AI Shopping Benchmark: Why 16 of 18 ChatGPT‑Shoppable Grocers Depend on Instacart

The Benchmark and What It Measures

AstraWorks, an AI enablement firm, evaluated 24 retailers on their agentic AI shopping capabilities. The scoring methodology covered three core dimensions: proprietary AI shopping tools, the presence of AI assistants on websites and apps, and direct integrations with third-party AI agents such as ChatGPT, Claude, Gemini, and Perplexity.

The resulting index runs to 100 points. Amazon leads with 63, Walmart follows at 49, and Albertsons and Wegmans tie for third at 38. The majority of retailers cluster between 10 and 24 points — a range AstraWorks designates as “starters.” That label is accurate in both senses: they have started, but they have not gone far.

Instacart’s Structural Advantage

Instacart launched its ChatGPT integration in December, and the timing proved decisive. With no proprietary AI shopping infrastructure of their own, most grocers defaulted to the integration that was already available — the same pattern that played out with online ordering more than a decade ago.

AstraWorks identifies 11 retailers, including Publix, H-E-B, Schnuck Markets, and Ahold Delhaize USA, as “platform-dependent.” The report’s language is precise and worth quoting directly: remove the Instacart integration, and most of their agentic commerce score approaches zero.

This is not a criticism of those retailers so much as a description of where the market stands. Instacart moved first, built the integration, and offered grocers a functional path to AI shoppers without requiring internal development capacity. The trade-off is structural dependency.

What Platform Dependency Actually Costs

The AstraWorks report draws a clear distinction between having a working path to the consumer and controlling that path. A retailer connected to ChatGPT via Instacart can be discovered and shopped through AI agents — but the interface, the intelligence layer, the data capture, and the long-term differentiation all belong to the intermediary.

This matters more than it might appear in the short term. Data generated through AI shopping interactions is precisely the kind of first-party behavioral signal that trains better models, improves personalization, and compounds over time. Retailers who route that data through a third party are, in effect, outsourcing the learning curve.

The parallel to early e-commerce is instructive. Many grocers that became dependent on Instacart for online orders eventually built their own websites and pickup services — but they did so years later, from a weaker position, and often while continuing to pay for the third-party channel anyway. The question is whether the same pattern repeats in agentic commerce, or whether some retailers move faster this time.

The Retailers Building Their Own Path

Two names stand out in the benchmark as exceptions to the platform-dependency pattern: Albertsons and Wegmans, both scoring 38 points and tying for third place behind Amazon and Walmart.

The Outliers Worth Noting

At the bottom of the index, the results are telling in their own way. Whole Foods Market scores just 9 points — a striking underperformance for a retailer owned by Amazon, the benchmark’s top scorer. Trader Joe’s, Dollar General, and WinCo Foods round out the lowest tier.

These results reflect deliberate positioning as much as capability gaps. Trader Joe’s has historically resisted digital commerce infrastructure. Dollar General’s customer base and price-point strategy may not align with the early adopter profile of AI shoppers. But the gap between Amazon’s 63 and Whole Foods’ 9 within the same corporate family is a reminder that agentic commerce readiness is not automatically inherited — it must be built or integrated at the retailer level.

Agentic Commerce: Discovery Surface or Transaction Channel?

The AstraWorks report is careful not to overstate where agentic commerce currently stands. Whether AI shopping agents remain primarily a discovery surface — helping consumers find products and compare options — or evolve into a significant automated transaction channel is, as the report notes, “an active debate.”

That uncertainty is itself strategically important. Retailers building proprietary capabilities now are making a bet on the latter scenario. Those relying on Instacart are, implicitly, treating AI shopping as a feature rather than a channel — something to have covered without committing significant resources.

The risk of the second posture is that if agentic commerce does become a primary transaction channel, the window for building proprietary infrastructure will have narrowed considerably. The cost of catching up tends to be higher than the cost of building early.

What This Benchmark Signals for the AI Tools Ecosystem

From a platform perspective, this benchmark illustrates a pattern that recurs across AI adoption cycles: a well-positioned intermediary captures the integration layer before most incumbents have defined their own strategy. Instacart’s ChatGPT integration is, in structural terms, similar to what payment processors, search advertising platforms, and social commerce tools have done in prior cycles — it solves a real problem quickly, at the cost of long-term dependency.

For AI tool observers, the more interesting signal is the divergence between retailers who are treating AI shopping as an infrastructure question and those treating it as a vendor relationship. Albertsons’ multi-agent direct integration strategy suggests a sophisticated understanding of where value accumulates in an AI-mediated commerce stack. Most of the industry is not there yet.

The Takeaway

The 16-of-18 figure is a snapshot, not a verdict. Instacart’s early advantage in AI shopping integration is real, but the grocery industry has navigated this dynamic before — and the retailers that built their own capabilities eventually recovered competitive ground.

What is different this time is the speed of the underlying technology and the compounding value of the data generated through AI interactions. Retailers that move from platform dependency to proprietary capability sooner will accumulate a learning advantage that is genuinely difficult to replicate later. The benchmark makes clear who has started that journey — and how much ground the rest still need to cover.