Check reviews against your policy before they go live
US and UK rules. A low rating or harsh opinion never holds a review, and each hold has a reason. Text only: fake review evidence stays with fraud tools.
Try it on this example
Product title: Glass cafetiere, 8 cup, with bamboo lid
Star rating, 1 to 5: 2
Review title and text
- What should happen to this review before it goes live?Hold so a moderator removes personal details87%
- Does the review describe the writer's experience of this product, its delivery or the seller, rather than unrelated content?Yes99%
- Does the writer say they received the product free, a discount, a refund, a draw entry or any other reward in return for the review?Yes99%
- Does the writer say or clearly imply a connection to the seller, the brand or a competitor?No84%
- Does the writer say they have not yet received or used the product they are rating?No97%
- Is the review clearly about a different product from the product title?No92%
- Does the review text plainly contradict its star rating?No88%
- Does the review include personal data the policy forbids?Yes98%
- Does the review contain abuse, hate, threats, sexual content or other content that may be illegal to publish?No92%
- Does the text describe a hazard or an injury from the product?Yes98%
- Does the writer say someone offered them a reward, or pressed them, to post a positive review or to change or remove a negative one?No88%
These are real answers stored from one run on this example.
The prism behind it
Check reviews against your policy before they go live
Fields
- Product title
- Star rating, 1 to 5
- Review title and text
Context
We are Mossvale Home, a homeware retailer selling online in the US and the UK. Customers review products on our site after they buy. Each review is read here when it is submitted, before it goes live. The answers suggest what happens next. Moderators decide every hold, and no review is removed or hidden on these answers alone. Code logs each suggestion with its reason, and a writer whose review is not published is told why. Rule set: - US: the Federal Trade Commission rule on consumer reviews and testimonials bans fake reviews, including reviews that falsely claim experience of a product; rewards for reviews that must say something positive or negative; reviews by officers or managers, or reviews they ask staff or relatives to write, that do not disclose the connection; and suppressing negative reviews, by threats or by presenting a filtered set as all the reviews. - UK: the Digital Markets, Competition and Consumers Act 2024 bans fake reviews and concealed incentivised reviews, and requires a business that publishes reviews to take reasonable and proportionate steps to prevent and remove them. The CMA explains those steps in its fake reviews guidance. Both apply to us, so the policy below meets both. Our review policy: - Every review of the writer's own experience of the product, its delivery or our service goes live, whatever its star rating. A low rating, harsh criticism, criticism of us, or a comparison with a competitor is never a reason to hold a review. - A writer who says they got the product free or at a discount, a refund, a draw entry or any other reward for the review: the review goes live with an "Incentivised review" label. - Held for a moderator to check: the writer says or implies a connection to us, the brand or a competitor, such as working for, owning, supplying or being family of one; the writer says they have not received or used the product but rates it; the review is about a different product, or is not about any experience of it (adverts, links, spam, unrelated text). - Held so a moderator can remove the details, then published: a phone number, an email address, a full street address, an order number, or a full name used to identify another person, including our staff. - Held for a moderator to look at first: abuse, hate, threats, sexual content or content that may be illegal. - A review that describes an injury or a hazard goes live as normal and is copied to the product safety team. - A writer who says anyone offered them a reward, or pressed them, to post a positive review or to change or remove a negative one: the review goes live as normal and is copied to the compliance team. - A review whose text plainly contradicts its stars goes live, and code asks the writer to check the stars. Code, not these answers, checks whether the writer bought the product, the age of the account, bursts of reviews, shared devices and our own incentive programme records. The text alone cannot show that a review is fake.
Questions
What should happen to this review before it goes live? Choice
Judge the review against the review policy in the context. A low rating or a negative opinion is never a reason to hold it. This is a suggestion; a moderator decides every hold. When more than one option fits, pick the one lowest in the list.
Does the review describe the writer's experience of this product, its delivery or the seller, rather than unrelated content? Yes / No
Short reviews count: "Great, does the job" is about the experience. Adverts, links, spam, political or unrelated text, and a question with no experience in it are not. Yes: The review is about the writer's experience of the product, its delivery or the seller. No: The review is about something else, or is only a question.
Does the writer say they received the product free, a discount, a refund, a draw entry or any other reward in return for the review? Yes / No
Count a reward the writer says they got for writing the review, or for writing it in a certain way. The word "free" alone does not count: "BPA free" and "free delivery" are not rewards. Yes: The writer says they got something in return for the review. No: The writer mentions no reward for the review.
Does the writer say or clearly imply a connection to the seller, the brand or a competitor? Yes / No
Count working for, owning, supplying or being family or a close friend of the seller, the brand or a competing business. A free product for a review is an incentive, not a connection. Yes: The writer says or clearly implies such a connection. No: The writer shows no such connection.
Does the writer say they have not yet received or used the product they are rating? Yes / No
Count "not arrived yet but five stars", "haven't tried it" and "bought it as a gift, can't say". A review of delivery or service alone is still an experience. Yes: The writer says they have not received or used the product. No: The writer has received and used it, or says nothing to the contrary.
Is the review clearly about a different product from the product title? Yes / No
Compare what the review describes with the product title. A different size or colour of the same product is the same product. Yes: The review plainly describes another product. No: The review fits the product title, or names no product at all.
Does the review text plainly contradict its star rating? Yes / No
Count a 1 or 2 star review that praises everything, or a 4 or 5 star review that condemns the product. Mixed reviews with a middling rating do not count. Yes: The text and the stars plainly point in opposite directions. No: The text and the stars fit, or the text is mixed.
Does the review include personal data the policy forbids? Yes / No
Count a phone number, an email address, a full street address, an order number, or a full name used to identify another person, including our staff. A first name alone, or a town, is allowed. Yes: At least one such item appears in the review. No: No such item appears.
Does the review contain abuse, hate, threats, sexual content or other content that may be illegal to publish? Yes / No
Strong, rude or angry criticism of the product or of us is allowed and does not count. Count insults or threats aimed at a person, attacks on a group for who they are, sexual content, and content that may be illegal. Yes: The review contains at least one of these. No: The review contains none of these, however harsh its opinion. A breach of the review rules, such as an incentive or a connection to the business, and personal details are not counted here; other questions record them.
Does the text describe a hazard or an injury from the product? Yes / No
Count burns, scalds, cuts, shocks, smoke, fire, a burning smell, breaking glass, tipping furniture or any injury. Figures of speech such as "this lamp is fire" do not count. These go to the product safety team; the review still goes live. Yes: The text describes at least one hazard or injury. No: The text describes no hazard or injury.
Does the writer say someone offered them a reward, or pressed them, to post a positive review or to change or remove a negative one? Yes / No
Count an offer of a refund, a voucher or a gift for a good review or for taking down a bad one, and threats or repeated requests to change a review, from us, the brand or anyone else. These go to the compliance team; the review still goes live. Yes: The writer says this happened. No: The writer says nothing of the kind.
Lens columns
suggested_action, suggested_action_probability, about_the_experience, about_the_experience_probability, mentions_incentive, mentions_incentive_probability, connection_to_business, connection_to_business_probability, not_used_yet, not_used_yet_probability, different_product, different_product_probability, rating_text_mismatch, rating_text_mismatch_probability, personal_data_exposed, personal_data_exposed_probability, abusive_or_illegal, abusive_or_illegal_probability, safety_mention, safety_mention_probability, pressure_to_change_review, pressure_to_change_review_probability
Run it on your own text
Add this prism in the app, change any question, and test it on a file of your own.