Instacart launches Clementine: Azoma on what agent-built baskets mean for CPG brand visibility

Instacart launched Clementine on 9 September, an AI assistant that turns a message, a recipe or a photo of a handwritten list into a ready-to-buy cart. It is live to most customers across the US and Canada, and it is grounded in what is actually on the shelf at the shopper's chosen store rather than in what exists in a catalogue somewhere.

A shopper types "high-protein easy dinners for two" or "order my usuals" and gets back a full basket. For a CPG brand, that basket is a set of decisions made without a shopper comparing anything.

What Instacart actually shipped

1. An assistant built on grocery data nobody else has.

Clementine runs on more than 1.6 billion lifetime orders, a catalogue of more than 2 billion items, and more than 10 million daily inventory signals from nearly 100,000 stores across more than 2,200 retail banners. Those figures are stated as of 31 December 2025, so the current numbers are larger.

The inventory layer is what makes it different from a general assistant. It knows what is on the shelf at your store right now, what is on sale there, and what that household buys week after week.

2. Cart Assistant on grocers' own storefronts.

Instacart's white-label version lets a grocer run the same technology under its own name, tone, loyalty integration and recipe content, using its own catalogue and customer data. Food Bazaar, Heritage Grocers Group and Woodman's are live, with ALDI U.S., Harmon's, The Save Mart Companies and Stew Leonard's to follow.

Every regional grocer can now have its own Sparky. For brands, that means one engine's logic is being applied across a long tail of retailers who could never have built this themselves.

3. Handoff from the general assistants.

Instacart has partnered with Anthropic, OpenAI and Google so grocery conversations starting in Claude, ChatGPT, Gemini or AI Mode in Google Search can be completed on Instacart, with the order placed at the shopper's preferred retailer.

This is consolidation, not entry

Coverage is framing Clementine as Instacart joining the agentic commerce race. The record says otherwise, and the difference matters for planning.

Instacart announced enterprise AI including Cart Assistant in November 2025, launched a ChatGPT app with embedded grocery checkout in December 2025, and told its May earnings call that Cart Assistant already reached roughly a quarter of US customers. Clementine is the consumer-facing front end on a stack that has been running for most of a year.

What is genuinely new is the completeness of the coverage. The same recommendation engine now decides the basket in three different places at once.

Three front doors, one engine behind them. A CPG brand that is invisible to Instacart's recommendation logic is invisible in all three.

The basket is the new shelf

Every agent story this year has been about being named in an answer. A grocery basket is a different problem, because thirty items get selected at once and the shopper is shown a result rather than a comparison.

Instacart says orders placed with Clementine contain more items and exceed the company's $115 average basket size, a figure stated as of 30 June 2026. It also says customers start with simple tasks like restocking and then graduate to recipe discovery.

That progression is the part worth reading twice. Restocking rewards whoever is already in the purchase history, and recipe discovery is where a brand with no history can enter, provided the agent can match it to the ingredient and find it in stock at that store.

What CPG brands should do now

Each item maps to one of the 5 C's of Agentic Commerce.

1. Get the attributes an agent filters on into structured data. (Completeness)

Clementine applies dietary and product preferences a household sets once, across everything that follows. Gluten-free, vegetarian, nut-free and organic are filters before they are claims, and a product missing the attribute is absent from the result rather than ranked lower in it.

2. Treat store-level availability as a visibility metric. (Correctness)

The assistant is grounded in real-time shelf availability at the chosen store, so an out-of-stock SKU is an invisible SKU. Availability has always been a supply chain number. It is now also a share-of-recommendation number, and the two teams reporting on it rarely talk.

3. Win the ingredient, not only the category. (Context)

Recipe-led discovery selects against an ingredient in a method, not against a category browse. A product described as what it is competes poorly with one described in terms of what it is used to make.

4. Keep earning the sources the general assistants read. (Citations)

A meaningful share of these baskets will begin as a conversation in Claude, ChatGPT, Gemini or AI Mode before the handoff to Instacart. Whatever shapes the recommendation in those tools shapes the basket that arrives.

5. Understand what a first purchase is now worth. (Customer acquisition)

When "restock my usuals" pulls from order history, the first time a household buys your product it enters a list that gets reordered without a decision. Trial has always mattered. In an agent-mediated basket it compounds, because the competitor never gets shown again.

➡️ Azoma measures how AI shopping agents describe and recommend products, at ASIN and SKU level, with the prompts and citation sources behind every mention.

What this does not mean

It does not mean the basket data is public. Instacart has published a direction of travel, not a breakdown. Nobody outside the company can currently see which brands Clementine selects in a category, which is a measurement gap rather than a settled advantage for anyone.

It does not mean paid placement stops working. Instacart runs a substantial ads business, and nothing in this announcement suggests sponsored positions are going away. What changes is that a basket assembled by an agent has fewer impressions in it than a shopper browsing an aisle.

It does not mean grocery is a special case. The same pattern arrived on Amazon with Alexa for Shopping, and on Meta last week with Muse. Grocery is simply further along, because repeat purchase and real-time inventory make the agent more useful there sooner.

Important caveat: the basket figures come from Instacart's own early data in its launch announcement. Early adopters of any new feature tend to be heavier users, so a higher basket size among Clementine orders reflects who is using it as much as what it does. Treat it as directional until there is a full quarter behind it.

Summing Up

Two agent surfaces launched in eight days. Meta put a browser and a payment card inside WhatsApp, and Instacart made the basket itself the unit of recommendation across three different front doors.

The surface list keeps growing and the principles do not. Complete attributes, accurate availability, content that matches how people actually ask, citations in the sources the assistants read, and an understanding of what a first purchase is now worth.

For CPG brands the uncomfortable part is the arithmetic. A shopper browsing an aisle sees your pack next to the competitor's. A shopper reading a finished basket sees whichever one the agent chose, and never learns the other existed.

Azoma is built for exactly this. We measure how AI shopping agents retrieve, compare and recommend products across Alexa for Shopping, Walmart Sparky, ChatGPT and Gemini, and fix the sources those answers are built from. Get in touch today for an assessment of how your catalogue reads to an agent.