Someone asked an AI assistant for a birthday gift yesterday. They didn't open a browser. They didn't navigate a category page or type keywords into a search bar. They described what they needed - price point, occasion, delivery window, the feeling they wanted the recipient to have - and received a recommendation they acted on. That shopper will never see your end-cap display, your trade promotion, or the planogram position you negotiated last fall. The question worth sitting with is whether your brand was in that conversation at all.
Microsoft's agentic commerce report, published this month, makes this case well. Bain & Company puts 30 to 45 percent of U.S. consumers already using generative AI to research and compare products. McKinsey projects $3 to $5 trillion in global retail revenue orchestrated by AI agents by 2030. The direction of travel is not ambiguous. But the report is written for a general CMO audience, and CPG brands face a version of this challenge that is more specific, more structural, and considerably more urgent than a platform-level analysis tends to capture.
What Microsoft Got Right
The Microsoft framing - "the front door to retail has moved from the search box to the conversation" - is accurate and important. Shoppers are now expressing intent in natural language: something sustainable for a coworker who cooks, under fifty dollars, arrives by Friday, feels premium rather than generic. The AI agent interprets that intent, weighs tradeoffs in real time, and surfaces a recommendation. The conversation is the discovery, the evaluation, and the decision, all compressed into a single exchange.
The report is also right that brands face two distinct imperatives: be discoverable on third-party AI platforms, and build owned conversational experiences that generate proprietary learning. Both are real. Both matter. The sequencing question - which comes first and why - is where we'd push back, and we'll get to that.
What the report doesn't fully address is what those imperatives actually require of CPG brands in practice. The language of "optimizing for AI discoverability" sounds like a channel extension, a new tab on the digital media dashboard. It isn't. It requires confronting something more uncomfortable: most CPG product data was built for a world that no longer leads the purchase decision for a growing share of shoppers.
The Shelf You Can't Negotiate
In physical retail, shelf presence is a negotiated outcome. You invest in slotting fees, execute trade promotions, win category captain status, and compete for planogram position through a set of rules that have been refined over fifty years. The investment is enormous - the American CPG industry collectively spends more than $200 billion annually on it - but the rules are known, the players are defined, and the metrics are well understood.
In agentic commerce, none of those rules apply. When a consumer asks a shopping agent to recommend a protein bar, a laundry detergent, or a baby formula, the agent doesn't consult a planogram. It parses structured product data: ingredient attributes, use case descriptions, compatibility claims, real-time availability, the language your brand uses across every platform where it appears. The brand that invested the most in end-cap displays is not the brand that wins that recommendation. The brand with the most coherent, structured, machine-readable product content is.
"The agent doesn't care that you won category captain. It cares whether your structured data answers the question a shopper just asked in natural language."
This is a structural mismatch, not a budget allocation problem. The assets CPG brands have spent decades building - retail relationships, promotional machinery, planogram position - are largely invisible to the systems now influencing a rapidly growing share of purchase decisions. That doesn't mean those investments no longer matter. It means they confer no advantage in the fastest-growing channel in retail.
The Sequencing Problem
Microsoft's two imperatives - third-party discoverability and owned agentic experiences - are both correct in direction. But the report implies a rough equivalence between them that doesn't hold for most CPG organizations. For brands with product catalogs built across multiple lines, multiple agencies, and multiple years of inconsistent data entry, the honest first step is neither of those things. It's an audit.
Before you can optimize for AI discoverability, you need to know what an AI agent actually sees when it parses your product data today. Before you can build a meaningful owned conversational experience, you need product attributes that can power one. The Bain finding that consumers trust a brand's own AI agent three times more than a third-party agent is extraordinary - and almost entirely unrealized across the CPG sector. Very few CPG brands have deployed anything approaching a genuine AI shopping agent on their owned properties. Most haven't established what their product data looks like to the agents already recommending or not recommending their SKUs right now.
The sequence that actually works: understand your current state first, then build, then optimize for external discoverability as the owned experience generates the learning that makes optimization meaningful. Starting with discoverability optimization before you've established a foundation is spending money to appear in a conversation you're not yet equipped to win.
The Challenger Brand Advantage
There's a reason challenger brands are winning in AI agent recommendations at a rate that legacy CPG cannot yet explain through product quality alone. IPG research showing challengers winning in sixteen of eighteen CPG categories in agentic recommendations isn't luck - it's a structural artifact of how those brands were built. Smaller catalogs, cleaner data, less accumulated inconsistency across product lines and agency relationships. Their product content tends to be structured more recently, with more attention to the attributes AI agents actually parse, because they had no choice but to build for the current environment rather than the environment of fifteen years ago.
Legacy brands still have advantages: scale, decades of brand recognition, resources to invest, and retail relationships that aren't going away. But those advantages don't transfer to the agentic channel automatically. They have to be rebuilt in a language that agents can read.
The Four Actions, Re-Sequenced
Microsoft's four actions for retail leaders - optimize for AI discoverability, launch owned conversational experiences, design for openness and portability, govern measurement for compounding learning - are all directionally right. Here's how we'd re-sequence them for CPG specifically.
Start with a data audit. You cannot optimize what you don't understand, and most CPG brand teams don't have an accurate picture of how their product attributes currently read to AI systems. This is the foundational step, and it's the one most organizations skip because it's unglamorous and doesn't produce an immediate deliverable. It produces something more valuable: an accurate starting point.
Then establish owned visibility before optimizing for third-party platforms. The 3x trust advantage for brand-owned agents means your own properties are the highest-converting surface in agentic commerce. Build that before you spend on appearing in someone else's recommendation engine.
Then pursue external discoverability - AEO, GEO, structured data relationships with Google Merchant Center and the emerging AI shopping platforms - informed by what you've learned from your own data. This sequence means every dollar you spend on external discoverability is backed by a product catalog that can actually win the recommendation.
Measurement and governance run across all of it. The Microsoft report is right that brands risk losing visibility into how decisions are made as AI agents mediate more of the shopper journey. Clear ownership of signal data, clear expectations around learning, and clear metrics that go beyond last-click attribution are non-negotiable from the start, not something to layer in later.
"The brands that will be recommended in 2028 are building their advantage in 2026, before the channel is crowded enough that catching up costs twice as much."
The Window Is Narrower Than It Looks
The Microsoft report says "the question for CMOs is not whether to participate in agentic commerce." We agree completely. The CPG-specific version is sharper: the brands that will be recommended when a consumer asks an AI agent for a snack, a cleaning product, or a personal care item in 2028 are not the brands that reallocate budget when the numbers become undeniable. They're the ones doing the foundational work now, when the work is unsexy and the competitive advantage is still available.
The agentic shelf is being built right now, in product databases most CPG brand teams have never audited for machine readability. The trade spend, the slotting fees, the planogram negotiations - none of that secures a position on it. Understanding what your product data actually looks like to an AI agent is the first question. Most brands don't yet have an answer.
Statistics in this article draw on research from Bain & Company, McKinsey Global Institute, Microsoft Advertising, Adobe Commerce, and IPG. Analysis and editorial perspective are original to TheoryNXT / The Invisible Shelf. This piece was written in response to Microsoft's "How AI Conversations Are Replacing Traditional Search Results for Retail," May 2026.