Last week, Pinterest launched an in-app generative AI feature, Pinterest Assistant, designed as a visual shopping aid that responds to users’ voice queries with a brief audio response and personalized product results. Meant to replicate how one might shop IRL or ask a friend for suggestions, the assistant draws from user input such as saves, pins and boards, as well as data from those with similar taste among the nearly 600 million monthly active users on the app, to tailor its visual suggestions.
The intent for the assistant is to guide discovery through casual and conversational searches, avoiding the mental block that can come with trying to articulate a specific vision or aesthetic. The assistant expands Pinterest’s AI capabilities by integrating image, voice and text, a multimodal approach it claims achieves over 30% higher relevance than generic models.
This is the latest move in the app’s journey to become an AI-enabled ecommerce platform. Over the summer, the company launched auto-collages for advertisers on Pinterest Ad Labs and announced the creation of its “proprietary generative retrieval model” on its Q2 earnings call, presumably the foundation for Pinterest Assistant. Earlier last week, the app added new tabs to user boards based on their saved pins, including “Make it yours” and “More ideas” with personalized, AI-surfaced suggestions.
“Pinterest Assistant represents the next evolution of our AI capabilities … a conversational experience where users can ask open-ended questions about Pins, boards, or their feed using voice input,” a spokesperson for the brand told Campaign. “Unlike text-based chatbots that return walls of links, Pinterest Assistant maintains [our] signature visual gallery experience.”
The AI feature, still in beta, has rolled out to users 18 and over in the U.S. and is expected to expand in the near future.
The data of ‘taste’
Pinterest’s comparison of style and aesthetic between users is made possible by its Taste Graph, a compilation of first-party data analysis collected from user activity on the app from at least the past decade. Pinterest claims this is what powers its AI recommendations and ad impressions and allows for the translation of voice input into visual results.
The Taste Graph, with a foundation of 500 billion pins grouped into 10 billion boards (and counting), “is powered by two key breakthroughs: production-grade engineering that runs large-scale vision-language models end-to-end on Pinterest content, and a learning loop fed by hundreds of billions of on-platform actions like saves, clicks, and searches,” the spokesperson said. “Pinterest Assistant taps into that same system.”
‘Thinking in images’
It’s well documented that Pinterest is popular among Gen Z, and the app continues to lean in with this demographic.
“Gen Z says they ‘think in images’ far more than in sentences, yet most digital tools still communicate back through text boxes and product grids,” the Pinterest spokesperson said. “They love how Pinterest ‘gets their aesthetics and vibes’ by matching visual inspiration to concrete realizations, but they’ve been limited to typing search terms when they could better express what they want through conversation.”
This guided the development of Pinterest Assistant as a tool to surface results in real time based on spoken descriptors, granting dedicated users “the ability to express intent the way they naturally think and communicate,” the spokesperson said.
Keeping the conversation going
Pinterest Assistant also moves the platform from the realm of social commerce into conversational commerce, poised to compete against the likes of Apple’s Siri and Amazon’s Alexa. GenAI-powered chatbots, voice assistants and shopping concierges are the latest iteration brands are using to tap into naturalistic language use to create shoppable experiences for their customers.
“Consumers are getting fatigued by AI recommendations that feel generic or pushy,” the Pinterest spokesperson said, adding that Pinterest Assistant maintains a “visual-first experience rather than feeling like you’re talking to a salesperson.”
However, Pinterest, among other social platforms, struggled as the influx of AI-generated content outpaced the ability to vet it. Earlier this year, the app was criticized for allowing a flood of “AI slop” on the platform, which it addressed with labeling and giving users new controls to adjust how much AI content could show up on their feeds for certain categories.
To help reduce fake AI products from appearing in Pinterest Assistant results, the brand encourages use of its Pinterest Performance+ offering of ad campaign automation tools, “which helps ensure the products surfacing through Assistant are tied to real inventory and legitimate businesses.”
“Our recommendation system leverages the same personalization signals that make Pinterest ads effective, prioritizing content from merchants who’ve uploaded verified product catalogs,” the spokesperson said.
