Searching on Amazon for a trainer to convert my road bike into a stationary cycle, I knew the make and model of my bike, but had no clue about its wheel size, or whether it would be compatible with the products I was considering.
So I pulled up from the bottom of my screen in the Amazon app, and called on Rufus for help: “Can you recommend a trainer for my bike, which is a Raleigh cadent 2?”
Rufus responded, “The Raleigh Cadent 2 is a versatile hybrid bike that can handle both road and light off-road riding. When choosing a bike trainer for this bike, look for a wheel-on trainer that is compatible with its 700c wheels and quick-release axles,” suggesting several compatible products.
That was very helpful, and accurate, as I learned when I double-checked the wheel size with a Google search. Not only did Rufus know the wheel size, based on just the name of the bike, but it was able to use that information to identify compatible trainers.
That was one of the high points when testing the beta of Amazon’s new AI shopping assistant on a wide variety of queries over the past few days. Amazon announced Rufus in early February, and GeekWire requested access to test and report on the experience.
I tried a wide range of queries, focusing primarily practical questions about products. For example, a chat thread about stick vacuum cleaners garnered lots of relevant details and product links.
I also attempted to troll the AI by asking questions such as, “Which is better, Mac or PC?” Rufus handled that one like a pro, as you can see in the image above.
Can Rufus help you navigate your Amazon account? Yes, a little. For example, I asked, “When does my next order arrive?” hoping that I might get a direct answer in the chat. Instead, Rufus said, in part, “You can see the latest time a shipment can be expected and track packages in Your Orders.” However, it did also provide a link directly to my orders page at the end of its response. This can be hard to remember how to find sometimes, so I found the link helpful, at least.
The low point came later in the AI chat session about bike trainers, when Rufus responded to my follow-up question about the compatibility of a different group of bike trainers with this reply, “Unfortunately, I do not have enough information about your specific bike to definitively say if all of these options are compatible.”
I reminded Rufus about the make and model, just in case it had forgotten, but the AI was not able to repeat what it had done earlier, inexplicably reiterating that it needed to know the bike’s specs to check the compatibility.
Chalk it up to AI amnesia.
Did I mention that this is a beta? These kinds of inconsistencies are familiar to those testing and using some of the latest AI services. Amazon is still working out the kinks, and asking for feedback from early users in the form of thumbs up or down on specific responses, and the ability to provide more detailed input.
Overall, the experience has left me impressed with the potential for AI to make it easier to be a savvy shopper.
How it works: Rufus can be activated by asking a question in the regular search box in the Amazon app, or by pulling the tab up from the bottom of the app to manually activate the AI chat interface.
Voice commands: You can use your phone’s microphone to input text, but Rufus doesn’t speak back yet. At this point in the beta, the answers come back as text.
Product search links: When suggesting a product, Rufus generally links not to the product page but instead to the search results page for that product’s name. It’s probably not a coincidence that this approach ensures sponsored listings aren’t bypassed.
Where the answers come from: Amazon says Rufus is based on a large language model trained on data including Amazon’s product catalog, customer reviews, community Q&As and information from the web.
Technical details: Rufus uses a variety of core AWS services, including compute and storage, and other infrastructure services. It runs on AWS-designed Inferentia chips, which are built for high performance and low cost.
Data retention: Recent chat history can be accessed by scrolling up in the Rufus menu in full-screen mode. Deeper history is accessible via a menu, with the option to erase the full history. In the beta, Rufus is one long running chat. It would be useful to save individual chat sessions about specific products. Hopefully Amazon is working on this.
Product pages: Rufus also offers the ability to ask a question in the context of a specific product page. For example, while on the listing for a particular type of cat food, I asked if it would be good for kittens, and got a useful response.
Availability: Amazon says Rufus will roll out in waves, starting with “a small subset” of its U.S. customers. It’s available on the Amazon app for iPhone and Android, not yet on the web. Accessing Rufus via desktop web browser would be better, in many ways, for deeper product research. Hopefully this is also on Amazon’s to-do list.