Workflow Recipes - Facebook comments + Workflow AI

Task-specific steps

This recipe combines a Facebook comment trigger with generated reply text and sentiment-based branching. Human review and a safe fallback are essential because generated classifications can be wrong.

1. 1Create the workflow from the Facebook Comments + Workflow AI recipe and leave it unpublished while configuring accounts and prompts.
2. 2Limit the trigger to first-level comments on the intended page and post, then define any qualifying phrase or exclusion rules.
3. 3Rewrite the response prompt with the brand voice, prohibited claims, response length, and instruction not to invent facts.
4. 4Configure the comment-reply action to use the generated output only after testing offensive, ambiguous, negative, and unrelated comments.
5. 5Review the sentiment prompt and branches. The positive branch may send the interactive direct message; neutral, negative, or uncertain results should follow a deliberate alternate path.
6. 6Run controlled comments, inspect the stored inputs and outputs, and verify each public and private response fires once.

Feature-specific checks

• Do not send a private offer solely because an AI model guessed the sentiment; include deterministic eligibility rules.
• Provide a handoff or no-response path for sensitive, abusive, or support-related comments.
• Monitor replies after launch and update the prompt when real examples expose an unsafe or inaccurate pattern.

Before you begin

“Workflow Recipes - Facebook comments + Workflow AI” is part of the AI Studio → Workflow AI area in Adaptive AI. Use AI-assisted actions and content inside workflows. Automation changes can affect many records at once, so the safest approach is to define the intended entry condition, action, and stopping point before publishing anything. The objective is not merely to save a setting; it is to confirm the result works for the staff member or customer who will depend on it.

• Write down the exact trigger, filters, expected action, and condition that should stop or bypass the automation.
• Use an internal test contact with complete sample data, a valid communication channel, and consent appropriate to the test.
• Confirm that every connected account, custom field, pipeline, calendar, or template referenced by the automation still exists and is available.
• Record the current published status and settings so you can compare behavior or restore the earlier configuration if necessary.
• Choose a quiet test window when a controlled enrollment cannot interfere with active customer communication.

Verify the result

After completing the instructions in this guide, use a controlled test before treating the setup as finished. A saved screen or success message confirms configuration, but only an end-to-end test confirms the business outcome.

1. 1Save the configuration, review every branch from top to bottom, and publish only after required fields and warnings are resolved.
2. 2Run one controlled test using known inputs and note the exact time the test record entered the automation.
3. 3Open execution history and confirm that the record followed the expected branch in the correct order.
4. 4Verify the intended message, task, field update, or status change occurred once and contains the correct merged information.
5. 5Run a negative test that should not qualify; it should remain outside the automation without producing side effects.

Troubleshooting checklist

If the result differs from what you expected, preserve the current state and isolate one variable at a time. Start with the checks below before deleting the configuration or rebuilding it from scratch.

• Confirm the automation is published and that the trigger filters match the test record exactly, including capitalization and date conditions.
• Review re-entry settings to determine whether the same record is allowed to enter again or is being correctly excluded.
• Check the first failed or skipped step in execution history instead of changing several steps at the same time.
• Verify wait steps, account time zone, business-hour windows, and scheduled-send rules when an action appears delayed.
• Reconnect or reauthorize any external service that reports an expired credential, missing scope, or unavailable resource.

Best practices

• Use names that state the audience, trigger, and outcome so another administrator can understand the automation without opening it.
• Keep entry filters narrow and add explicit stop conditions to prevent duplicate messages, tasks, or opportunity changes.
• Respect contact consent, do-not-disturb status, quiet hours, and channel-specific opt-out requirements in every communication path.
• Change one behavior at a time, test it, and document the result before expanding the automation to more records.
• Review execution errors and unexpected enrollments after launch, then again after any connected integration or template changes.