Partner Content from SPINS
CPG marketers have spent decades winning the physical shelf through tactics like distribution, packaging and promotion. Then they learned to compete on the digital shelf with owned sites, retailer ecommerce and SEO fundamentals. Now a third layer has emerged, transforming product discovery: AI-assisted search.
Data intelligence firm SPINS refers to this environment as the “agentic shelf,” where AI systems can influence which products surface from a shopper’s natural language query. Before ever researching a brand or visiting a retailer’s site, shoppers can ask an AI assistant for a short list of recommendations matching their intent.
But that list relies on the evidence assistants gather across brand sites, retailer pages and third-party sources. The risk: despite products being available for purchase everywhere, incomplete or inconsistent product content can keep them out of AI recommendations.
To expand on this new layer of product discovery, Chief Marketer spoke with Jessica Wright, SVP of Product at SPINS, author of the company’s ebook “Win the Agentic Shelf.” In this conversation, Wright explains why the attribute is becoming the new SKU, which metrics belong in front of a board, where to invest across owned, retailer and earned sources, and what separates a durable advantage from table stakes.
Chief Marketer: You’ve characterized the agentic shelf as the “third shelf.” How is AI-mediated discovery changing actual CPG purchase behavior, not just search behavior? Can you share any data to support this?
Jessica Wright, SVP of Product at SPINS: The real change is that the consideration set now forms before the shopper ever reaches a shelf. On the physical shelf, a shopper browses an aisle, and on the digital shelf, they browse through listings on a retailer’s ecommerce website. But on the Third Shelf, they describe a need and an AI assistant weighs the evidence and narrows the options before a retailer or brand site has even loaded. Brands are now competing for algorithmic eligibility before they compete for the click or the cart, which means discovery and purchase decisions are collapsing into the same moment.
The early commercial evidence points in the right direction. For MadeGood, a purpose-driven, better-for-you snack brand, their improved ChatGPT visibility was accompanied by a 26% increase in purchase-intent clicks within 90 days. That is a leading indicator rather than proof that AI visibility caused incremental sales, and the industry will lose credibility quickly if it presents correlation as causation, but it is exactly the kind of signal that shows where demand is starting to form.
CM: Many marketers are layering GEO and AEO onto their existing SEO practices. How does the agentic shelf approach in your report differ from adding new tactics to an SEO foundation? And what should a CMO stop doing?
JW: Layering GEO and AEO onto an SEO foundation treats AI discovery as a ranking problem, when it is really a product-understanding problem. SEO taught brands to help a page rank for a query, whereas AI systems synthesize brand sites, retailer pages and third-party sources and then decide which products credibly answer a shopper’s need. The unit of competition is shifting from the page and the keyword to product truth, evidence and eligibility.
What a CMO should stop doing is treating every mention as equally valuable. A brand can improve its visibility across hundreds of prompts and make no commercial progress if those prompts have little connection to what its portfolio can credibly satisfy. That is why measurement should be anchored in Demand-Backed Prompts, the shopper questions that reflect real demand in the need states where the brand has a right to win, rather than whatever queries are easiest to win. The question worth asking is whether the brand is being accurately recommended for the shopper needs that matter commercially and that its catalog can credibly satisfy.
CM: The report argues that counting mentions is vanity optimization. Which AI visibility metrics should marketing leaders still report to their boards?
JW: Boards need a small set of measures that connect AI representation to commercial demand, not a single visibility score. The scorecard I’d put in front of a board covers AI Share of Voice and average position for priority need states, representation accuracy, citation share by source type, inclusion in the competitive consideration set and progress in closing high-value content gaps. Where relevant, each should be segmented by assistant and tracked over time.
Downstream indicators such as AI-referred visits, purchase-intent activity and retailer destinations belong on the same page as evidence that shoppers are moving through the journey — though not as dollar-for-dollar sales attribution, which current measurement does not yet support. The board-level question is whether the brand is becoming more visible, more accurate and more competitive in the discovery moments that matter commercially, and whether stronger discoverability is showing up as stronger engagement.
CM: You say the “attribute is the new SKU.” How should a brand with a large portfolio decide which need-states each product should own?
JW: A brand should not chase every growing attribute or need-state. Category demand shows where shoppers are heading, but the catalog decides which of those needs a brand should own. The right portfolio sits where demonstrated demand, product eligibility, brand credibility, competitive whitespace and economic importance overlap, which usually means protecting the needs tied to today’s core business while investing selectively in emerging ones where the brand has a genuine right to win.
Calling the attribute the new SKU does not make the attribute the strategy; it makes the attribute the evidence. A parent asking for a snack for a child with food allergies who avoids animal products is describing a vegan, allergen-friendly product without typing either term, and an assistant can only make that match if the product’s data supports the attribute across every source it reads. The attribute’s job is to prove that a product credibly satisfies that intent. Each product should have a clear role in the need-states its attributes can substantiate, so the lineup reinforces a focused set of commercially meaningful needs instead of competing with itself for every prompt the category generates.
CM: The report separates owned, influenced and earned sources. In your experience, which source type moves AI recommendations most, and are brands underinvesting in earned authority?
JW: No single source type moves recommendations most, and the brands that go looking for one tend to overinvest in their own websites. Owned pages carry the product truth a brand controls, retailer pages connect discovery directly to availability and purchase, and earned sources add independent corroboration, with the weight of each varying by assistant, category and shopper question.
The citation data makes the imbalance clear. In our analysis of more than 347,000 AI citations, retailer websites accounted for roughly half of the sources referenced, while brand-owned sites represented less than a quarter. Many brands are underinvesting both in the consistency of their retailer content and in external evidence that independently validates what the brand says about itself, which is a costly gap when the assistant is far more likely to be reading the retailer’s page than the brand’s.
CM: If every brand optimizes for the same assistants, does the advantage disappear? What is the durable edge once the fundamentals are table stakes?
JW: The advantage doesn’t disappear, because every brand wins for a different reason. Technical readiness will become table stakes, and what drives growth for one brand is often completely different from what drives growth for another. One brand may grow on gut-health benefits while a peer in the same category grows on kid-friendly lunchbox convenience. Category demand shows where shoppers are heading, and the catalog decides which of those needs a brand can credibly own. The durable edge comes from optimizing around the need-states and use cases the catalog has a right to win, and from the operating discipline to keep responding as consumer demand and the source environment change.
That is why optimizing around whatever is relevant to the category is a weak strategy. A category-wide list of trending needs sends every brand after the same prompts, and most will lack the products, proof or credibility to win them. Any brand can improve its schema or rewrite a product page. Far fewer can identify where their catalog has a credible right to win, back it with proprietary product knowledge and coordinate evidence across owned, retailer and earned sources. Over time, the advantage goes to the brand that is the most relevant, trusted and commercially informed answer for the needs its own portfolio genuinely serves.
CM: Twelve months from now, what will leaders wish they had started doing today?
JW: Leaders will wish they had established a baseline, because many brands still cannot say which shopper questions matter, whether their products enter the AI consideration set, what assistants believe about those products or which sources are shaping the answers. Without that baseline they have no way to tell whether their position is improving or whether a competitor is quietly taking ground.
The practical starting point is a focused set of commercially important need-states, an audit of how the portfolio is represented against them, fixes for the highest-value content gaps and clear ownership across marketing, ecommerce, data and IT. What turns that into an advantage is running it as a repeatable cycle of measure, prioritize, act and validate as demand, sources and competition move, so that a year from now the brand is optimizing while its competitors are still working out where they stand.

