Conclusion First: The ‘Buyer’ Will No Longer Be Human

ChatGPT has taken the wallet.

OpenAI’s newly announced “ChatGPT Wallet” enables AI agents to search for, compare, and complete purchases of products on behalf of users. Currently, this feature is limited to Pro users in the U.S., requiring user confirmation before each purchase, but the direction is clear.

The ‘主体 of shopping’ is shifting from humans to AI.

This is not just a discussion about payment methods. For small and medium enterprises (SMEs), it signifies a fundamental change in the structure of order placement itself.

What Will Change? — The Shocking Reduction of Order Costs to ‘Almost Zero’

Let’s first look at it from the ordering side.

Imagine the purchasing operations of an SME. An administrative staff member opens multiple supplier websites, compares prices, checks inventory, creates purchase orders, obtains approvals, and sends them out. Each order takes about 30 minutes to an hour. If there are 100 orders a month, that amounts to 50 to 100 hours of work. Calculating at an hourly wage of 1,500 yen, that translates to 75,000 to 150,000 yen a month. Over a year, that adds up to 900,000 to 1,800,000 yen just for ‘placing orders.’

What happens when an AI agent takes over this process?

By simply prompting the AI with specifications, budget, and deadlines, it can generate candidates, compare them, and suggest options that meet the criteria. With just a click of a confirmation button, it takes less than five minutes per order. For 100 orders a month, that totals just 8 hours of work, costing only 12,000 yen a month. Annually, that amounts to 144,000 yen.

The annual cost drops from 900,000 to 1,800,000 yen to just 144,000 yen. This represents an 80-90% reduction.

This is what it means for ‘order costs to become almost zero.’ Large corporations have been doing similar things with ERP and EDI systems, but those implementations cost millions to tens of millions of yen. Now, with a ChatGPT subscription costing just 200 dollars (about 30,000 yen) a month, this is a game-changing structure for SMEs.

The Real Fear Lies in the ‘Receiving Side’

Now we get to the main point.

If the ordering side becomes AI, it means that AI will also be the one to ‘choose’ your company’s products.

Human purchasing agents might decide on suppliers based on long-term relationships, the personality of sales representatives, hospitality, or simply inertia of ‘going with the usual supplier.’ Much of the order volume for SMEs has relied on such ‘relationships.’

AI does not consider such factors.

What AI looks at are structured data: prices, delivery times, specifications, reviews, inventory status, and information accessible via APIs. In other words, companies that do not have accurate product information structured on their websites will not even be considered by AI.

This is not just a matter of ‘let’s improve our SEO.’ It represents a more fundamental change.

Three Conditions to Be ‘Chosen’ by AI

1. Product information must be machine-readable — Websites that only have PDF catalogs or images cannot be read by AI. It will be necessary to implement JSON-LD, structured data, and API endpoints.
2. Prices and conditions must be explicitly stated — ‘Please contact us’ is synonymous with ‘no information’ for AI, which will exclude it from comparisons.
3. Inventory and delivery times must be referenced in real-time — AI cannot make judgments based on ‘probably will arrive next week.’

How many SMEs currently meet these three criteria on their websites? Probably less than 10%.

In other words, 90% of SMEs risk not appearing on the ‘AI shopping list.’

The Cost of Placing Products on the ‘AI Shelf’

So, how much will it cost to adapt?

Let’s talk about realistic figures.

• Adding structured data (Schema.org) to existing websites: 100,000 to 300,000 yen (if outsourced)
• API-izing product information (basic version): 200,000 to 500,000 yen
• Automatic linking of inventory and delivery times (connecting to existing systems): 300,000 to 1,000,000 yen

In total, a full set would cost between 600,000 and 1,800,000 yen. This is not a trivial investment for SMEs, but compared to the risk of losing several million yen in orders by not being ‘chosen by AI,’ it is well worth it.

Moreover, AI can be utilized here as well. Generating structured data, creating API specifications, and formatting existing data are all areas where LLMs excel. If done in-house, costs could potentially be reduced to less than half.

A Risk Not to Be Overlooked — AI Can Be Deceived

Here, let me throw a cold splash of water.

In a world where AI agents make purchasing decisions, there is a new risk: attacks that intentionally distort AI judgments.

Recent research (2024, a collaborative study among multiple universities) has demonstrated that AI model decision-making can be significantly distorted by misinformation. For instance, generating large volumes of fake reviews about competitors’ products to manipulate AI evaluations, or disseminating false benchmark data favorable to one’s own products.

What a human might instinctively reject as ‘suspicious,’ AI will accept as data without question.

SMEs should take two measures.

As the ordering side: Do not take AI proposals at face value. Maintain a human check for final decisions. Especially for high-value orders or new suppliers, implement a system to verify the reasoning behind AI’s judgments. The design of ChatGPT Wallet, which includes ‘confirmation every time,’ embodies this philosophy.

As the receiving side: Accumulate data that can prove your company’s reliability. Achievements, certifications, third-party reviews, transaction histories — make these materials available in a structured format so that AI can determine ‘this company is trustworthy.’

The Agent Economy is Already in Motion

LangChain is valued at 3 billion dollars, with over 7,000 customers. This pertains to the ‘development platform’ for AI agents, and the number of agents operating on top of it is exponentially larger.

Shopify has already begun developing commerce APIs for AI agents. Stripe is building a system for ‘inter-agent payments.’ Amazon, Google, and Apple are all heading in the same direction.

‘AI shopping’ is not a distant future; the infrastructure is being built right now.

The question is whether ‘your products will be listed on this infrastructure.’

So, What Should You Do?

You only need to do three things.

1. Structure your product information on your company website. Start with Schema.org’s Product type markup. If you do it yourself, it costs nothing; if outsourced, it will be around 100,000 yen. Just this step will significantly increase the chances of being ‘visible’ to AI.

2. Reduce ‘please contact us’ instances. Clearly state prices, delivery times, and minimum order quantities — as much information as possible that AI can use for comparisons. If you can’t provide everything, at least include price ranges and standard delivery times.

3. Try placing an order through an AI agent. Whether it’s ChatGPT Plus or Claude, do a test run where ‘you let AI search for and compare the consumables you usually buy.’ This will give you a firsthand experience of how your website appears from the AI’s perspective.

The important thing is not to aim for perfection. The goal is to meet the minimum conditions to appear on the ‘AI shopping list’ by the end of this month.

In six months, when ordering through AI agents becomes commonplace, the gap between companies that were prepared and those that were not will be irreparably wide.

Change is always over before you realize it.