What Agentic Commerce Means for Ecommerce Websites

AI Is Getting Involved in the Transaction

OpenAI has developed the Agentic Commerce Protocol, or ACP, an open standard intended to let AI agents and merchants exchange the information needed for product discovery and transactions. Merchants can supply structured information about their products, including identifiers, descriptions, images, prices, availability and variants. OpenAI also has Instant Checkout for approved merchants, allowing a transaction to move further along inside ChatGPT while the retailer remains responsible for payment, fulfillment and the customer relationship.

Google is working on the same basic problem from its side. Its Universal Commerce Protocol, or UCP, is an open standard for commerce across AI experiences. Google’s work includes product catalogs, carts and checkout, as well as ways for an AI system to retrieve current information about things like price, inventory and product variants.

There are plenty of technical details behind both approaches, and both will undoubtedly change. The more interesting question for most retailers is simpler: what does your store look like to an AI that is trying to shop it?

That can expose some problems that aren’t particularly obvious when looking at the homepage.

Imagine a Store with Great Design and Bad Product Data

Consider a fictional WooCommerce retailer we’ll call Green Trail Supply. It sells insulated water bottles in several sizes and colors. The site was redesigned last year and looks terrific. There are beautiful photographs, a good mobile experience and a homepage featuring hikers using the company’s bottles in the mountains. But behind the scenes, things aren’t nearly as tidy.

One bottle is called the “GT Adventure Bottle” in one place and the “Green Trail Insulated Bottle” somewhere else. A few colors were set up as separate products years ago, while newer colors are variants. The 20-ounce and 24-ounce sizes are easy to find, but the dimensions are buried in the descriptions. Inventory updates sometimes lag behind the warehouse. And one useful fact—that the 24-ounce bottle fits most standard bicycle bottle cages—isn’t in the product information at all. A customer happened to mention it in a review.

None of this necessarily stops a determined person from buying a bottle.

Now imagine someone asking an AI assistant, “Find me a stainless-steel water bottle under $40 that holds at least 20 ounces and fits in my bike’s bottle cage.”

The attractive homepage doesn’t provide much help with that question. The product data does.

OpenAI’s current product specifications give some indication of the information that matters in this new environment. They include fields for product IDs, titles, descriptions, links, images, availability, prices and brands, with additional information available for variants, attributes, shipping, returns and reviews.

For years, retailers have thought of the product catalog mainly as the material used to build the store. Increasingly, the catalog may have to function as a version of the store itself.

Product Copy May Need to Work a Little Harder

This also puts some pressure on a familiar piece of ecommerce marketing: the product description.

Consider an office chair described this way: “Experience uncompromising comfort with our most versatile seating solution yet.” There is nothing especially wrong with the sentence. It sounds like thousands of other product descriptions. Unfortunately, it doesn’t answer many questions.

Now consider a chair whose product information makes it clear that it supports up to 300 pounds, has adjustable lumbar support, works well for users over six feet tall, arrives fully assembled and carries a ten-year warranty. Those details become very useful when someone asks, “Find me an ergonomic office chair under $900 for someone who is 6’3″, sits at a desk eight hours a day and doesn’t want to assemble it.”

An AI shopping system may be working with information from several places, including product pages, structured catalog data, merchant feeds and integrations. Retailers therefore shouldn’t respond by stuffing awkward lists of specifications into every product description. The better lesson is that useful product facts need to exist somewhere reliable and in a form that other systems can understand.

It sounds obvious. Anyone who has worked on a large ecommerce site knows that it often isn’t.

Even the Warehouse Is Part of This

Inventory accuracy isn’t the exciting part of ecommerce, but agentic shopping makes it unusually important.

Say someone asks for a pair of running shoes in size 10, under $150, that can arrive before a race next weekend. An AI finds a good match because the merchant’s data says size 10 is available for $139. Except it isn’t. The last pair sold yesterday and the inventory feed hasn’t caught up. From the customer’s perspective, the AI gave a bad recommendation. From the retailer’s perspective, the problem started much farther back in the system.

