Automatic on positive reviews, approved by you on the others
A tool that publishes everything automatically is frightening, and rightly so: a clumsy reply under a one star review is visible to everyone and does not go away. A tool that asks for approval on everything saves almost nothing, since half the work is thanking happy people.
Reply splits the decision by rating, and the default protects the expensive case.
No commitment · Cancel in one click
Each rating, and its default handling
| Rating | What Reply does | What you have to do |
|---|---|---|
| Four and five stars | Writes and publishes a personalised reply, after a natural delay | Nothing |
| Three stars | Writes a reply acknowledging the problem and sends it to you | Publish, rewrite, or skip |
| One and two stars | Writes a draft and sends it to you, publishing nothing | Decide |
| Reviews flagging something serious | Come to you in every case | Read, and decide |
How each band is handled is set in the dashboard. This table describes the default behaviour, on an account nobody has touched.
Why the line sits at four stars
A four or five star review calls for a thank you. There is no judgement to make, no legal risk, no version of events to dispute. The only quality expected is that the reply talks about what the customer talked about, which a model does very well from the text of the review.
Below that, a reply commits you. It acknowledges a problem or not, offers a gesture or not, contradicts the customer or not. Those are owner decisions, not writing decisions.
Three star reviews stay on your side deliberately: it is the band where the reply most often needs a judgement a model cannot make, because the customer liked one part and not the other.
The guard that cannot be switched off
Some reviews always come back to you, whatever your settings: the ones the model reads as flagging hygiene, safety, discrimination, or a legal threat.
The reason is simple: publishing a polite paragraph under a review about food poisoning, and telling the owner nothing, is the worst possible outcome. That guard is not a setting, it is a rule of the product.
What this split guarantees
- No negative review published without your approval, by default.
- Thank yous on positive reviews handled without you, which is most of the volume.
- A review flagging something serious always coming back, whatever happens.
- A pending reply visible in the dashboard immediately, even at night.
What it implies
- There are still decisions to make: the product does not remove mixed reviews.
- A pending review you never answer is never published, so it stays unanswered.
- The line sits at four stars, not elsewhere: it is not configurable rating by rating outside the bands provided.
- A positive review published automatically went out without you reading it, if you leave the default in place.
How to check it
A single definition in the code decides what goes out unreviewed, and it is read both by the pipeline that publishes and by the prompt builder. A disagreement between the two would produce a reply written as a draft for a human, published as is.
When a reply goes out unreviewed, the prompt stops telling the model it will be read over. That is not cosmetic: some writing latitude is only granted because a human approves afterwards, and it disappears with them.
An inactive business generates nothing and publishes nothing. That is what makes the "configured but not yet paid" state safe.
Common questions
Can I approve positive reviews as well?
Yes. Automatic replying to positive reviews can be turned off in the business settings, and then everything comes through you.
What happens if I never answer an approval request?
Nothing is published. The review stays pending, visible in the dashboard, until you decide.
Can I rewrite a reply rather than publish it as is?
Yes, by simply answering the message you received. The text you send is published as is.