Amazon's AI shopping assistant Rufus is often wrong

Testing is showing that Amazon's new AI shopping assistant is far from perfect. Amazon sellers are also raising concerns about the chatbot. Image (c) ConsumerAffairs

If it gets smarter, Rufus could make online shopping even easier

Amazon's artificial intelligence shopping assistant isn't doing a good job.

Rufus, an AI chatbot which Amazon has offered since July, regularly makes errors, doesn't recommend the products asked for and sometimes doesn't suggest products at all, according to tests of the AI shopping asssistant. 

"All AI systems suffer from hallucinations, and they are glaringly obvious in Rufus," said Juozas Kaziukėnas, founder of e-commerce data provider Marketplace Pulse, in a November blog post.

For example, Kaziukėnas said Rufus doesn't give the cheapest options when asked or will list items that aren't TVs when asked for recommended TVs for gaming.

"If asked for the best marathon running shoes, it will respond with options no runner or coach would recommend," Kaziukėnas said. "And if asked for winter gloves, it will respond with a list seemingly picked at random."

In March, The Washington Post also tested Rufus, finding that it made poor recommendations for cycling gloves and didn't recommend products at all when asked for help with composting at home, among other issues.

"In my testing over the past several days, the chatbot wasn’t a disaster," the newspaper's Shira Ovide wrote. "But I also found it mostly useless."

In response to the concerns, an Amazon spokesperson told ConsumerAffairs they encourage customers to leave additional feedback with the chatbot if an answer isn't helpful, including pressing the thumbs down button.

“It’s still early days for generative AI and the technology won’t always get it exactly right, but we care a lot about accuracy, and will keep training our AI models to continually improve the customer experience," the Amazon spokesperson said. "The AI models are continuously learning and improving, and customer feedback is the best way to improve the experience for everyone.”

What are Amazon saying about Rufus?

Amazon sellers are also complaining about Rufus.

"Rufus has check marked “bitterness” as a positive coffee attribute. Bitterness is one of the worst attributes which is hurting our sales. We have a 4.4 rating and bitterness has hardly ever been a complaint and not reflective in our reviews. This is totally unacceptable!," one seller said on Amazon's forums.

Another seller said: "I can't believe Amazon is doing this to sellers. The rufus AI is giving disparaging false info about our [product] to shoppers, while it talks glowingly about our competitors products. Thanks amazon."

Still, AI assistants like Rufus could make shopping much more convenient very soon—if they get smarter.

"The promise of AI is analyzing the more than 40,000 cycling gloves in Amazon’s catalog to save shoppers from having to click around, watch videos, and read reviews," Marketplace Pulse's Kaziukėnas said. "Once Amazon adds a buy button to the AI response, buying a product will take just one click."