Rules for Reviews & Predictions
Written By Chad McGuire (Sparrow Intel)
Overview
Rules let you say "when X happens to a review (or prediction), do Y." Once you have a few rules in place, the platform starts doing routine work for you — replying, marking, escalating, creating tasks, requesting feedback, posting host-to-guest reviews — without anyone touching a button.
This lesson covers the building blocks of a Review rule and a Prediction rule, with concrete examples for each.
If you've read Rules for Conversations, this will feel familiar — it's the same Rules engine with different target types.
How a rule works
Every rule has three parts:
A trigger — what event the rule reacts to (a new review posts, a new prediction generates)
Conditions — checks that must be true for the rule to fire (rating ≤ 3, OTA = Airbnb, has cleaning issues)
Actions — what happens when the conditions match (auto-respond, mark for removal, create task, email)
Rules evaluate every time a matching event occurs. They are not retroactive — turning on a rule today won't process yesterday's reviews.
You'll find rule management in Rules in the main navigation. Review rules use the Review target type. Prediction rules use the Prediction target type.
Review rules
Conditions you can use on Reviews
A non-exhaustive list of what you can match on:
Conditions combine with AND by default. A rule with three conditions fires only when all three match.
Actions you can take on Reviews
Example Review rules
Auto-respond to clean 5-stars.
When a new review posts AND rating = 5 AND no cleaning issues AND no maintenance issues, auto-respond with Chirp AI.
This rule alone clears a substantial percentage of the review queue on most properties. It's the safest auto-respond pattern.
Auto-create maintenance tasks from review-detected issues.
When a new review posts AND has maintenance issues, create task in Breezeway (or your task system).
One review mentioning the dishwasher → one targeted maintenance task with the issue text pre-filled.
Escalate 1-star Airbnb reviews to a supervisor.
When a new review posts AND rating ≤ 2 AND OTA = Airbnb, email the supervisor team.
A simple, opinion-free flag so nothing slips while a busy front-line agent is working through the queue.
Prediction rules
Conditions you can use on Predictions
Predictions don't have a star rating to match on (no review has posted yet) — they have a predicted sentiment and the issue extractions from the stay.
Actions you can take on Predictions
Example Prediction rules
Send a recovery touchpoint when a stay is predicted negative.
When a new prediction generates AND predicted sentiment = Negative, send message (template: "We saw a few hiccups during your stay — we'd love to make sure everything's resolved before you head out.").
This is the highest-ROI prediction rule. It opens a recovery channel automatically on every red-flagged stay.
Email a supervisor on negative-prediction stays at flagship properties.
When a new prediction generates AND predicted sentiment = Negative AND property group = "Flagship," email the supervisor team.
For your most-visible listings, you want a human eye even when the recovery message has already gone.
Create a cleaning task for predicted cleaning issues.
When a new prediction generates AND has cleaning issues, create task in the cleaning queue.
Catch the issue before the next stay even checks in. Especially valuable for back-to-back turnovers.
Common pitfalls
Over-triggering. A rule that emails on every review will flood inboxes. Narrow with rating or sentiment conditions.
Conditions ANDed when you meant OR. Build narrower rules and let multiple rules fire if you want OR-like behavior.
Forgetting forward-only. A rule turned on today won't act on existing reviews or predictions.
Doubling up on auto-respond. Don't have two rules that both auto-respond to overlapping conditions. Pick one source of truth.
Auto-responding to negatives without supervision. Chirp can draft a response to a 1-star review, but you almost certainly want a human glance before it posts. Restrict auto-respond to clean positives.
Testing a rule
Before turning a rule loose, the safest pattern is:
Build the rule with the action set to "Star" only (e.g., as a "would-fire" marker)
Let it run for a day
Filter to "Starred" and confirm the rule fired on the right reviews/predictions
Replace the test action with the real ones
Turn the rule on for real
This is especially worth doing for rules that post responses, mark for removal, or send messages — the irreversible kind.
Up next
Host-to-Guest Reviews: Automating Outbound Reviews (Airbnb) — your team's side of the review, drafted by Chirp AI under your guardrails.