Amazon uses AI to write reviews of people's product reviews

Amazon is now using generative AI to produce written summaries of customer reviews for products on the shelves of its sprawling e-commerce empire.

These machine-made digests, visible in the internet giant's mobile app, are supposed to inform shoppers about an item they're thinking of buying. Instead of scrolling through all the reviews, they can read a short paragraph generated by an Amazon AI model, which highlights common words folks have used to describe the product.

As an example, for an LG 3D Smart TV, the snippet of automatically generated text mentions it being easy to set up with excellent streaming speeds. Underneath the paragraph is a list of product attributes, such as "ease of use" and "stability," that people can click on to be directed to the relevant comments that mention these keywords.

The new feature may save shoppers time researching products online, though of course, the AI-generated summaries may not be trustworthy if the reviews are fake or misleading to begin with. Negative comments could be hidden and outnumbered by false positive ones, which could, in theory, push the model to generate a description making an item seem better than it really is.

Amazon is aware of the issue, and said it is trying to crack down on fake reviews to avoid misleading customers.

• Uncle Sam cracks down on faked reviews and bad influencers
• Amazon sues 10,000 Facebook Group admins for offering fake reviews
• No reliable way to detect AI-generated text, boffins sigh

"We welcome authentic reviews—whether positive or negative—but strictly prohibit fake reviews that intentionally mislead customers by providing information that is not impartial, authentic, or intended for that product or service," Vaughn Schermerhorn, Director, Community Shopping at Amazon, said today. 

Amazon said it already uses machine-learning algorithms to detect whether a review looks suspicious. The software reportedly analyses "thousands of data points," looking at the poster's account, review history, and activity to assess their authenticity. The web titan also employs a team of human moderators to oversee content. 

"The new AI-generated review highlights use only our trusted review corpus from verified purchases, ensuring that customers can easily understand the community's opinions at a glance," Schermerhorn said.

Fake reviews, however, continue to plague the site and AI has made it easier for bots to increase the volume of trash. Some comments on Amazon, previously spotted by CNBC, even include the words: "As an AI language model," which is often used in responses produced by ChatGPT. A document from the US Federal Trade Commission requesting public comment examining customer reviews online cited a report estimating that 27.6 percent of reviews on Amazon were fake.

Last year, 125 million customers posted a total of 1.5 billion reviews and ratings to Amazon stores, the mega-corp said. ®