Amazon tests AI to answer shoppers' product questions

If you have ever spent twenty minutes scrolling through hundreds of Amazon reviews just to figure out if a toaster actually toasts evenly, Amazon has a new fix.

The e-commerce giant is rolling out a mobile AI tool that answers specific questions about individual products, as confirmed by Amazon spokesperson Maria Boschetti after researchers at Marketplace Pulse first spotted the test.

Fast summaries with tight guardrails

Built directly into Amazon's mobile app, the feature prompts shoppers to ask questions about the product they are currently viewing.

Within seconds, the tool generates a concise response by scanning and summarizing both the official listing details and the customer review section.

Unlike general-purpose chatbots like ChatGPT, this tool is strictly bounded to the item on the page. It will not hold casual conversations. If you try asking it something off-topic like "Who is Jeff Bezos?", it simply returns an error message.

Yet it still retains some playful LLM quirks. In tests on a women's vest listing, it happily wrote a custom haiku about the garment and even described the product in the voice of Yoda from Star Wars.

The test follows several other AI additions to Amazon's platform over recent months, including automated review summaries launched last June, listing-generation tools for third-party sellers, and enterprise offerings like Amazon Q and Bedrock.

CEO Andy Jassy also noted in Amazon's recent earnings call that the company uses generative models behind the scenes to forecast inventory and map last-mile driver routes.

Why it matters

We think Amazon's approach here is remarkably smart because of what it refuses to do.

Instead of building a sprawling personal shopping assistant that tries to guide your whole life, Amazon constrained the model to a single, high-friction bottleneck: parsing messy customer feedback on a specific product page.

By locking down open-ended chat and slapping an error screen on irrelevant queries, Amazon limits expensive compute costs and significantly lowers hallucination risks right where buying decisions happen.

The real test will be review authenticity. If an item is already flooded with fake positive feedback, an AI summary simply turns garbage data into an authoritative-sounding pitch.

Still, if Amazon can keep the summaries honest, this could quickly kill the traditional chore of review-scrolling for millions of mobile shoppers.