Instacart has launched Clementine, a conversational assistant that turns a request like "find me simple, high-protein dinners for two" into a filled grocery cart. It is available now in the United States and Canada. Clementine writes personal recipes and pulls matching ones from vetted publishers, filters for dietary requirements including gluten-free, vegetarian, nut-free and organic, surfaces applicable discounts and promotions, and suggests cheaper alternatives when a basket runs long. Uber Eats and DoorDash already operate assistants that do much of the same.
The other prompts Instacart puts forward are deliberately ordinary: "I need cheap lunches for the kids this week" and "order what I usually buy." The assistant will also read a photograph of a handwritten shopping list, or a screenshot of a digital one, and convert it into a cart.
Chris Rogers, Instacart's chief executive, said in an emailed press release that millions of families ask every evening what to make for dinner, and that Instacart has spent nearly 15 years studying how families shop and eat. Clementine, he said, draws on that knowledge to assemble orders.
The strategic logic is stated openly in the launch: delivery services want to own the decision that happens before the order, not only the fulfilment that happens after it. That moves the competitive ground. Speed and selection still matter, but the prize now is becoming the household's default meal-planning tool. The second motive is defensive, and it is also acknowledged: keeping users inside the app so they do not take their planning to a general chatbot like ChatGPT.
Read that second motive first, because it explains the timing. Nothing about grocery shopping suddenly required a chat interface this quarter. What changed is that planning a week of dinners is already something people do in a general assistant, and every session that begins in ChatGPT and ends at somebody else's checkout is demand Instacart never sees. Clementine is a retention product dressed as a convenience product. Rogers' fifteen-years line is the honest form of the defense: a general model can propose five weeknight dinners, but it does not know what this household actually bought last month. Whether that is a moat or a speed bump depends entirely on how long that purchase history stays exclusive to Instacart, and nothing in the announcement suggests it will.
There is also a question the launch leaves alone. Clementine both surfaces discounts and proposes cheaper substitutes, which means a system is choosing which products to place in front of a shopper who has just delegated the choosing. Instacart does not say which model runs Clementine, and it does not say how a given promotion earns its place in the cart. An assistant that assembles the basket is also an assistant that decides whose brand is in it, and that mechanism is the part of the product the announcement never describes.
The pitch is that a shopper stops deciding. The exposure Instacart takes on with it is that a shopper who stops deciding also stops having a reason to prefer the app that decided for them, only the one that decides first.