Comment Moderation

The Comments page at /comments is the moderation queue for comments under your ad posts. Every comment the scanner finds is classified by AI, recorded in the queue with the platform it lives on and the action taken, and can be hidden, shown again, deleted, approved or answered — by you on the page, by the auto-moderation rules, or by the Agent in chat. All three paths go through the same moderation service, so the queue always shows what was done and who did it.

What Meta requires

Meta lets an app read and moderate comments under a post only with the token of the Page that owns the post — never with the ad account’s user token. AdWitch therefore:

• asks for pages_read_user_content (reading and deleting comments), pages_manage_engagement (hiding, showing and replying) and instagram_manage_comments (the same for Instagram) when you connect Meta — see Ad accounts setup;
• resolves the owner Page of each ad from its creative and takes that Page’s token from the Pages your connected Facebook user manages;
• refuses honestly, before any call to Meta, when the connection is known to lack one of those permissions (the account card offers Reconnect via Facebook) or when the connected user has no role on the owner Page (a Page admin grants roles in Meta Business Suite). A connection whose granted permissions are unknown is never treated as missing them.

Comments are moderated only for posts of ads in the ad accounts you can see; the Agent cannot act on a comment id that does not belong to one of them.

Facebook and Instagram

The scanner covers two surfaces of active ads and ads stopped within the configured window:

• the Facebook post of the ad (effective_object_story_id);
• the Instagram media of the ad, when the creative was published through a professional Instagram account linked to the Page.

Both are classified with the same AI rules, keyword lists and sensitivity. Each queue row shows a Facebook or Instagram badge and the feed can be filtered by platform. Instagram allows replying to top-level comments only, not to replies.

Comment feed

Use Scan Now to check comments right away; the scheduler also runs on the interval from the settings. The feed offers text search and filters by status, sentiment, platform and actor. Sentiment values are Positive, Neutral, Negative, Spam and Pending (still being analysed).

Each row includes the author, text, platform, sentiment, the moderation action and who performed it (Auto-moderation, the named team member, or the AI assistant), the timestamp, campaign and ad. When available it also shows the AI confidence score and reason, and the reply that was posted under the comment — by auto-reply, by a team member or by the Agent.

Row actions are Hide, Unhide, Delete, Approve and Reply, depending on the current state. Deletion is permanent and asks for confirmation.

Summary

The cards above the feed show Total, Hidden, Deleted, Flagged, Approved and Auto counts.

Moderation settings

The settings tab provides:

• an Auto-Moderation switch — scanning (scheduled or manual) runs only while it is on;
• separate automatic hiding for spam, negative comments, competitor promotions and URLs;
• Scan interval (minutes): leave empty for the platform default of 30 minutes; allowed values are 5–1440 minutes. Recently stopped ads (days) defaults to 7 and accepts 0–30;
• AI sensitivity: Low, Medium, High or Max;
• default action: Hide, Delete or Flag;
• blocked keywords, processed without AI analysis;
• whitelist keywords, which are automatically approved;
• notifications: Send summary (one notification per ad account and scan in which comments were hidden, deleted or flagged — each summary covers that account only) and Notify on each hide (a separate notification for every comment hidden or deleted automatically). They arrive in the bell; e-mail and digests follow each member’s Settings → Notifications (Autopilot category). Slack and webhooks are team destinations on Integrations. In agency mode a member restricted to some ad accounts sees only those accounts’ moderation notifications;
• Auto-reply — see below.

Save changes with the save control that appears after editing. The page also shows the last scan time and cumulative scanned, hidden and deleted counts once a scan has run.

Auto-reply

When Auto-reply is on, one templated reply is posted, as the Page, under each approved comment in two classes: a positive template and a question / neutral template. The reply goes out at the moment the comment is approved — either when the scan records it and the classifier approves it right away, or later, when a comment held for review is approved by a team member on /comments or by the Agent. Auto-reply posts at most one reply per comment, ever: a comment that already carries a reply (automatic, a team member’s or the Agent’s) never gets an automatic one, even if it is approved a second time, by two people at the same moment, or picked up by two scans at once. Replies you or the Agent write explicitly are always posted and become the reply of record. An empty template means that class gets no reply. Auto-reply never answers spam or negative comments, never answers a reply, a hidden comment or the Page’s own comments, and skips excluded campaigns; the scan-time reply is also paused while the team balance is exhausted. Each reply is recorded on the queue row, so you can see what was said and when.

If the reply cannot be posted — the Meta connection lacks a permission, the connected user has no role on the owner Page, the ad account is disconnected or Meta rejects the reply — the approval still stands and the page (or the Agent) tells you that the auto-reply was not sent. Once the cause is fixed, approving the comment again posts the reply. The one exception is an attempt whose outcome is unknown (the connection dropped before Meta answered, or Meta accepted the reply but it could not be recorded in the queue) — this holds for a reply you or the Agent sent as well: that comment is never auto-replied again, so a second reply cannot appear under it. The message says so; check the thread on Meta and, if needed, reply by hand — that reply is always posted and becomes the reply of record.

Working through the Agent

In chat, the Agent moderates through the same service and appears in the queue as the AI assistant. It can:

• show the queue with filters — “Show today’s negative comments”, “List comments under campaign X”;
• act on comments by their queue id or Facebook / Instagram id — “Hide the spam in campaign X”, “Approve this one” (with auto-reply on, the approval posts the templated reply and the Agent reports whether it went out), “Reply: thanks, we’ve sent you a message”;
• start a scan now — “Check comments now”;
• read and change moderation settings — “Turn on auto-hiding spam and send summaries” (changing settings needs an owner or admin role).

Deleting a comment and switching on auto-hide or auto-delete always require your confirmation in chat. Actions on accounts you cannot see are refused, and a missing permission or Page role is reported before anything is sent to Meta.

Not covered

Private replies to a comment (Messenger), real-time Page webhooks and comments outside your ads are out of scope — moderation is polling-based and limited to the posts of your ads.