The same problem applies to delivery dates, prices, colors, sizes and other variants. OpenAI’s merchant documentation emphasizes keeping product information current, while Google’s commerce work includes retrieving information such as current inventory, pricing and variants.

In other words, some fairly unglamorous ecommerce housekeeping could become part of whether your products are useful to an AI shopping agent in the first place.

Your Website Isn’t Going Away

It would be easy to take all this too far and conclude that people won’t need ecommerce websites anymore. That’s not what the current technology suggests.

Google provides a particularly useful example. Its commerce infrastructure can support a shopping experience in which a cart is assembled through a Google surface and then passed to the retailer’s website, where the customer continues the purchase.

Think about how odd that would have sounded a few years ago. The customer arrives at your store with a cart already filled. Your website still matters. It just enters the story later.

There are also plenty of purchases where people are unlikely to hand the entire experience over to an AI. If you’re buying a $28 replacement filter for your refrigerator, you probably want the right model at a reasonable price and would be delighted to skip ten minutes of searching. A $4,000 dining table is different. You may want to look closely at the wood, see it in several rooms, read the delivery policy, learn something about the manufacturer and come back three days later after measuring your dining room again.

Retailers will have both kinds of customers, and sometimes the same customer will behave differently depending on what he’s buying.

The merchant also still has plenty to do after an AI helps make the sale. Someone has to process the order, ship the product, handle the return, answer the customer’s question and maintain the account. Under both OpenAI’s and Google’s current approaches, the retailer remains the merchant of record.

A Practical Place for WooCommerce Retailers to Start

Most WooCommerce retailers probably don’t need an “agentic commerce strategy” meeting next Tuesday. They would get more value from looking closely at the ecommerce foundation they already have.

Pick one complicated product in your catalog and pretend you’ve never seen it before. Can you tell exactly what it is? Is the price current? Can you determine which versions are actually in stock? Are size, color, material and other important characteristics stored as attributes, or are they hidden inside a paragraph written four years ago? If there are 35 variations, are they modeled consistently? Does the shipping information live in WooCommerce or in another system? Which system actually knows what’s in the warehouse?

Then try asking questions a real customer might ask. 

“Which of your jackets is waterproof, comes in XL and costs less than $200?”

“Which replacement part works with the 2023 model?”

“Do you have this in walnut, under 60 inches wide, and can I get it before November 15?”

A human employee who knows the catalog might answer those questions immediately. The challenge is making sure your ecommerce systems can answer them too.

That’s useful work even if agentic commerce develops more slowly than expected. Better product architecture can improve filtering, internal search, shopping feeds, paid advertising, marketplaces and integrations today.

What Happens When AI Shops Your Store?

The familiar ecommerce journey isn’t going to vanish overnight. People will still browse stores. They’ll still click ads, look at pictures, abandon carts, come back later and spend an unreasonable amount of time deciding between two nearly identical products. But another kind of customer journey is being built alongside that one.

Someone may tell an AI assistant, “Find me a navy blazer under $400, available in a 44 long, mostly natural fibers, free returns, and I need it by Friday.”

That’s not really a search query. It’s closer to giving instructions to a personal shopper.

To be useful in that situation, the retailer needs to provide more than a photograph of a handsome man wearing the blazer. Somewhere in its systems there needs to be dependable information about the size, color, price, material, availability, return policy and delivery options.

At New Target, we build and support WooCommerce and other ecommerce experiences with both sides of that equation in mind: what customers see and what makes the experience work underneath. Increasingly, that means product architecture, structured data, APIs and integrations deserve as much attention as the pages customers browse. For a long time, ecommerce teams have asked, “How do we get more people to shop on our website?”

It’s worth adding a second question now: What happens when their AI shops our store for them? Contact us. We can help.