AI Shopping Agents Don’t Read Packaging
The shelf is going invisible; and most brands aren't yet visible to what's replacing it.
Shopify activated agentic storefronts for 5.6 million merchants on March 24th. In the twelve months prior, AI-attributed orders on the platform grew 11x and AI-referred traffic was up 7x. According to Adobe’s holiday 2025 data, AI-referred shoppers converted 38% higher than non-AI traffic on Black Friday, and revenue per visit from AI-referred sessions was up 254% year-over-year. Across e-commerce broadly, traffic from AI assistants has been doubling every two months since September 2024, with a 1,300% year-over-year increase during the November-December 2024 holiday window.
Those are not early-adopter numbers, that’s a channel forming in real time.
Bain estimates that 30-45% of US consumers already use generative AI to research and compare products. I would argue this number is already much higher. McKinsey projects the US agentic commerce market alone could reach $1 trillion by 2030. Another piece of Bain’s research suggests AI agents could account for 15-25% of US e-commerce sales by the end of the decade (with grocery and household essentials leading adoption since those are the categories where convenience matters more than brand loyalty).
The structural problem isn’t awareness. Most CPG executives have heard of agentic commerce. The problem is that the entire branded CPG playbook, from package design to shelf placement to trade promotion to retail media, was built to influence a human being standing in an aisle making an emotional decision. AI shopping agents don’t stand in aisles, they don’t respond to end caps or eye-level placement or the color psychology on your bag. They parse through structured data, they compare verified claims, they optimize for consumer-specified parameters (i.e. price, ingredients, ratings, availability, sustainability certifications, etc.) and they surface the product that best matches the query. Not the product with the biggest media spend. Not the product with the most shelf facings. The product that has the best data.
That’s a different competitive game than the one most brands are playing today. For example, when Amazon Rufus recommends products, independent analyses suggest it favors Amazon’s own brands at rates significantly disproportionate to their market share. That tells you the architecture of this new shelf - the platform that controls the agent controls the recommendation; and the recommendation is the new shelf placement.
Moreover, the infrastructure buildout is accelerating faster than most CPG companies are tracking. Google launched the Universal Commerce Protocol at NRF in January 2026 (co-developed with Walmart, Target, Shopify, Etsy and Wayfair) creating an open standard that lets AI agents discover products, negotiate purchases and complete checkout without a human ever touching a screen. OpenAI and Stripe have their own competing protocol, the Agentic Commerce Protocol, already live inside ChatGPT’s instant checkout. Amazon Rufus handles over 250 million daily queries (roughly 14% of all Amazon searches) and converts those users at 60% higher rates than traditional search. Perplexity launched Comet, an AI agent that shops across retailers on your behalf, and Amazon immediately sued to block it. A federal court issued this temporary block in March but was struck down by the Ninth Circuit court just days later. The legal fight over who controls agentic commerce is being litigated right now; and the outcome will shape how products get discovered, compared and purchased for at least the next decade.
While Google’s UCP and OpenAI’s ACP are the early attempts to create a standardized protocol for any AI agent to discover and transact with any merchant, “open” in platform economics has a history that usually ends with the platform extracting more value than it distributes. The brands that built their digital strategies around Facebook organic reach in 2014, then Amazon search ranking in 2018, then retail media in 2022, know exactly how this cycle works. The channel starts open, the economics look favorable and then the platform tightens the algorithm and charges rent.
The difference this time is that the intermediary isn’t just controlling the shelf. It’s replacing the shopper. When an AI agent makes a purchasing decision based on structured data and preference algorithms, the consumer relationship (the very thing that brand equity is supposed to represent) runs through a layer of code that the brand doesn’t own, can’t see and increasingly can’t influence through traditional marketing. Right now, most brands aren’t visible to that code. Product data lives in retailer-managed systems, uploaded through portals designed for human category managers, stored in formats optimized for planogram software - not for agentic parsing. The brand’s own website often hasn’t even been meaningfully updated since the direct-to-consumer push of 2019.
An AI agent trying to compare a mid-market protein bar against competitors has to scrape inconsistent data from five different retailer sites, none of which match the brand’s own nutrition claims format. Meanwhile, the private label alternative has clean, structured, machine-readable data because the retailer built it that way from the start. The agent doesn’t know or care which product is the “brand leader”, it surfaces the product that has the better data.
Most of the conversation around AI in CPG right now centers on internal deployment (demand sensing, supply chain optimization, pricing models, etc.). Those matter, but they don’t touch the question agentic commerce is forcing - Is your product being found by the AI agent that’s making the buying decision on your consumer’s behalf?
Structured product data that agents can parse without scraping. Standardized claims with third-party verification an algorithm can read. Commerce infrastructure that lets your product be discovered, compared and purchased inside an AI conversation without the consumer ever visiting your website or the retailer’s. The gap between where that data needs to be and where it actually sits today, for most brands, is enormous.
The companies treating agentic commerce readiness as a 2028 initiative are making the same mistake as the brands that treated e-commerce as a 2015 initiative in 2010 - risking arriving only after the shelf is already built, the protocols are already set and the algorithms have already learned which brands to recommend (and which to skip). You can’t shortcut your way to AI discoverability, tou can’t acquire it in a deal unless you’re specifically buying data infrastructure and digital commerce capability and every adoption curve in the data suggests critical mass is closer to 2027 than 2030.
The shelf is going invisible, and agents don’t read packaging.