Instacart Clementine arrived on 9 September 2026 as the company’s first consumer-facing AI grocery shopping assistant, available immediately to most customers across the United States and Canada. The pitch is disarmingly simple: type “I need a week of budget-friendly kid lunches” and get a ready-to-buy cart. TechCrunch, Engadget, Forbes, Modern Retail and the WSJ all ran it the same day.
So we read every word Instacart published about it. The Instacart Clementine press release runs to 1,102 words of body prose. The consumer product page it points shoppers to carries another 110 words of actual product copy. In those 1,212 words, the word “model” appears zero times. So does “LLM”. So do “accuracy”, “trained”, “hallucination”, “human-in-the-loop” and “privacy”. Not one of the eight appears anywhere a shopper will look.
Now walk two clicks sideways, to the enterprise pages Instacart aims at grocers. There, across 2,762 words, the same eight terms appear 18 times. “Model” seven times. “Accuracy” four. “Trained” three. One page states plainly that the white-label version is “Trained on the same grocery data models that power Instacart Marketplace.”
That last sentence is the whole story, because the Instacart Clementine announcement itself confirms the two products share an engine: Cart Assistant is described as “made smarter by the same technology and intelligence behind Clementine.” One system, two audiences, and only one of them is told what it is made of. This article counts every word, does the date arithmetic the announcement leaves implicit, and flags two figures Instacart’s own live pages currently disagree about. We are not alleging deception — we are measuring disclosure.
Table of contents
- What Instacart Clementine Actually Launched
- Counting Every Word Instacart Clementine Says to Shoppers
- What Instacart Tells Grocers About the Same Engine
- The Instacart Clementine Numbers Measure Everything Except Clementine
- Instacart Clementine and the Dates Behind Its Figures
- Two Live Instacart Pages, Two Different Companies
- Cart Assistant Is the Instacart Clementine Engine, White-Labelled
- What Instacart Clementine Means for Anyone Buying AI Systems
- Frequently Asked Questions About Instacart Clementine
- References and Further Reading
What Instacart Clementine Actually Launched
The product, in Instacart’s own words
Instacart (NASDAQ: CART) introduced Clementine as “its AI-powered grocery shopping assistant,” live across North America on the Instacart Marketplace. The company frames it around three problems: deciding what to make for dinner, shopping the way a family actually eats, and staying on budget. A message becomes “a personalized, ready-to-buy cart in seconds.”
Six capabilities, named
The release lists what Instacart Clementine does: generate recipes on the spot or surface curated ones from publishers; apply household dietary preferences such as gluten-free, vegetarian, nut-free or organic; surface deals and lower-cost alternatives; rebuild past orders from “shop my usuals”; assemble occasion lists such as “everything for a football watch party for 20 people”; and turn a photographed handwritten list into a cart.
Where a shopper finds it
Availability is precise. On mobile, the Instacart Clementine icon sits at the top right above the search bar. On desktop, it is at the top of the left-hand navigation. It is live “today to most U.S. and Canadian Instacart customers” through both app and website.
Who is quoted
CEO Chris Rogers frames it as fifteen years of learning “how families shop and eat” put to work. John Adams, Vice President and Head of Product, describes “a system that understands both of those things — not just what you need, but how you eat, what you love, and what’s on the shelves at your store right now.”
| Capability | Example prompt Instacart publishes |
|---|---|
| Meal planning | “I need a week of budget-friendly kid lunches” |
| Dietary targeting | “high-protein easy dinners for two” |
| Reordering | “order my usuals” / “restock my pantry staples” |
| Occasion shopping | “everything for a football watch party for 20 people” |
| List upload | Photograph a handwritten list, or screenshot a digital one |
| Deal finding | Surfaces promotions and lower-cost alternatives |
Counting Every Word Instacart Clementine Says to Shoppers
The method
We fetched Instacart’s press release and the instacart.com/p/clementine product page directly, stripped scripts, styles, site navigation and footer chrome, and counted the body from the dateline to the last sentence before the “About Instacart” boilerplate. The press release body is 1,102 words. The product page carries 110 words once its 119-word department navigation is removed. Counts are case-insensitive substring matches, so “models” counts under “model”.
