bar. MIA knows the section you're in, so you can ask it about what you see on screen — and it can also answer how-to questions about using MB Suite.