Can Amazon AI voice replace customer reviews? I’m starting to try

Anyone who has been looking for product reviews on Amazon knows how valuable the experience of other shoppers is, and how easy it is to defeat the rabbit hole in customer comments, from five-star raves to one-star takedowns, sometimes hundreds of words. It says that shopping will be easier if Amazon continues to deploy AI capabilities. The AI-generated product descriptions are scraped by mixing the customer’s commentary.

The AI ​​Voice product description, called “Hear the Highlights,” relies on a large language model to script summaries based on a variety of sources, draw from Amazon’s product catalogs, customer reviews, and information from the web, and convert content into short form audio clips. The summary has been deployed over the summer to a subset of US customers with specific products, reaching all US customers as buttons in mobile shopping apps, covering over a million products.

The unique appeal of Amazon and other e-commerce options was their ability to retrieve information from real users, not just product descriptions. Of course, fake reviews have been a problem for Amazon for a long time. You still need to organize different efforts to game the system, and in recent years it has shown that the latest review writers are chatbots like ChatGPT. But real people share real and singular experiences with products – customers as sources and were an important part of informed decision-making – from clothing and shoe sizes to safety and narrower questions, it was a critical part of the learning process. Is the timer setting accurate when purchasing a new toaster?

Can AI improve that? The reviews are inherently cumbersome in terms of mass personality, but they prove that human readers are rather skilled at distilling what they need from human confusion. In a sense, the average Amazon Review Reader brain is a pretty good, big language model, skimming and choosing keywords and key information, so AI must at least do it the same way as humans when it comes to adding it to the customer experience that is as promised.

AI and cognitive overload

AI has its advantages. One is that you won’t experience cognitive overload in the face of an electric tea kettle on Amazon with over 32,000 reviews. It can integrate data, but can it give us only what we need or want to know? It can be difficult for AI to get right, and combining product catalogs, customer reviews, and web information into a single distillation may be inherently problematic. Each source of this information has its own intent.

“We’re excited to announce that we’re a great place to go,” said Ankur Edkie, CEO and co-founder of AI Voiceovers. “The key question is whether there is a way to consider the customer context as input while generating these summaries,” he said.

According to Edkie, AI values ​​find the right problem response suitability. If that is not achieved, the sense of the gimmicky can pass through doors that have been left open due to AI fatigue.

The Amazon’s Hear The Highlights Ai-Audio summary is currently the same for all users.

“AI summary needs to capture nuance and context. For example, even some negative reviews about safety can outweigh many positives in other features,” Edkie said, adding that price-sensitive summaries may not be relevant if customers focus solely on product performance.

Considering the context, the potential for dialogue between consumers and machines may be heading towards where technology is headed, in other words, the agent side of AI. AIAMON has also been actively added to AI commerce tools such as Rufus, using a shopping tool called AI. The feature, which was unfolded in the spring, is a separate feature from the main search bar on Amazon’s website.

The chat and audio summary remains within the engagement overview, but having real-time conversations with AI voice agents – asking specific questions, clarifying concerns, and gaining deeper insights from reviews, according to Edkie, shifts experiences from one-way delivery to two-way discovery, and is much more personalized. “Currently, you can interact with Rufus using text or voice input, but Rufus cannot reply. The response is text only,” he said. “But with a voice agent you can have two-way conversations with the bot that tells you,” he added.

For now, one segment of customers who may see immediate benefits are visually impaired shoppers, making accessibility an interesting aspect of the functionality, but the audio is of high quality and the content needs to be delivered accurately.

Brian Nomainville, principal of consumer research firm Feedback Group, said that by providing an audio-based alternative to visually presented information, these types of features could make shopping more accessible and translate detailed text into easily consumed audio summaries. However, this feature must be thoughtfully designed to truly benefit people with visual impairments. According to Numainville, this includes ensuring full compatibility between screen readers and keyboard navigation, providing a clear, structured, concise summary, and avoiding overly long or confusing audio presentations in audio presentations. The quality and clarity of the audio generated by AI also has a major impact on ease of use.

“The transition from diverse human reviews to AI-generated summaries may mean losing important nuances, context, and personal touch,” Numainville said.

The risk of losing unique shopper insights

The tendency for AI to focus on general themes can dilute the response even when it is distilled.

Human reviews tend to contain subtle stories and details about very specific use cases – think about the motivation to write a review first and how this will work well with shopper anxiety and decision-making processes – all of which will help make your shopping experience more personal and insightful.

“AI may overlook unique insights and niche needs that don’t match most of the answers,” Numainville said. “In addition, the ability to critically interpret reviews, such as finding bias and trusting a particular reviewer, is reduced in AI summary.”

Cutting down a lot of content to generate summary also costs you at the expense of product descriptions, customer reviews, and annotations that can distinguish between web information.

“The current implementation doesn’t make it 100% clear how much product descriptions and reviews are being used. Amazon doesn’t fully explain this mix,” says Numainville.

Amazon declined to comment and introduced CNBC to publicly available information about the feature.

“Product descriptions are usually marketing-driven and emphasize positivity, but reviews reflect the real user experience and include both the advantages and disadvantages,” says Numainville. Mixed without attributes – by definition flatten the source – it is difficult for customers to identify advertising speaking from authentic reviews, and the results are like native ads.

“If this blend occurs, it can unintentionally mislead consumers by lending de facto authority to subjective opinions and disguising promotional material as a non-discount review,” Numainville said.

Research shows that customers tend to trust Voice as a form of information delivery.

Tama Lieber, professor of internet studies at Curtin University in Australia, said:

Another concern is that “AI can weigh average and overall scores, but buyers often see a lot of negative reviews (one star),” Leaver said.

Dr. Nauman Dawalatabad, a research scientist at Zoom Communications, said in his personal opinion that technology is moving towards a better customer experience. “I’m using it as a technique that helps us make informed decisions,” he said.

If a streamlined explanation leads to impulse purchases and regret, he says, it is for the buyer, and it is no different to the way things always work – you can charge the same subtle forced fee for every marketing effort. He believes that voice-based agent AI continues to evolve, and consumers start talking (instead of typing and searching for AI agents) explaining, “get exactly what you need.”