Zero, eight times over
Across those 1,212 consumer-facing words describing Instacart Clementine, here is the complete tally: model 0, LLM 0, vision-language 0, accuracy 0, trained 0, human-in-the-loop 0, hallucination 0, privacy 0. There is no partial credit hiding in the margins. The word “accurate” appears once, in the phrase “fast, accurate, and accessible” — and that phrase is on the enterprise page, not the consumer one.
The product page is 110 words
The page Instacart links from its own release with “Find more information on Clementine here” is thinner than the release. Stripped of navigation, it is six feature blurbs: “Clementine knows your go-tos.” “Clementine remakes any recipe.” “Clementine spots discounts on your favorite items as you shop.” “Snap a photo. Clementine builds your cart.” The name appears eight times in 110 words. No mechanism appears at all.
What the release does name
To be fair to Instacart, the Instacart Clementine release is not silent about technology in general. It names Anthropic, OpenAI and Google — but for a different feature entirely, namely that grocery conversations begun on Claude, ChatGPT, Gemini or AI Mode in Google Search can be completed on Instacart. Those are inbound traffic partnerships. None of them is described as powering Instacart Clementine, and the release never says what does.
What Instacart Tells Grocers About the Same Engine
The AI Solutions page
The same press release ends with “Find more information on… Cart Assistant here,” pointing at company.instacart.com/enterprise-platform/ai-solutions. That page runs 673 words and is written for retailers. It describes Cart Assistant as a “White-label, grocery-trained AI assistant” and states, in a bullet a shopper will never see, that it is “Trained on the same grocery data models that power Instacart Marketplace.”
The enterprise launch post
Behind that sits a 2,089-word company update from 4 November 2025, “Introducing New Enterprise AI Solutions to Democratize AI for Grocers of All Sizes.” It is specific in a way nothing consumer-facing is. Catalog Engine “leverages AI-driven extraction of more than 1.3 billion data points, combining Vision-Language Models (VLMs) with scaled Human-in-the-Loop (HITL) verification to ensure accuracy and eliminate duplicates.”
Store View and computer vision
The same post explains Store View, which “combines images and videos captured by Instacart shoppers or store associates with our advanced computer vision and AI models to automatically analyze inventory.” Agentic analytics is “powered by Instacart’s proprietary AI machine learning models.” Caper Carts use “edge AI and advanced sensor fusion for item recognition.” This is the vocabulary of a company that knows exactly how to describe its own systems.
The asymmetry, measured
Set the two corpora side by side and the pattern is not subtle. The grocer-facing pages are 2.28 times longer than the consumer-facing ones, but they carry 18 mentions of the six machinery terms against zero. That is 6.52 mentions per thousand grocer words and 0.00 per thousand shopper words.
| Document | Audience | Words | “model” | “accuracy” | “trained” | “LLM” |
|---|---|---|---|---|---|---|
| Clementine press release | Shoppers / press | 1,102 | 0 | 0 | 0 | 0 |
| instacart.com/p/clementine | Shoppers | 110 | 0 | 0 | 0 | 0 |
| Enterprise AI Solutions page | Grocers | 673 | 3 | 0 | 3 | 0 |
| Enterprise AI launch post | Grocers | 2,089 | 4 | 4 | 0 | 2 |
The Instacart Clementine Numbers Measure Everything Except Clementine
Thirteen figures, sorted by what they describe
The Instacart Clementine release is not a thin document numerically. Strip out dates and footnote markers and it carries thirteen substantive figures. Sort them by subject and the distribution is striking: seven describe Instacart’s scale, three describe the consumer problem, three describe survey methodology, and zero measure how Instacart Clementine performs.
The scale numbers
Seven figures establish the foundation: nearly 15 years of operating history, over 1.6 billion lifetime orders, a catalogue of more than 2 billion items, more than 10 million daily inventory signals, nearly 100,000 stores, more than 2,200 retail banners, and a $115 average basket size. These are real and impressive. None of them is about Clementine.
The problem numbers
Three more come from a Harris Poll commissioned by Instacart: more than 80% of Americans say figuring out dinner causes stress, nearly a third say deciding is harder than cooking, and more than three in five would be interested in an AI assistant that helps. The methodology footnote adds three: 2,059 adults, plus or minus 2.7 percentage points, 95% confidence.
