Actions reference

Actions are what a branch does when its conditions match a comment. This page lists every action in the order they appear in the Action type dropdown, what each does on Meta (Facebook and Instagram), whether it can be undone, who gets notified, and how it shows up in your logs.

Hide vs delete, and can you undo it?

Hiding is what the Flag action does, and it’s reversible: a Workflow’s Allow action, or Unhide on the comment in the Inbox, un-hides it. Approving the comment in the Inbox only changes its status; it does not un-hide it. Delete removes the comment from the platform permanently and cannot be undone. There is no separate “Hide” action inside a workflow: Flag is the hide. To bring a hidden comment back, see Restoring a comment.

Actions run seconds after a comment arrives (roughly one second to evaluate, then up to about ten seconds to carry out on Meta). If a workflow is in observe-only mode, its actions are simulated: they are recorded in the log but never sent to Meta. Only the highest-priority matching workflow fires for a given comment, so only that branch’s actions run.

The Action type dropdown, showing the available branch actions.

Which actions cost you anything

Unlimited moderation, pay for the work. Your plan meters the automations that reach a follower in your name — and the ones where the AI model runs on your behalf. Defensive actions are free at any volume, so no amount of spam can run your allowance down:

Never counts toward your allowance	Counts toward your allowance
Flag (the hide)	Reply
Delete	Private reply (DM)
Allow	Send message
Promote	Like
	Reply with AI

An AI draft is an automation. Reply with AI used to be free on the grounds that it drafts text and sends nothing — but the model still ran, and running it is the work you are paying for. So it costs exactly one automation, whether the rule drafted it for your review or you pressed Generate AI draft in the composer yourself, and whether or not you go on to send what it wrote. Never two: if you send the draft, the send is the send’s own cost. A generation that fails — the provider is down, or times out — costs nothing at all.

Two more rules cover everything else. Anything you do by hand from the Inbox is never metered, and a workflow running in observe-only mode bills zero at any volume — it simulates its actions and sends nothing. Metering also lands only when the platform accepts an action, so a reply blocked by an expired token costs nothing.

For the allowances, the rates, and what happens when you reach them, see How automation billing works.

Flag

Flag hides the comment on Meta and sends you a notification. On Facebook the tool sets is_hidden=true; on Instagram it sets hide=true. Flag has one setting, Severity (Low, Medium, or High), which labels the flag for triage but does not change what happens on-platform.

Flag also sets the comment’s status to Flagged.

Reversibility: reversible

Flag is reversible. The hide is undone by a Workflow’s Allow action or by Unhide on the comment in the Inbox, which un-hide it on Meta. Approving the comment alone does not. See Restoring a comment.

Who gets notified: you do. Flag fires a push notification. Turn workflow-match notifications on or off under Settings → Notifications.

How it appears in the log: the comment’s status becomes Flagged (find it in the Inbox under the Flagged filter), and the action is written to the Audit Log (Analytics → Audit Log) with the workflow name and action type.

Input comment

spammy_bot Facebook

Check my page for a FREE giveaway!!! link in bio

↓

Matched condition

Keywords: "free", "giveaway" (match any)

↓

Resulting action Flag

Flag hides the comment on Facebook and sends you a push notification. Reversible via Allow.

Allow

Allow marks the comment as Approved and takes no destructive action. On Meta it un-hides a comment that was previously hidden (it sets is_hidden=false on Facebook, hide=false on Instagram). Allow has no settings.

Reversibility: reversible

Allow is itself the reverse of a hide. A later Flag (or a Hide by hand) can hide the comment again.

Who gets notified: no one. Allow does not send a notification.

How it appears in the log: the comment’s status becomes Approved (Inbox → Approved filter), and the action is recorded in the Audit Log.

Promote

Promote records the action but makes no change on Meta and has no settings. It is an internal signal only; nothing is hidden, deleted, or posted. It also leaves the comment’s status as it was — a promoted comment is not marked Approved or Flagged.

Where a promoted comment shows up:

• On the comment in the Inbox, as an automation chip naming the Workflow that promoted it.
• In Analytics → Audit Log, as a Promote entry, and in the Promoted This Month card there.
• In your webhooks, as a comment.promoted event, if you have a webhook subscribed to it — the way to send promoted comments to another tool.

Reversibility: nothing to reverse

Promote makes no on-platform change, so there is nothing to undo on Meta.

Who gets notified: no one.

How it appears in the log: the action is recorded in the Audit Log with the workflow name; no platform effect is sent.

Reply

Reply posts a public reply to the comment using a text template you write. The template can include the variables {{author}}, {{comment}}, and {{platform}}, which are filled in when the reply is sent. This is a real post to Meta.

Reversibility: not reversible

A posted reply cannot be undone from the tool. To remove it, delete the reply manually on the platform.

Who gets notified: the comment author sees your reply on the post.

How it appears in the log: the reply is recorded in the Audit Log with the workflow name and action type.

Reply with AI

Reply with AI generates a reply instead of using a fixed template.

