The Shelf Is Not the Store

You spent twenty years learning how to win the shelf.

Endcaps. Eye level. Planograms. The digital version was barely easier: rank in Google Shopping, hold the buy box on Amazon, fight for position on a retailer page you do not control. All of that was one shelf. A single, physical-ish battlefield.

That battlefield is gone. And the thing that replaced it does not even have a fixed location.

Brandtech's Jellyfish dropped a study this week that should scare every ecommerce marketer who still thinks in rankings. Their Share of Model tool went and asked three AI shopping assistants the same product questions across eight categories, fashion, athletic wear, men's suits, home appliances, furniture, in the US, UK, Australia, and Singapore. Then they compared what each assistant put on the "shelf."

There is no shelf. There are many shelves, and they barely agree with each other.

What it is

On October 1, Jellyfish launched Shopping Optimization, a new capability inside its Share of Model platform. It analyzes AI shopping behavior at the individual product level, showing marketers how AI systems evaluate, compare, and recommend specific SKUs, and which retailer the AI would actually buy from.

The launch is backed by genuinely interesting research. Jellyfish measured how ChatGPT, Google's AI Mode, and Amazon's Alexa for Shopping recommend products, and the results read like a stress test for every assumption brands hold about discovery.

The numbers are worth your attention. Across the eight categories, ChatGPT accounted for roughly 80 percent of AI product recommendations and Google's AI Mode around 20 percent, a four-to-one gap. But that gap swings wildly by category. In home appliances, it was 97 percent to 3 percent. Nearly thirty to one.

A single AI shopping question surfaced as many as 290 competing brands. In US fashion, recommendations spanned 120 brands, and the most recommended brand held just 7 percent of the shelf. Seven percent. That is not a leaderboard. That is a lottery ticket distribution.

And price positioning, the thing premium brands pay agencies millions to defend? Collapsed. In one AI response, men's suits in the UK ranged from £27 to £3,295. Gaming chairs in the US went from $79 to $3,479. Your premium product now sits one line below a budget alternative in the same recommendation, and nobody asked your permission.

What changed

Here is the shift that actually matters. For years, digital shelf tracking measured how you rank on a retailer's own website. Then the first generation of AI visibility tools measured whether a chatbot mentions your brand at all. Both are now outdated.

Share of Model measures something different: the specific products an AI recommends, where they rank, what they cost, how they are rated, and which retailer the AI would buy them from. That last bit is the tell. The AI does not just recommend. It decides where the money goes.

The research also exposes how differently each assistant "shops." Asked for toys, Amazon's assistant recommended products from 177 brands but sourced them almost entirely from itself. For the same request, ChatGPT drew on 24 retailers and Google's AI Mode on 37. In US athletic wear, ChatGPT considered 171 retailers and Google's AI Mode 126.

John Dawson, Jellyfish's VP of Strategy, put it bluntly: "Our own data shows the same product question can return a 30-retailer shortlist on one assistant and a closed, single-store answer on another, so 'winning AI' isn't one race, it's many, and most brands can't yet see the starting line."

That is the new reality in one sentence. You are not competing for a position anymore. You are competing across parallel universes of recommendation, each with its own rules, and you cannot see any of them.

What works

The brands that win this will be the ones that stop optimizing for the answer and start optimizing for the decision.

Jellyfish's platform pushes toward what it calls Generative Engine Marketing, or GEM: measuring and improving product performance across owned, earned, and paid AI touchpoints throughout the customer journey. SKU-level visibility, not brand-level vanity. Understanding why specific products get recommended or overlooked. Finding the attributes, content signals, and information sources that actually influence AI recommendations.

Natasha Wallace, Jellyfish's Chief Solutions Officer, framed the gap plainly: "Until now, marketers have had little visibility into why AI shopping systems favor one product over another. Share of Model's Shopping Optimization gives brands a clear understanding of the factors driving AI recommendations, so they can focus their optimisation efforts and measure the business impact."

The interesting wrinkle is that this pairs with what Constructor is doing on the retail side. Constructor announced October 2 that companies deploying its conversational AI agents are up 900 percent year over year, and its Discovery MCP server puts a retailer's personalized product discovery inside external engines like ChatGPT. Agents are starting to talk to agents. Your catalog is being queried by software you do not operate, on behalf of shoppers you will never see, and the shelf is assembled in milliseconds by a model, not a merchandiser.

What breaks or stays fenced

Plenty of this stays fenced, and you should read the research with the right skepticism.

First, it is vendor research. Jellyfish sells exactly this capability, and the study is the marketing for it. The data is real and specific, which gives it weight, but the conclusions conveniently point at the product.

Second, the assistants move fast. ChatGPT at 80 percent of recommendations today tells you about distribution, not destiny. Google's AI Mode is a rounding error in fashion and dominant nowhere yet, but Google owns the intent layer underneath half the internet. Anyone betting the next two years look exactly like this chart is kidding themselves.

Third, conventional tools are not dead, they are demoted. Ranking on a retailer's site still drives the actual conversion for most categories. The AI shelf decides what gets considered; the retail shelf still closes. Brands that abandon one for the other get it wrong in both directions.

And the closed ecosystems are the wildcard. Amazon's assistant sourcing almost everything from Amazon itself is the most important single finding in the whole study. When the assistant and the store are the same company, the "shelf" is a moat. Winning inside Alexa for Shopping is a different game from winning inside ChatGPT, and there is no unified strategy that covers both.

Who it is for

If you sell physical products online and you are not yet tracking how AI assistants recommend your category, you are flying blind. This matters most for:

  • Brand marketers in fashion, beauty, home, and consumer electronics, the categories where AI recommendations are already dense and fragmented.
  • DTC teams whose product feeds, reviews, and content are the raw material AI models ingest. Your structured data is your new packaging.
  • Retail media buyers. When the AI decides which retailer to buy from, retail media strategy and AI visibility strategy are the same meeting.
  • Anyone who spent 2025 arguing about whether GEO was real. It is real now. It has a dashboard, a launch date, and a four-to-one leader.

What to do this week

  1. Ask the question your customers ask. Open ChatGPT, Google AI Mode, and Alexa. Ask each how they would pick a product in your category. Screenshot everything. You now have your first AI shelf audit, and it cost you nothing.
  2. Check your product data, not your rankings. The AI ingests structured product info, reviews, pricing, and content signals. If your feeds are thin, inconsistent, or outdated, you are not losing a ranking war. You are invisible in the consideration set.
  3. Segment by assistant, not just by channel. Build separate tracking for ChatGPT, AI Mode, and Alexa for Shopping. The research shows winning on one tells you almost nothing about your position on another. One dashboard for "AI visibility" is a lie.
  4. Fix the listing before you buy the bid. When the AI weighs inventory, ratings, and content alongside price, the highest-ROI move this week might be better product content and review velocity, not more spend. The cheapest shelf space in AI commerce is a well-fed product feed.

Sources