The two claims that stop just short
The release does gesture at performance, twice, and both times it declines to put a number on it. “Early results show a clear progression: customers start with simple tasks like restocking, then graduate to higher-order use cases like recipe discovery.” How many customers? Over what period? And: “orders placed with Clementine include more items than Instacart’s typical basket and exceed the company’s industry-leading $115 average basket size.” More by how much? Exceed by how much? An adjective sits where a percentage would go.
What is not measured at all
No adoption figure. No accuracy or substitution-error rate. No cart-abandonment comparison. No user count for a product now live to “most” customers in two countries. No statement of what happens to the conversation data, and no privacy language anywhere in 1,212 words.
Instacart Clementine and the Dates Behind Its Figures
The as-of dates are in the footnotes
Two footnotes carry vintages that the body text does not. The scale figures — 1.6 billion orders, 2 billion items, 100,000 stores, 2,200 banners — are marked “As of December 31, 2025.” That is 252 days before the Instacart Clementine launch. The $115 average basket is “As of June 30, 2026,” or 71 days before. The Harris Poll fieldwork ran 2–4 June 2026, ending 97 days before launch.
The groundwork was public for months
Instacart Clementine did not appear without warning. The enterprise AI push was announced 309 days earlier. The ChatGPT integration press release is dated 8 December 2025, 275 days before. The Claude page carries “Last Updated: Apr 23, 2026” — 139 days. Gemini followed on 19 May 2026, 113 days. AI Mode in Google Search on 16 July 2026, just 55 days before Clementine.
Why the gap matters
The pattern is a company that spent roughly ten months building distribution inside other companies’ assistants, then launched its own. That is a coherent strategy, and it is visible only if you read five separate pages and do the subtraction. The Instacart Clementine release links all five and sequences none of them.
Two Live Instacart Pages, Two Different Companies
Lifetime orders: 1.6 billion or 1.5 billion
On the day Instacart Clementine launched, the release said “over 1.6 billion lifetime orders.” On the same day, the AI Solutions page the release links to said “1.5 billion orders,” and the enterprise post said “over 1.5 billion lifetime orders.” Both pages were live and reachable simultaneously. The gap is 100 million orders, or 6.7%.
Retail partners: 2,200 or 1,800
The same split runs through the partner count. The Instacart Clementine release says “more than 2,200 retail banners.” The enterprise post says Instacart “provides access to over 1,800 retail partners,” and the December 2025 ChatGPT release says “more than 1,800 retailers.” That is a 22.2% difference between two numbers a reader would reasonably treat as the same fact.
The charitable reading, and its limit
These are almost certainly the same underlying business measured at different times, with older marketing pages left unrefreshed — an ordinary content-maintenance problem, not a misrepresentation. The limit is that Instacart’s own launch document links directly to the stale page, and neither carries a visible as-of date in the body. A reader comparing them has no way to tell which is current.
| Fact | Clementine release | Linked enterprise pages | Gap |
|---|---|---|---|
| Lifetime orders | over 1.6 billion | over 1.5 billion | +6.7% |
| Retail partners | more than 2,200 banners | over 1,800 partners | +22.2% |
| Stores | nearly 100,000 | nearly 100,000 | match |
Cart Assistant Is the Instacart Clementine Engine, White-Labelled
What Instacart is selling grocers
Cart Assistant is described as Instacart’s “white-label AI shopping assistant, built for grocers who want to offer their own branded agentic shopping experience.” Grocers can customise “the name and tone to loyalty integration and recipe content.” Crucially, the release says it is “made smarter by the same technology and intelligence behind Clementine” — which is why the enterprise pages’ technical vocabulary is fair evidence about Instacart Clementine itself.
Three live, four named to follow
The release names seven grocers. Food Bazaar, Heritage Grocers Group and Woodman’s are “now live.” ALDI U.S., Harmon’s, The Save Mart Companies and Stew Leonard’s are listed “to follow,” alongside unnamed “additional partners.” So three of seven named grocers, or 42.9%, had actually shipped on announcement day.
The strategic logic
Read together, the two products are one bet placed twice. Instacart Clementine defends the Marketplace against shoppers who start their week in a general-purpose assistant. Cart Assistant sells the same capability to the grocers who might otherwise build it themselves, or buy it from someone else. Both depend on the inventory data management and analytics layer the enterprise pages describe in detail.
