Customer reviews have been a core part of why customers love shopping at Amazon ever since we opened in 1995. We make sure that it’s easy for customers to leave honest reviews to help inform the purchase decisions of millions of other customers around the world. At the same time, we make it hard for bad actors to take advantage of our trusted shopping experience. That’s where artificial intelligence (AI) comes in.
Amazon’s product review process using AI
So, what happens when a customer submits a review? Before being published online, we use AI to analyse the review for known indicators that the review is fake. The vast majority of reviews pass our high bar for authenticity and get posted right away. However, if potential review abuse is detected, there are several paths we take. If we’re confident the review is fake, we move quickly to block or remove the review and take further action when necessary, including revoking a customer’s review permissions, blocking bad actor accounts, and even litigating against the parties involved. If a review is suspicious but additional evidence is needed, our expert investigators, who are specially trained to identify abusive behaviour, look for other signals before taking action. In 2023, we proactively blocked more than 250 million suspected fake reviews from our stores worldwide.
“Fake reviews intentionally mislead customers by providing information that is not impartial, authentic, or intended for that product or service,” says Josh Meek, Senior Data Science Manager on Amazon’s Fraud Abuse and Prevention team. “Not only do millions of customers count on the authenticity of reviews on Amazon for purchase decisions, but millions of brands and businesses count on us to accurately identify fake reviews and stop them from ever reaching their customers. We work hard to responsibly monitor and enforce our policies to ensure reviews reflect the views of real customers, and protect honest sellers who rely on us to get it right."
Other ways Amazon is using AI to ensure authentic customer feedback
Among other measures, we use the latest advancements in AI to stop hundreds of millions of suspected fake online reviews, manipulated ratings, fake customer accounts, and other abuses before customers see them. Machine learning (ML) models analyse a multitude of proprietary data including whether the seller has invested in ads (which may be driving additional reviews), customer-submitted reports of abuse, risky behavioural patterns, review history, and more.
Large language models (LLMs) are used alongside natural language processing techniques to analyse anomalies in this data that might indicate that a review is fake or incentivised with a gift card, free product, or some other form of reimbursement. We also use deep graph neural networks (GNNs) to analyse and understand complex relationships and behaviour patterns to help detect and remove groups of bad actors or point towards suspicious activity for investigation.
Detecting a fake review: a challenging task
Our Senior Data Science Manager, Josh Meek says: “The difference between an authentic and fake review is not always clear for someone outside of Amazon to spot. For example, a product might accumulate reviews quickly because a seller invested in advertising or is offering a great product at the right price. Or, a customer may think a review is fake because it includes poor grammar.”
This is where some of our critics get fake review detection wrong—they have to make big assumptions without having access to data signals that indicate patterns of abuse. The combination of advanced technology and proprietary data helps us identify fake reviews more accurately by going beyond the surface level indicators of abuse to identify deeper relationships between bad actors.
“Maintaining a trustworthy shopping experience is our top priority,” says Rebecca Mond, Head of External Relations, Trustworthy Reviews at Amazon. “We continue to invent new ways to improve and stop fake reviews from entering our store and protect our customers so they can shop with confidence.”
Learn more about our efforts to combat fake reviews here.
