Skip to content
Get Started

Review moderation that lets an unhappy customer speak

The Review preset validates the rating, scores the text and judges each review against AI instructions you can edit.

Get Started FreeSee Pricing
  • Templates on Starter and up
  • A review costs one check
POST/v1/moderations
Sample data
Validation0 of 2 fields valid
  • Ratingrating7Rating must be at most 5
  • Reviewbodyok.5 to 3,000 characters
PoliciesNo model called
  • Toxicity
  • Spam & advertising
  • Off-topic
  • AI instructions

Validating1 error, 0 checks

Sample data

Two fields, one rule and the policies a review needs

Pick the Review preset when you create a template in the dashboard, and you get your own copy to edit before you publish.

Review preset
Preset as shippedpreset: "review"

Fields

FieldTypeRuleValidationTriggers
RatingratingNumberRequired1 to 5Validated only. The AI instructions read it with the text.
ReviewbodyTextRequired5 to 3,000 charactersToxicity, flagsHate speech, flagsPersonal data, adds a review signal

Policies

FlagThe review becomes FLAGGED.
ToxicityHate speechViolenceSexual contentSwear wordsSelf-harmIllicit & regulated goodsAI instructions
ReviewAdds a review signal without flagging.
Personal dataSpam & advertisingLinks & contact detailsLookalike charactersMeaningless textOff-topic
ObserveScored and returned, never affects the verdict.
Made with AIGrowth and up

Rules

01Zero tolerance for hateif Hate speech is flaggedReject
Severity triage0.400.80AllowReviewReject

When no rule matches, severity triage decides.

Preset as shipped

Say what a review should be, in plain words

The AI instructions are one paragraph in the template's Instructions tab. Edit it and publish, and the next review is judged against the new text.

instructions
A score from 0 to 1. From 0.70 it flags the review.
explanation
Why the review got that score, in a sentence or two.
matches
A short quote from the review, as evidence.
AI instructions
Sample data
Reviews should describe a real experience with what is reviewed. Flag reviews that are clearly fake or paid for, that attack a person rather than the product or service, that promote a competitor, or that are unrelated to the purchase or visit.
Evaluated once per submission across all fields, with your project context and guidelines. Returned as the instructions policy.
Form · reviews

The shop offered a refund for five stars, so here they are. The kettle is fine.

AI instructionsbodyTriggered0.76
confidence:80"offered a refund for five stars"

The reviewer says the rating was given in exchange for a refund, so the review reads as paid for.

FlaggedTriggeredSent to reviewReview

Sample data

A one-star review is not a violation

A low rating triggers nothing by itself, so two one-star reviews can get opposite verdicts.

  • Harsh about the productEvery score stays under 0.40, so severity triage allows the review.

  • Abusive about a personToxicity flags the text from 0.70, and severity triage rejects from 0.80.

Two one-star reviewsSample data
  • Form · reviews

    Leaked from the base on day two, and support took a week to reply. I would not buy this kettle again.

    SafeNot flaggedPublishedAllow
    Toxicity0.18
    AI instructions0.05
    Off-topic0.02
  • Form · reviews

    The owner is a pathetic little man. Anyone who shops here is as stupid as he is.

    FlaggedTriggeredRejectedReject
    Toxicity0.94
    AI instructions0.83
    Swear words0.21

Start moderating your users' content

Create an API key, send your first request and get a verdict back.

Get Started FreeSee Pricing