Shoppers are increasingly asking ChatGPT, Claude or Gemini for product recommendations before they ever open a retailer app, and that shift is changing how brands position themselves online. Tech writer Deana Burke recently outlined the concept of the AI shelf in her newsletter, then tested it by inventing a completely fictional natural deodorant brand to see whether she could get it recommended. The experiment worked, which is precisely what makes the concept uncomfortable for brand teams.
A Fictional Deodorant Brand That Got Recommended
The test mattered because it removed every advantage a real brand normally relies on. There was no manufacturing history, no reviews, no retail distribution and no advertising budget behind the invented label. What the experiment isolated was the mechanism itself: how these algorithms decide which products to surface, and what signals they treat as evidence of legitimacy. Tech writer Deana Burke outlined the AI shelf concept and tested it
For brands accustomed to optimizing for physical shelves and digital storefronts, the AI shelf adds a third surface that behaves nothing like the first two. Shelf space is bought, storefront ranking is earned through conversion data, and the AI shelf is assembled at the moment of a question. That assembly is invisible to the brand being described.
Why Machine Readability Beats Copywriting
Jessica Wright, senior vice president of product at Spins Foundry, joined the Modern Retail Podcast this week to unpack the shift, covering how personalized AI shelves are shaping CPG trends and why brands need their products to be machine-readable for AI agents. The practical implication is blunt: content that reads well to a human but resolves poorly to a parser contributes almost nothing to the new surface. Jessica Wright of Spins Foundry unpacked how AI shelves shape CPG trends
Machine readability is a structural problem rather than a stylistic one. Product attributes, ingredient lists, pack sizes, certifications and price bands all need to exist as discrete, consistent fields rather than as sentences inside a paragraph. When those fields are missing or contradictory across a brand's own pages, the model has to guess, and guessing tends to favor whichever competitor states things plainly.
Three Signals That Decide Whether a Brand Appears
Consistency across every owned surface
A brand that describes the same product differently on its own site, on a marketplace listing and in a press release gives the model conflicting evidence. The cheapest fix is also the most boring one: a single canonical description, reused everywhere, with the same numbers and the same claims in the same order.
Third-party corroboration the model can cite
Models prefer claims that can be checked against an independent source. A volume claim supported by a category report is worth more than the same claim standing alone, because the citation gives the answer a place to point. Brands that publish their own numbers without any external anchor are effectively asking to be taken on trust.
Structured answers to real questions
Buyers ask comparative questions, not descriptive ones. Content organized around "which one is better for sensitive skin" or "what is the difference between these two pack sizes" gets retrieved far more often than a brand story, because it matches the shape of the query rather than the shape of a brochure.
What Changes Inside the Marketing Budget
If the AI shelf becomes a primary discovery surface, the split between organic and paid marketing stops being a clean line. Paid placement does not buy a mention inside a generated answer, and organic reach does not guarantee one either. What the two now share is a dependency on the same underlying evidence base, which means budget decisions have to be justified against a metric most teams do not yet track. Retailers are already rewiring how they present value, with Kohls leaning into clearer pricing and gift bundles for the holidays, and that message has to survive being paraphrased by an assistant. Kohls is leaning into clearer pricing and gift bundles for the holidays
That metric is citation frequency: how often a brand appears in answers, in what context, and alongside which competitors. It behaves more like a share-of-voice measure than a conversion measure, and it moves slowly. Brands that wait for a clean attribution model before investing will likely find the surface already occupied by competitors with worse products and better documentation.
The Operational Cost Nobody Budgets For
Maintaining machine-readable product data is ongoing work, not a one-time cleanup. Packaging changes, formulations get reformulated, pack sizes shift with seasonal promotions, and every one of those changes has to propagate to every surface at the same time. Retail organizations built around campaign calendars are not structured for that cadence.
There is also an internal coordination problem. The team that owns product data, the team that owns content and the team that owns retail media rarely report to the same person. On the AI shelf, all three feed the same answer, so a discrepancy between them becomes a visible inconsistency rather than an internal footnote. Closing that gap is the unglamorous prerequisite for everything else. Albertsons is accelerating turnaround efforts amid an identity crisis
