In search of the perfect gifts this holiday season? Artificial intelligence chatbots may offer some assistance, albeit with specific limitations.
As Cyber Monday approaches, many shoppers will likely encounter interactive chatbots developed by various retailers and e-commerce platforms aimed at enhancing the shopping experience. These shopping aids, which include advanced generative AI technologies, allow users to pose questions in a conversational manner, such as, “What’s the best wireless speaker?”
Retailers envision these chatbots, often referred to as shopping assistants, as virtual companions designed to help customers explore or compare products. Previously, chatbots primarily facilitated straightforward tasks, like tracking online orders or managing returns.
Amazon, the leader in online retail, has introduced Rufus, an AI-driven shopping assistant. Customers have engaged with Rufus, asking questions about the maintenance of specific coffee makers or seeking suggestions for birthday party games for children. Rufus is currently accessible to holiday shoppers in the U.S. and select other countries. Meanwhile, Walmart is also testing a similar AI chatbot, though it will be limited to specific product categories such as toys and electronics for a select number of customers.
In a recent development, Perplexity AI launched a feature on its AI search engine, enabling users to query for products like “the best women’s leather boots” and receive direct results that the company claims are not influenced by sponsorships. According to Mike Mallazzo, an analyst at Future Commerce, the uptake of these technologies has been remarkable.
The rise of AI chatbots in retail gained momentum following the popularity of ChatGPT, a text-based AI developed by OpenAI, which sparked widespread interest in generative AI technology. Companies like Victoria’s Secret, IKEA, Instacart, and Canada’s Ssense are among those experimenting with AI chat solutions, some leveraging technology from OpenAI.
Even before the advent of sophisticated chatbots, online shops were curating product recommendations based on individual shopping history and prior purchases. Amazon was particularly proactive in this regard, making Rufus’ capacity for product suggestions less revolutionary. Rajiv Mehta, Amazon’s VP of search and conversational shopping, highlighted that Rufus is programmed to ask clarifying or follow-up questions to enhance its recommendations, with shoppers also utilizing it to discover personalized deals.
However, there are drawbacks, as chatbots can produce inaccurate information, described as “hallucinations.” Juozas Kaziukenas, founder of Marketplace Pulse, remarked that in their testing of Rufus, the assistant suggested products unsuitable for gaming TVs when asked for recommendations. Similarly, when asked for budget-friendly options, Rufus did not respond with the most economical choices.
In a separate inquiry, Rufus provided suggestions for gifts for a brother that included a T-shirt and engraved multifunctional knife. Although it later refined its suggestions to include soccer jerseys, it failed to identify which seller offered the best price. When prompted to find a price comparison on a popular skin serum, Rufus inaccurately provided the product’s price before discounts.
“Rufus is always learning,” Mehta asserted in a recent interview. Similarly, Shop AI, introduced by Shopify last year, supports users in discovering products by asking relevant questions about features or recipient preferences. Nonetheless, Shop AI struggles with pinpointing specific items or determining the lowest price in its category.
These shortcomings underline that the technology remains in its early stages and has substantial progress to make before attaining the functionality that consumers and retailers desire. According to a McKinsey & Company report, for shopping assistants to genuinely revolutionize the user experience, they must be deeply personalized, with the capability to remember customer order histories, preferences, and purchasing habits.
Amazon has indicated that Rufus’ insights originate from product listings, customer reviews, and community Q&As—this includes potentially misleading reviews that can skew product visibility on the platform. The AI model powering Rufus was trained on the complete product catalog and public data, as outlined by Trishul Chilimbi of Amazon in an IEEE Spectrum article.
However, how retailers balance various training factors, such as customer reviews, in their recommendations remains uncertain, as noted by analyst Nicole Greene from Gartner. Meanwhile, Perplexity AI has instigated a new shopping feature that allows users to search for items like “best phone case,” compiling results from different retailers. They have encouraged retailers to share product data for better recommendation algorithms.
Nonetheless, Perplexity’s CEO, Aravind Srinivas, acknowledged uncertainty in how products are recommended, a sentiment echoed by the company’s Chief Business Officer, Dmitry Shevelenko. He clarified that generative AI outputs cannot always be predicted based solely on training input data.
Shevelenko emphasized that for products to gain visibility in Perplexity’s search results, retailers must demonstrate superior product quality rather than rely on keyword manipulation for improved search rankings.