Starbucks has begun testing a ChatGPT-powered ordering experience that turns a customer’s “vibe” — or a photo of their outfit — into a drink suggestion and routes the selection into a purchase completed in the Starbucks app or on the company’s website. A beta is available to customers in the US as of Wednesday, the company’s rollout indicated.
Users can ask for ideas that match a mood or a scene, specify preferences like protein or sugar-free options, and choose a store location from within the ChatGPT interface before finalizing their order, the company’s senior vice president for digital and loyalty, Paul Riedel, said as the feature debuted.
“Over the past year, one thing has become clear: customers aren’t always starting with a menu. They’re starting with a feeling,” Riedel said. He called the beta “an opportunity for us to listen, learn, and refine as we go,” and added, “This is only the beginning,” according to Business Insider.
Agentic commerce
The test positions Starbucks among the first large food brands to plug generative AI directly into discovery and ordering. It is part of a wider shift toward “agentic commerce,” in which chatbots guide users from discovery to checkout across retail and delivery platforms. Several major brands have already integrated ChatGPT into shopping flows — including Walmart and Target for product discovery and purchases — while operators like DoorDash and Uber Eats let users turn recipes into shoppable lists or browse menus and place delivery orders. Fast-food chains such as Burger King and Firehouse Subs help customers find nearby locations and deals through the chatbot, according to Business Insider.
Starbucks has also been rolling out Green Dot Assist, an AI virtual assistant for baristas. It was piloted at 35 locations before broader deployment this year, indicating a parallel focus on tools for employees alongside customer-facing AI.
The system correlates opt-in mood inputs with historical purchases, weather, time of day, and regional patterns. An engineering push yielded a 22% rise in average transaction value during pilot tests in Seattle and Austin, according to internal metrics leaked to The Verge.
Baristas in pilot stores have been using the AI suggestions as conversational openers rather than rigid scripts. Internal surveys indicated that most felt the tool supported their ability to connect with customers.
Latency is a key operational constraint. A 1.8-second pause between mood selection and suggestion adds up to a significant annual time cost at global scale. Engineers are experimenting with edge caching and quantized model distillation to reduce delays.
Can't you decide on a coffee?
Early usage shows both flexibility and limits. In a test that blended outfit details, warm weather, a walk in Central Park, and a request for a summery, polished-but-not-too-sweet drink with non-dairy milk, the tool pointed to a Cold Brew with almond milk as a fit. The system also sometimes surfaced unconventional pairings—like espresso with lemonade.
Critics question the need for a chatbot to choose a latte or cold brew. They argue that if customers struggle to recall drink names, the menu may lack clarity. Some critics also warn about dependence on AI for small, daily decisions, suggesting that if indecision is so acute that a chatbot must settle a coffee order, the issue may not be technological.
A 2025 study found that algorithmic suggestions on e-commerce platforms tended to produce more negative feelings, with many shoppers reporting they felt stuck in information cocoons; separate survey data indicates only 39% of consumers in the USA currently trust AI to make everyday purchases on their behalf.
Starbucks framed its personalization strategy as especially relevant to Gen Z audiences that gravitate toward tailored experiences and like to engage with brands early in the decision process.