The Subject of Shopping Shifts from ‘Humans’ to ‘AI’

Your product may no longer be chosen by a human.

The AI shopping assistant announced by Shipt in May 2025 can create a cart for “a tailgate party for 25 people” simply by saying so, gathering all the necessary food, drinks, and supplies into the cart. Instacart’s conversational AI assistant, “Clementine,” chats with users to understand their preferences and budget, selecting the optimal products to add to the cart.

Then there’s Meta’s “Muse.” This AI agent performs tasks on behalf of users, not limited to shopping, but also including sending emails, booking travel, and negotiating bills. Purchasing behavior is an extension of this.

The essence of what is happening here is simple. The decision-making process of ‘what to buy’ is beginning to shift away from human hands.

Consumers no longer need to browse supermarket shelves and think, “Maybe I’ll go with this one.” AI assesses past purchase history, reviews, prices, nutritional information, and stock availability to determine, “This is the best choice.” Humans simply press the “OK” button.

I want to consider this structural change from the perspective of small and medium enterprises.

What Is Happening: Organizing Three Services

First, let’s establish the facts.

Shipt’s AI Assistant

• Allows cart creation requests in natural language
• Automatically generates suggestions based on events and number of people by linking to Target’s product database
• Approximately 50 million households in the U.S. are within Target’s commercial area

Instacart’s Clementine

• A conversational interface that gathers user preferences, allergies, and budget
• Selects from over 80,000 products across more than 1,500 retail banners
• Instacart’s annual gross transaction value (GTV) is approximately $30 billion

Meta’s Muse

• A general-purpose AI agent not limited to shopping
• Meta’s user base is about 3.9 billion people worldwide
• If purchasing agent functionality is implemented, the impact will be unprecedented

What stands out is the scale. Instacart alone handles $30 billion in annual shopping, equivalent to about 4.5 trillion yen. The transition of the “selection logic” for this distribution amount to AI means that the rules of competition for 4.5 trillion yen worth of shelf space will fundamentally change.

The Rules for Being ‘Chosen’ Will Change

In the traditional retail world, simply getting your product on the shelf was the competition. You would pitch to supermarket buyers, secure shelf space, and make your product stand out with end displays and POP advertising. Once in the consumer’s view, it was a battle of package design and price.

In a world where AI handles shopping, this premise collapses.

Because there will be no ‘eyes’ of consumers.

When AI selects products, it does not refer to human vision. Instead, it relies on information registered in product databases—product names, ingredients, prices, review ratings, stock status, past purchase data, and the total quantity and quality of information available online about that product.

In other words, the following will happen:

• Spending 1 million yen on package design will be invisible to AI
• Distributing samples to 200 people in-store will not be recorded in AI’s data
• Conversely, if the product description page lacks information, it may not even be considered by AI

The value structure of costs will be turned upside down.

The value of large-scale advertising and TV commercials, which small and medium enterprises have thought they couldn’t afford, will relatively decrease. On the other hand, the value of the task of “accurately and richly organizing product information digitally,” which can be done for just a few thousand yen, will skyrocket.

From SEO to GEO—From an Era of Being Searched to an Era of ‘Being Chosen by AI’

There is a concept I want to emphasize here: GEO (Generative Engine Optimization).

SEO was about optimizing to rank high in Google search results. GEO is about creating a state where your information is referenced when generative AIs like ChatGPT and Clementine provide answers or recommendations.

The differences between the two are clear.

In the world of GEO, the competition will be whether AI has access to information that allows it to judge “who this product is suitable for, in what situations, and why.”

What specifically should be done?

Five Things Small and Medium Enterprises Should Do Right Now

1. Organize Product Data in a ‘Format Readable by AI’

Product names, ingredients, capacity, price, origin, allergen information, storage methods—these should be described on your own site using structured data (like schema.org). The cost is almost zero. If using WordPress, free plugins can handle this.

The key point is to present it in a form that “AI can analyze and judge,” rather than just “human-readable.” Simply placing a PDF catalog on the site will not be readable by AI.

2. Accumulate Reviews and Testimonials in ‘Quantity’

In AI’s recommendation algorithms, the number of reviews and ratings are extremely important decision-making factors. Large companies can gather reviews through advertising budgets, but small and medium enterprises have no choice but to “ask those who bought the product directly.”

However, this is also an area where small and medium enterprises have an advantage. They are closer to their customers. After a purchase, sending a message via LINE asking, “Could you share your thoughts?” will increase the number of reviews. Using tools can also automate this process. Services costing a few thousand yen per month are sufficient.

3. Tell the Story of ‘Why This Product’

Generative AI tends to refer to background information about products, not just specifications. Information like “a miso brewery with 80 years of history that makes miso only from local soybeans” can serve as a differentiating factor when AI suggests a “miso soup set suitable for tailgate parties.”

This is difficult for large companies to replicate. Adding a “story” to mass-produced products in factories will be recognized by AI as inconsistent information. The genuine stories of small and medium enterprises are a competitive advantage in themselves.

4. Provide Product Information That Hits Niche Needs

AI responds to specific requests like “for 25 people at a tailgate” or “gluten-free within a budget of 3,000 yen.” In such cases, products from small and medium enterprises that perfectly match specific needs may be chosen over generic products from large companies.

However, there is a condition. That information must exist digitally.

“Our product is gluten-free”—if this is not written anywhere on the site, AI cannot make that judgment. It may seem obvious, but when looking at actual small and medium enterprise e-commerce sites, it is surprisingly common to find such basic information missing.

5. Prioritize Product Registration on Platforms

AI from Instacart and Shipt selects from product databases on their own platforms. In other words, if your product is not registered on the platform, it won’t even be considered.

In Japan, the completeness of product page information on platforms like Amazon, Rakuten, and Yahoo! Shopping is likely to directly impact AI assistant recommendations in the future. Monthly listing fees range from a few thousand to tens of thousands of yen. Skimping on this investment and missing out on annual sales opportunities worth hundreds of thousands of yen is not worth it.

The Real Threat Is ‘Being Ignored by AI’

Some readers may think, “This doesn’t concern us because we don’t sell online.”

However, consider this.

As AI agents like Meta’s Muse become prevalent, consumers will start asking AI to “buy the ingredients for this week’s dinners.” The AI will complete the order online. The frequency of visiting physical stores will undoubtedly decrease.

In the U.S., the online grocery market is projected to reach about $150 billion (approximately 22 trillion yen) by 2024, with predictions of exceeding $200 billion by 2027. In Japan, while the food e-commerce penetration rate is still in the 4% range, the spread of AI agents could rapidly boost this figure.

The real threat for small and medium enterprises is not being ‘chosen’ by AI. It is ‘not existing’ in AI’s selection candidates.

So, What Should Be Done?

It’s not a difficult topic.

1. Present your product information digitally in a ‘format readable by AI.’
2. Collect reviews diligently.
3. Write specifically about ‘why this product.’
4. Register products on major platforms.

All of these can be done for a few thousand yen if outsourced, or nearly zero yen if done in-house.

Even if you can’t run a 3 million yen TV commercial, you can organize product page information. Even if you can’t send a nationwide sales team, you can still get listed in AI’s recommendation lists.

The era of AI handling shopping is both a threat and an unprecedented opportunity for small and medium enterprises. The “wall” of large corporate advertising budgets is being replaced by a “flat arena” of AI recommendation algorithms.

However, to enter that arena, minimal preparation is necessary. Digitally organizing product information—this is the first step in ‘shelf acquisition’ in the AI era.

Start by opening your product page and reading it from the perspective of AI. “Can this information recommend this product to someone?” If the answer is No, then what you need to do today is clear.