What Instacart Clementine Means for Anyone Buying AI Systems
Disclosure follows the buyer, not the risk
The most transferable lesson from the Instacart Clementine launch has nothing to do with groceries. Instacart disclosed model architecture, verification method and accuracy language to the audience that signs contracts, and none of it to the audience that will actually use the product daily. That is not unusual. It is close to standard practice, and it is worth naming as a pattern.
The consumer bears the substitution risk
A grocery assistant that picks items on your behalf makes decisions with real consequences: an allergen swap, a wrong quantity, a substitution that breaks a recipe. The enterprise page talks about accuracy four times because a grocer will ask. A shopper has no equivalent lever, and Instacart Clementine ships to them with no stated error rate at all.
Questions worth asking of any assistant
If you are evaluating autonomous AI agents for your own business, the Instacart Clementine documents are a useful template for what to demand. Ask which model family is used and whether it changes. Ask for a measured accuracy or error rate on your task, not an adjective. Ask what human-in-the-loop verification exists. Ask where conversation data goes and how long it is kept.
Read the page written for the other audience
The practical trick this article rests on is cheap and repeatable. When a vendor publishes a consumer page and an enterprise page for the same system, read both and diff the vocabulary. The enterprise page is written for someone with procurement questions, so it answers them. On Instacart Clementine, that diff was 18 mentions to zero. Our note on what Cresta declined to publish about pricing found the same shape in a different market.
| Question a buyer should ask | Answered on grocer pages? | Answered on shopper pages? |
|---|---|---|
| What kind of model runs it? | Partly — “VLMs”, “machine learning models” | No |
| Is there human verification? | Yes — “Human-in-the-Loop (HITL)” | No |
| Is accuracy discussed? | Yes — four mentions | No |
| What data was it trained on? | Yes — “the same grocery data models” | No |
| What is the error rate? | No | No |
Frequently Asked Questions About Instacart Clementine
Is this article saying Instacart lied?
No. Every claim we checked in the Instacart Clementine release is defensible on its own terms, and nothing obliges a company to name its model vendor. Our claim is narrower and checkable: across 1,212 words written for shoppers, eight technical-disclosure terms appear zero times, while the same company uses them 18 times across 2,762 words written for grocers about the same underlying system.
What model actually powers Instacart Clementine?
Instacart has not said publicly, in any document we could find. The enterprise pages describe “proprietary AI machine learning models” and Vision-Language Models for catalogue work, and name OpenAI, Google and Microsoft as agentic-commerce partners — but partnership for inbound traffic is not the same as supplying the engine. Treat any specific attribution you see elsewhere as unsourced unless it cites Instacart directly.
Does Instacart Clementine cost extra?
The release states no price and does not mention Instacart+ or any subscription requirement. It says the assistant is available “today to most U.S. and Canadian Instacart customers,” which implies it is included, but the word “free” does not appear in a pricing sense anywhere in the announcement.
Is it available outside the US and Canada?
Not per this announcement. Instacart Clementine is described as live “across the U.S. and Canada on the Instacart Marketplace” and “to most” customers there. No other market is named, and no timeline for expansion is given.
How is Clementine different from using Instacart inside ChatGPT?
Those are separate products with separate histories. The ChatGPT, Claude, Gemini and Google AI Mode integrations let a conversation that starts in someone else’s assistant finish as an Instacart order. Instacart Clementine is Instacart’s own assistant, inside Instacart’s own app, with access to your order history and store-level inventory.
What is the strongest single finding here?
That one sentence exists on the grocer page and has no equivalent anywhere a shopper looks: Cart Assistant is “Trained on the same grocery data models that power Instacart Marketplace.” Instacart is willing to describe the training basis of this technology in writing. It simply does not do so in the Instacart Clementine materials aimed at the people using it.
References and Further Reading
Meet Clementine: Instacart’s AI Shopping Assistant That Takes “What’s for Dinner?” Off Your Plate
Shop with Clementine — Instacart
Instacart Enterprise: AI Solutions
Introducing New Enterprise AI Solutions to Democratize AI for Grocers of All Sizes
Instacart launches an AI grocery shopping assistant called Clementine — TechCrunch
Instacart launches custom AI assistants for retailers, and its own — Modern Retail
Instacart is now available in Claude
Instacart Brings Agentic Grocery Shopping to Gemini
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.