Amazon is testing AI summaries for customer reviews

Scrolling through hundreds of product reviews before buying anything has become an exhausting online ritual. Amazon thinks large language models can finally kill that chore.

As reported by CNBC, the e-commerce giant is quietly testing a new generative AI feature in its mobile shopping app that summarizes customer feedback left on product listings. Instead of making you dig through pages of conflicting opinions, the tool provides a concise snapshot of what people loved and hated about an item.

Mark Wieczorek, chief technology officer at Amazon marketing agency Fortress Brand, first spotted the feature in the wild. Amazon subsequently confirmed the test, though it declined to share which AI models power the summaries or how the underlying pipeline works.

Pulling no punches on bad quality

What stands out most is that the summaries do not read like marketing fluff. Each summary comes with a clear disclaimer that it is AI-generated from customer review text, and it is surprisingly blunt about product flaws.

In one mobile listing for a children's cauldron toy, the summary acknowledged positive mentions around fun factor and appearance, but immediately followed up by noting that the majority of customers expressed negative opinions on those exact same aspects. It specifically cited buyers who felt the toy was not worth its over-$100 price tag, alongside frequent complaints about quality defects, charging failures, and leaks.

Amazon spokesperson Lindsay Shanahan noted that the company is significantly investing in generative AI across all of its businesses. The move aligns with comments from Amazon CEO Andy Jassy, who said in April that generative AI and foundation models stand to transform virtually every customer experience.

The garbage in, garbage out dilemma

While summarizing reviews is an obvious win for shopper convenience, it runs directly into Amazon's oldest headache: review credibility.

Amazon has spent years battling fake and paid reviews using a mix of automated tools and human moderators. The retailer previously introduced a one-tap ratings system to encourage feedback without writing, but that diluted written nuance. More recently, shady sellers and lazy shoppers have started using chatbots like ChatGPT to generate fake reviews.

If the training data feeding the summary consists of bot-written praise or targeted review-bombing, the AI risks synthesizing noise instead of signal. Amazon will need rock-solid spam filtering before these summaries can be trusted at scale.

Why it matters

We think this is one of the most practical, high-impact applications of generative AI in consumer tech so far. For everyday shoppers, getting an unbiased digest of a product's chronic failure points in five seconds completely changes how we buy things online, especially for complex electronics or expensive toys.

By prominently highlighting negative feedback rather than burying it, Amazon seems willing to prioritize shopper trust over vendor appeasement. This is going to put immense pressure on rival marketplaces like Walmart and eBay to ship similar automated review digests, while forcing merchants to fix genuine product flaws instead of masking them behind review volume.

However, the technical risk cannot be ignored. Summarization engines do not inherently know truth; they only know frequency and phrasing. If Amazon cannot successfully purge bot-generated reviews before feeding them to the summarizer, it risks turning automated manipulation into official shopping advice.