Amazon review integrity: detect, act on and track fake or manipulated reviews

You are a product manager at Amazon. Design a system that detects and keeps track of review abuse in the Amazon store. ```hint Define abuse first "Review abuse" covers several behaviors with different actors and different victims. Decide which kinds you will track and who each one hurts before choosing signals, because the signals, the action you take and the cost of a mistake all differ by kind. ``` ### Clarifying Questions - Which kinds of abuse are in scope: paid or incentivized positive reviews, negative reviews posted to attack a competitor, reviews from people who never used the product, or offensive and off-topic content? - Who is the main customer of the system: shoppers who rely on ratings, sellers whose ratings are distorted, or the internal trust and safety team that investigates and enforces? - Does tracking mean screening each review as it is submitted, monitoring review health across products over time, or both? - May the system act on its own (hide a review, leave it out of the star rating, restrict an account), or does it only flag reviews for human investigators? ### What a Strong Answer Covers - A breakdown of abuse types, the actors behind each, and the harm each does to shoppers, honest sellers and trust in ratings. - Detection that combines review content, reviewer behavior, purchase history and links between accounts into a risk score, rather than relying on a single rule. - Graduated responses with human review and appeals, and an explicit view of the cost of removing a genuine review. - How to measure abuse that is mostly never labeled, with success metrics and false-positive guardrails. - How abusers adapt, what sellers are told without revealing detection logic or shopper data, and what ships first. ### Follow-up Questions - Abusers start buying the product before posting a paid review, so the review carries a verified purchase badge. Which signals still catch them? - Sellers begin reporting their competitors' honest negative reviews as abuse. How do you stop the reporting channel from becoming a weapon? - Leadership asks what share of reviews in the store are fake today. How would you estimate it when you only see the abuse you catch?