Instacart Launches Its Own AI Assistant Clementine While Opening Its Shelves to Rival Agents

Two moves in opposite directions, announced the same day

Instacart launches AI assistant Clementine, agentic commerce integrations

Instacart launched an AI assistant, Clementine, for grocery shopping; it also added Cart Assistant partners and agentic commerce integrations.

www.digitalcommerce360.com

Instacart (NASDAQ: CART) announced Clementine, its AI-powered grocery shopping assistant, on September 9, 2026. Most U.S. and Canadian customers can use it from that day through the app and the website: the Clementine icon sits above the search bar on mobile, and at the top of the left-hand navigation on desktop.

What makes the release worth reading twice is that the same document carries a second message. More grocers are now live with Cart Assistant, the white-label version that retailers run on their own sites and apps, and conversations that begin on ChatGPT, Claude, Gemini or AI Mode in Google Search can now be completed on Instacart.

A move that strengthens Instacart's own front door and a move that hands its shelves to someone else's front door were published on the same day, in the same press release. That combination is the real story here.

Clementine is betting on the decision, not the search

Clementine is not aimed at product search. It targets the decisions that happen before a purchase: what to cook tonight, how to shop the way a family actually eats, and how to stay inside a budget. A message such as "I need a week of budget-friendly kid lunches," "high-protein easy dinners for two," or "order my usuals" becomes a ready-to-buy cart in seconds.

Instacart cites a Harris Poll survey conducted June 2 to 4, 2026 among 2,059 U.S. adults aged 18 and older. More than 80% said deciding what to make for dinner causes them stress, and nearly a third said the hardest part is not the cooking but the deciding. More than three in five said they would be interested in an AI assistant that could help.

The differentiation argument rests on inventory data

As a feature set, a generative AI shopping assistant is no longer novel. TechCrunch noted in the same news cycle that Uber Eats and DoorDash have already shipped comparable assistants that turn natural-language requests and grocery lists into carts.

What Instacart put forward as its differentiator is not the feature list but the data layer underneath: more than 1.6 billion lifetime orders, a catalog of more than 2 billion items, more than 2,200 retail banners across North America, and more than 10 million daily inventory signals from nearly 100,000 stores. That foundation is what lets Clementine build a cart from what is physically on the shelf at the customer's chosen store, rather than from what is generally recommended.

The quality of a recipe suggestion matters far less than whether the suggested items can actually be bought. In a world where agents execute the purchase, the second question decides the experience. Instacart's July acquisition of computer vision company Arpalus, which sharpened its read on real shelf conditions, sits on the same line.

Instacart also reports a behavioral progression among early users. Customers start with simple tasks such as restocking, then graduate to higher-order use cases such as recipe discovery. Orders placed through Clementine, the company says, exceed its stated $115 average basket size.

How far does the opening to rival agents actually go?

The integrations announced alongside Clementine differ in character from one surface to the next. The sharpest difference is where checkout happens.

The deepest integration is with ChatGPT. Instacart's ChatGPT app, launched on December 8, 2025, is built on the Agentic Commerce Protocol (ACP), and Instant Checkout completes payment without leaving the conversation. Stripe processes the transaction, and wallets such as Apple Pay and Google Pay are supported. At the time, Instacart described itself as the first app in ChatGPT's ecosystem to offer in-conversation checkout.

The Claude and Google integrations currently work differently. In Claude, the cart is assembled in conversation and syncs with the customer's Instacart account, so the order itself is placed on Instacart. The AI Mode integration in Google Search, published July 16, 2026, follows the same pattern: checkout happens in the Instacart app or on the web, with Instacart+ benefits and discounts applied. It is available to U.S. users 18 and older on web and mobile web.

In other words, "opening up to third-party agents" covers two very different arrangements: handing over cart creation, and handing over checkout as well. That distinction is what determines your negotiating position.

What Cart Assistant sells is a position, not a technology

Cart Assistant is the white-label product that lets a grocer run a branded agentic shopping experience on its own site and app. It combines Instacart's grocery intelligence with the retailer's own catalog and customer data, and grocers can customize the name, tone, loyalty integration and recipe content.

Food Bazaar, Heritage Grocers Group and Woodman's are live today, with ALDI U.S., Harmon's, The Save Mart Companies and Stew Leonard's named as partners to follow. According to PYMNTS, Instacart said on its May earnings call that Cart Assistant was available to roughly 25% of U.S. customers, so this announcement marks the move into named retailer deployments.

What the product sells to a grocer is not the conversational UI. It is the ability to own an agent experience without letting go of the customer relationship. Seen from Instacart's side, the more retailers strengthen AI-assisted shopping on their own sites, the more the grocery intelligence running underneath belongs to Instacart.

What the announcement does not answer

Read only through the numbers the company chose to publish, this looks like a clean success story. Several material terms, however, were not disclosed.

Pricing and revenue sharing for Cart Assistant are not disclosed. Fees or referral terms for the OpenAI, Anthropic and Google integrations are also not disclosed, so how agent-originated orders affect Instacart's take is unknown. Neither Clementine's adoption rate nor the share of North American users actively using it has been published.

The basket-size comparison deserves care as well. Saying that Clementine orders exceed $115 is not proof that Clementine raised order value. Customers who try a new feature early tend to be heavier users to begin with, and no causal analysis separating that selection effect has been offered. In its Q2 2026 results, the company did not quantify the assistant's contribution to quarterly performance.

The external environment carries its own caveats. PYMNTS Intelligence found that 64% of shoppers expect to shop with AI agents at least occasionally within two years, and 30% expect to do so frequently or almost always. Interest is high, yet agent-originated traffic still accounts for a small share of overall e-commerce today, a point covered in our analysis of Q2 AI referral sources.

How e-commerce operators should read this

Translated to your own business, Instacart's moves resolve into three options.

The first is running a conversational buying experience in your own UI. What Clementine demonstrates is that the value of that experience comes from the accuracy of the inventory and purchase history behind it, not from the elegance of the dialogue. Bolting a chat interface onto unstructured real-time inventory, pricing and promotion data produces the worst possible outcome: recommending items the customer cannot actually buy.

The second is opening your catalog to outside agents. The question is not whether to open, but how far. Handing over cart creation is one thing; handing over checkout is another. Full checkout handoff minimizes friction, but customer data and the relationship shift toward the agent. Instacart draws that line differently on each surface rather than applying a single policy.

The third is becoming the supplier of inventory and product intelligence to others, which is what Cart Assistant is. That path is unrealistic for most operators, but the underlying exercise, taking inventory of which of your own data is genuinely hard for others to reproduce, is worth doing. For Instacart the answer was 1.6 billion orders and 10 million daily inventory signals.

The common thread is that the asset worth defending in the agent era has moved from the UI and the branded site experience to the freshness and accuracy of the data underneath.

Conclusion

Announcing a first-party assistant and third-party openness on the same day reads as a judgment that neither bet can be made alone. Where consumers will start their shopping has not been settled.

Two things are worth watching next. Whether Cart Assistant deployments actually reach large chains such as ALDI U.S., and whether the Claude and Google integrations move to in-conversation checkout the way ChatGPT already has. If the second happens, the industry-wide line on where payment occurs will shift. The day agent-originated order share is actually disclosed will be the moment this strategy gets graded.