At BrightonSEO San Diego, Jordan Koene of Previsible offered a provocative prediction about the future of ecommerce: Your next customer may never visit your website.
That does not mean websites will disappear tomorrow. It means their role in the customer journey is changing. Instead of browsing multiple sites, comparing products, reading reviews, and completing transactions themselves, consumers will increasingly delegate that work to AI agents.
For retailers and marketers, the implications extend far beyond adding a chatbot or generating more content. Success in agentic commerce will depend on whether AI systems can find, understand, trust, and recommend a company's products.
Agents Don't Browse, They Match Data
Koene's simplest explanation of AI shopping was also one of his most useful:
Agents don't browse. They match data.
An AI agent may already know a shopper's budget, preferences, location, schedule, loyalty memberships, and purchase history. Its job is to match those requirements with the products most likely to satisfy them.
Imagine someone looking for an engravable gift for a knitter. The shopper has a budget of $30 to $50, needs the gift by a specific date, and prefers handmade products from small businesses. The agent must match those criteria against product data such as:
- Product attributes and descriptions
- Price and inventory
- Shipping time
- Ratings and reviews
- Seller reputation
- Structured metadata
The retailer controls the product side of that equation. If its data is incomplete, outdated, inconsistent, or difficult for machines to interpret, the product may never become part of the agent's recommendations.
Quality and Scale Must Work Together
Koene described two ways companies commonly get AI commerce wrong.
The first is pursuing scale without quality. Producing enormous volumes of automated content may create more pages and data, but it also amplifies weak signals. Generic FAQs, thin category pages, and repetitive product descriptions become noise that agents cannot trust.
The second is achieving quality without scale. A company may improve a small number of pages or create a few excellent AI experiences, but those isolated successes never extend across the full catalog.
Koene described these as "boutique wins": valuable experiments that fail to compound.
True AI shopping leadership requires both: Quality x scale.
When trusted information is supported by systematic execution, a brand can become a reliable answer across search engines, AI assistants, shopping agents, and other generative experiences.
Koene presented a framework based on two dimensions: data readiness and execution scale.
- Invisible: Low trust, limited repeatability, gaps in structured data and ownership, and inconsistent pages.
- Noisy scale: Large volumes of weak or irrelevant automated content.
- Boutique wins: High-quality work with limited reach across the catalog.
- AI shopping leaders: Trusted data, systematic execution, consistently refreshed information, continuous experimentation, and optimization for recommendations rather than rankings.
Agentic Commerce Is Promising, but Still Early
The technical infrastructure for agentic commerce is developing faster than consumer adoption.
Koene discussed a large Agentic Commerce Protocol implementation covering approximately 180 million products, 60 markets, and 11 languages. The project took roughly two months and required about $500,000 in infrastructure investment.
Although transactions increased around Black Friday, the system produced only about 3,000 sales per day. That may sound substantial, but it was not financially significant at the retailer's scale.
The lesson was not that agentic commerce had failed. It was that technical capability does not automatically create customer demand. Companies must balance experimentation with realistic expectations and connect their investments to meaningful business results.
Writing for Agents
Traditional SEO advice has emphasized writing for people rather than search engines. Koene deliberately challenged that idea by telling the audience to begin writing for agents.
His point was not that human readers no longer matter. It was that AI systems will increasingly mediate the relationship between companies and consumers. Content must therefore be structured so that machines can accurately retrieve, interpret, compare, and recommend it.
In a CarGurus case study, product overview pages were reorganized around the information AI models needed to understand individual vehicles. The updated model pages generated approximately 31,000 additional impressions.
The goal was not simply to improve conventional rankings. It was to increase the likelihood that CarGurus information would appear during AI-assisted vehicle research.
Etsy: Written by Agents, Judged by Agents, Approved by Humans
Etsy presents an unusually difficult data challenge. Its marketplace contains millions of distinctive, handmade, and sometimes highly unconventional products. Many lack standardized identifiers, specifications, or sizing conventions.
Koene illustrated the problem with one memorable example: chicken hats. Even a category that unusual must have enough organized information for an AI agent to understand the products and match them with interested customers.
Etsy addressed the challenge through a multi-stage agentic content system:
- Etsy catalog and listing data were combined with AI-assisted market research.
- Writer agents drafted unique FAQs for different page groups in more than 11 languages.
- LLM judges and moderation systems evaluated each answer for quality, safety, and accuracy.
- Human reviewers made the final approval or rejection decision.
- Rejected content entered a nightly revision queue.
- Approved content could be deployed across millions of live pages and multiple markets.
The system was deployed across approximately 3.5 million category and product pages and contributed to a 3% increase in visits and engagement.
This example also clarified the continuing role of people. At this scale, human reviewers cannot write or individually inspect every piece of content. Instead, they establish the rules, supervise the agents, review exceptions, and retain authority over what gets published.
Preparing for the "Cartless Buyer"
Etsy uses the term "cartless buyer" for someone who discovers, evaluates, and purchases a product through an AI agent without visiting Etsy's website.
This is the customer Koene believes companies must begin preparing for now.
Traditional measures such as rankings, clicks, bounce rates, and time on site become less meaningful when the customer never arrives on the site. Companies will need to measure whether their products appear in AI recommendations and whether that visibility ultimately generates sales.
Koene summarized the new model with a straightforward formula:
Data quality x scale = AI visibility = revenue
The ecommerce winners will not necessarily be the companies producing the most content or receiving the most website visits. They will be the companies with reliable, structured, current information and the systems required to make that information available everywhere AI agents are helping customers make decisions.