Testing rules before going live

You do not have to guess. Before a Workflow (an automation that acts on comments) touches a single comment on Facebook or Instagram, you can check exactly what it would do. There are two safety nets, and you can use both:

1. Preview — run a sample comment through your branches with no calls to Meta at all.
2. Observe-only mode — turn the Workflow on but have it only record what it would do, without doing it.

And even after you go live, the most common first action, Flag, is reversible. So the worst realistic outcome — a comment hidden that shouldn’t be — is one you can undo. This page shows the safe path from “I built a Workflow” to “it’s running in production.”

Preview a branch without touching Meta

The preview runs a comment you supply through the Workflow’s branches (a branch is a set of conditions plus the actions to run when they match) and reports what would happen. It is completely safe: no comment is hidden, no reply is posted, nothing is written to your Audit Log, and no call is made to Facebook or Instagram. It is a pure dry run.

For each branch, the preview tells you:

• whether the branch matched,
• for every condition (a test the comment must pass) in that branch, whether that condition matched and a short human-readable reason, and
• which actions would fire if that branch matched.

You can preview a Workflow before you save it, or preview a saved one.

Preview cannot change anything

The preview uses the same matching logic as the live engine, but it only reads — it never hides, replies, deletes, or logs. Running a preview as many times as you like has zero effect on your comments or your usage.

Preview shows every branch, not just the winner

There is one important difference between the preview and live moderation. The preview reports the match status of every branch in isolation. Live moderation does not: for a given comment, only the first matching branch in the highest-priority matching Workflow actually fires, and everything below it is suppressed.

So a preview can show two branches both matching, when in production only the first of them would ever run. Read the preview top to bottom and treat the first matching branch as the one that would fire. For the full precedence rules, see How workflows are evaluated.

Observe-only mode: run it live, act on nothing

Preview tests one comment at a time against comments you type in. Observe-only mode is the next step up: it runs the Workflow against your real incoming comments, but every action is simulated — recorded in the Audit Log with a simulated status, with nothing sent to Meta.

This is the honest way to answer “would this Workflow have caused problems?” You let it watch live traffic for a while, then read the log to see exactly which comments it would have hidden or replied to — before it has the power to actually do so.

An observe-only Workflow still wins its slot

Observe-only does not mean “inert.” A higher-priority observe-only Workflow still counts as the match for a comment and still suppresses lower-priority Workflows — it just simulates its own actions instead of performing them. If a live Workflow below it stops firing while you are testing, the observe-only one above it is matching first. See How workflows are evaluated.

The safe rollout, step by step

1. Preview against sample comments. Type in a comment you expect to match, and one you expect not to match. Confirm the right branch matches each and the actions listed are the ones you intended.

2. Turn the Workflow on in observe-only mode. It now evaluates real comments but only simulates its actions.

3. Let it run, then read the Audit Log. Go to Analytics → Audit Log and look at the simulated actions. Every entry shows the Workflow name and the action that would have fired. Ask: is it matching the comments you wanted, and only those?

4. Escalate to live actions. Once the log looks right, take the Workflow out of observe-only so its actions run on Meta.

5. Start with reversible actions. For the first live version, prefer Flag (which hides the comment and can be undone) over anything one-way. Widen to replies or direct messages once you trust the matching. See Start with reversible actions.

Screenshot — pending capture The Audit Log under Analytics, showing an automation that fired

The Audit Log (Analytics → Audit Log) is where you review what a Workflow did — or, in observe-only mode, what it would have done.

”It hid a comment it shouldn’t have”

This is the fear that keeps people from enabling a Workflow, so here is the plain answer: on Facebook and Instagram, a wrongly hidden comment is recoverable.

Symptom: a Workflow hid a comment (its status is Flagged) that you wanted to keep visible.

Cause: a branch’s conditions were broader than you expected and matched the comment, and its Flag action hid it. Flag hides the comment on Meta (it sets is_hidden=true on Facebook, hide=true on Instagram) and sends you a push notification.

Fix: un-hide it. Approving the comment, or applying an Allow action, un-hides it on the platform (it sets the hidden flag back to false). The comment becomes visible again. See Restoring a comment.

Prevention: preview and run in observe-only first (above), and narrow the condition that over-matched. Keyword matching is case-insensitive and matches partial text — the keyword art also matches “smart” and “party” — so tighten broad keywords and prefer match all over match any when you need every term present.

Delete is the one that isn't reversible — and it doesn't work on Meta anyway

Flag hides (reversible). Delete is described in the editor as “Irreversible action… It cannot be undone.” But on Facebook and Instagram the tool has no delete capability, so a Delete action against a Meta comment does nothing — it is a no-op, and the comment stays visible. To take a comment down on Meta, use Flag, which is both what works and what you can undo. See Actions reference.

Start with reversible actions

The safest first live action is one you can take back. On Meta, that is Flag: it hides the comment and notifies you, and an Allow or a manual approval un-hides it. One-way actions — Reply, Reply with AI, Private reply (DM), Send message, Like — post real content or send real messages that cannot be recalled from the tool. Add those only after observe-only has shown the matching is clean.

Input comment

curious_customer Facebook

Is this a scam? Seems too cheap to be real

↓

Matched condition

Keywords: "scam" (match any)

↓

Resulting action Flag

A broad keyword like 'scam' will match genuine questions, not just spam. In preview this branch matches; in observe-only mode you'd see it in the Audit Log as a simulated Flag before it ever hides anything. If it goes live and hides this real question, Allow (or approving the comment) un-hides it.

One workflow acts per comment

Only the highest-priority matching Workflow fires for a comment, and only its first matching branch. When you test a new Workflow, remember an existing higher-priority one may already be handling the comments you are aiming at. How workflows are evaluated covers the full order.

Related

• How workflows are evaluated — first-match-wins, priority, and why the preview shows more matches than production.
• Conditions — the four condition types and exactly how each one matches.
• Actions — which actions are reversible and which are one-way.
• Restoring a comment — un-hide a comment a Workflow hid.