The first setting, What happens when this fires, is the one that matters. Two of its three choices post to the public thread with nobody reading the reply first:

Choice	What it does
Draft it for review — nothing is sent	The reply is generated and waits in Inbox → Held until a person sends, edits or discards it. Nothing reaches the platform on its own. This is what a new action starts on.
Write a reply and send it automatically	The model writes the reply and it is posted, unattended, in your name.
Fill in my template and send it automatically	The model fills in your Reply Template and the result is posted, unattended, in your name.

The other settings shape the text: Tone (Friendly, Professional or Empathetic), Length (Short, Medium or Long), and — for the template choice — the Reply Template itself.

A reply can be held even when you asked us to send it

Both sending choices are still subject to the safety checks described in the action editor. If a reply cannot be safety-checked — the language isn’t one we can evaluate, the check itself fails, or the safety gateway isn’t evaluating — it is held in Inbox → Held rather than sent, and the queue says which of those happened. It still counts as one automation, because the model ran.

Reversibility: not reversible

Once a reply is posted it cannot be undone from the tool; remove it manually on the platform.

Who gets notified: the comment author sees the reply on the post when one is sent.

How it appears in the log: the action is recorded in the Audit Log with the workflow name.

What it costs: one automation per generation — see Which actions cost you anything. This is charged when the model runs, not when a reply is sent, so a draft you never send still costs one and a generation that fails costs none. Sending a held draft from Inbox → Held costs nothing extra: you did that, not the automation.

Private reply (DM)

Private reply (DM) sends a direct message to the comment’s author instead of replying in public. It is only available for Facebook and Instagram. The action tile shows the policy chips 7d, 1/comment, and fb·ig:

• 7d — the private reply can be sent only within 7 days of the comment being posted. After that, Meta no longer allows it, and no message is sent.
• 1/comment — one private reply per comment. Once a private reply has been sent for a comment, no second one is sent for it.
• fb·ig — Facebook and Instagram only.

If the workflow isn’t scoped to a Facebook or Instagram page, the tile is locked with the message: “Private reply is only available for Facebook or Instagram. Scope this workflow to a Facebook/Instagram page to enable it.”

Reversibility: not reversible

A sent direct message cannot be recalled or undone.

Who gets notified: the comment author receives a direct message.

How it appears in the log: the message is recorded in the Audit Log, and the direct message appears in the Inbox under the DMs tab. Automated DMs carry an “Auto-replied by workflow” badge.

Send message

Send message replies in an existing direct-message conversation: when someone messages your Page or Instagram account, the Workflow can answer them in that thread. Like Private reply, it is Facebook and Instagram only, and it needs the DMs channel turned on in the Workflow’s scope — the tile says so if it isn’t. Its policy chips are 24h, 1/message, 750/h/account, and fb·ig:

• 24h — the reply must go out within 24 hours of the person’s message. That is Meta’s messaging window.
• 1/message — one reply per incoming message.
• 750/h/account — at most 750 sends per hour from each connected account. That is Meta’s own Send API rate limit.

Private reply or Send message?

They both send a DM, but they start from different places. Private reply answers a public comment by DM — the first message in a new conversation, allowed within 7 days of the comment, once per comment. Send message answers a direct message someone sent you, in the same thread — within 24 hours of their message, once per message. Use Private reply on comment Workflows and Send message on DM Workflows.

Reversibility: not reversible

A sent message cannot be recalled or undone.

Who gets notified: the recipient receives a direct message.

How it appears in the log: the message is recorded in the Audit Log and appears under the Inbox DMs tab and in outbound messaging.

Like

Like likes the comment on Meta. It has no settings. Like is not supported on TikTok, where the dropdown shows it as disabled with the suffix “(TikTok unsupported)”.

Reversibility: not reversible from the tool

There is no “unlike” action in a workflow, so a like placed by a workflow can’t be removed from within the tool.

Who gets notified: the tool sends no notification for a Like.

How it appears in the log: the action is recorded in the Audit Log with the workflow name and action type.

Delete

In the editor, Delete is labeled destructive: “Irreversible action. This action permanently deletes the comment. It cannot be undone.”

That warning is real. On Facebook and Instagram, Delete removes the comment from the post on the platform — for you and for everyone else. Like every on-platform action, it runs only if the connected account granted the permission it needs; if it didn’t, nothing is removed and the Audit Log entry shows that the action did not run.

Reversibility: irreversible

A deleted comment cannot be restored — not by the tool, and not on the platform. If you want a reversible way to take a comment down, use Flag, which hides it instead.

Who gets notified: the tool sends no notification for a Delete.

How it appears in the log: the action is recorded in the Audit Log with the workflow name and action type.

Input comment

troll_account Instagram

This product is garbage and so is everyone who buys it

↓

Matched condition

Keywords: "garbage" (match any)

↓

Resulting action Delete

Delete removes the comment from Instagram for good. If you might want it back, use Flag to hide it instead.

Related

• How workflows work
• Conditions reference
• Testing a workflow
• Restoring a comment