With AI platforms like Perplexity adding new ways to shop with generative AI, there’s a growing need for AI to justify its recommendations — just like reviews from publishers and content creators.
This week, Perplexity became the latest AI search platform to debut new shopping assistant tools to help people search and buy products. The platform, “Buy With Pro,” offers product recommendations based on text-based chats along with a “Snap To Shop” visual search similar to Google Lens. Meanwhile, a new program for merchants offers an increased chance of being featured in recommendations, payment integrations, free API access to build Perplexity search into retail websites, a dashboard with search and shopping trends.
Arriving just a week after Perplexity introduced ads, “Buy With Pro” is just one of the ways major AI players are more deeply integrating generative AI into how people find products — and how companies market them. Others include Google’s new updates for Google Shopping, Amazon’s chat-based AI shopping assistant Rufus, eBay’s “Shop The Look” and Klarna’s AI shopping assistant.
The convergence of AI and commerce also raises new opportunities and challenges for marketers looking to reach shoppers in the new AI search era. For example, are AI platforms choosing product recommendations based on news articles, branded content or customer reviews on e-commerce sites?
Generative AI is reshaping the concept of search beyond traditional platforms to areas like social networks, shopping apps and retail platforms. Many of these are focused on discovery, but AI-generated product recommendations might require explaining to users why it’s recommending an item. That’s where citations play an important role in helping users know where the information came from — and in helping companies analyze which websites are showing up in results about their products or categories.
How does AI explainability work?
Much of this relates to the topic of AI explainability, which addresses the ways AI models explain how and why they made certain decisions — consider it a form of transparency. AI models prioritize features using weights to tailor product recommendations based on user preferences. For example, when someone asks for the “best laptop” the model evaluates factors like performance, brand reputation, price, and user data to rank items aligned with user’s query. The credibility of sources is also weighted. For example, it might give more weight to trusted reviews from Wirecutter over branded content. But can AI consistently differentiate between credible and biased information?
Another question for the future role of generative AI in shopping could include how Perplexity and others plan to compensate the websites used in making its product recommendations — especially if it siphons revenue away from traditional affiliate models. Perplexity’s partnerships with publishers include an advertising revenue-share model, but it’s still unclear how much money that could provide.
In the context of e-commerce product recommendations, AI explainability could help build trust by helping users see the rationale behind suggestions. That in turn could improve transparency, accuracy, and fairness for shoppers and for the companies whose products are being mentioned.
But there’s a catch: AI models aren’t always good at explaining where answers come from. Some say Perplexity’s citations help users get a sense if they should trust an answer from a familiar source or if they should be more skeptical. That’s something Chad Stoller, UM’s chief innovation officer Chad pointed out last week when discussing Perplexity’s ads roll-out.
“If I’ve never heard of these sites, maybe my skepticism level is going to be a little bit higher on this result,” Stoller said. “That’s a good thing… I think that’s going to be the norm and it has to be, because people have to know how the sausage is being made.”
AI personalization
While generative AI content has begun to scale, personalization hasn’t — at least not yet. In a recent survey of 5,000 global consumers conducted by Boston Consulting Group, more than 80% said they want and expect personalized experiences. However, two-thirds mentioned experiencing inaccurate or inappropriate personalization.
There’s also a chance users who share information willingly with chat-based search platforms might be more receptive to having their data used for personalization, according to Mark Abraham, who leads Boston Consulting Group’s Marketing, Sales & Pricing practice. Retailers are also exploring ways to make generative search more engaging, helpful or more exciting.
Marketers might soon have to market to both humans and AI models. Abraham — who recently wrote a book about personalization in the AI era — said some large advertisers are starting to place aggressive bets on preparing for the future of generative search. One example Abraham gave was an undisclosed large CPG that is planning for one-third of its marketing spend to potentially be directed at marketing to AI agents in just a few years.
“There is a world where the platforms themselves — the Geminis and the Perplexity’s of the world — become curation points for discovery,” Abraham said. “Virtual shopping assistants and brands will have to position themselves and pay for influence on the rankings.”
How AI could evolve recommendations
Some of these changes could lead to a shift from AI-powered search to “search-powered AI,” said Keri Rich, vp of product at LucidWorks, which provides companies with AI search tools. Ads also will have to become more sophisticated to keep up with how they’re triggered in a chat-based environment. As a result, AI experience design could become more popular as UX designers adapt to how people use generative search for their shopping. That could also mean back-end updates with prompts asking users follow-up questions to better understand what they’re shopping for, what they want to buy and what they’ll actually buy.
“The Perplexities and others are going to need to have that sophistication to their weighting [and how they surface answers],” Rich said. “Because users don’t want to see the same product over and over again. They want to see what’s new and trending. The cool thing for them is that because you’re having a conversation with the user, you can understand where the user is at … Are they a user who really likes what’s trending?”