DEV Community: Dor Shany

Your Product Data Is Your Most Valuable AI Asset (And Most Retailers Are Wasting It)

Dor Shany — Tue, 31 Mar 2026 06:47:14 +0000

Product data enrichment for AI commerce refers to the process of expanding structured product attributes (material, dimensions, certifications, care instructions) beyond basic catalog fields so that AI shopping agents can accurately match, compare, and recommend products to consumers. As AI-driven shopping scales rapidly, enriched product data has become the primary factor determining whether an agent surfaces your products or a competitor's.

Every retailer has product data. Titles, prices, descriptions, images, maybe some categories. It sits in a PIM, gets pushed to a website, and occasionally gets cleaned up when someone notices errors.

That product data is now the single biggest factor determining whether AI shopping agents recommend your products or your competitor's. And most retailers are treating their most valuable AI asset like a maintenance task.

Why Has Product Data Become the Competitive Moat?

When customers shopped through search engines, product data was important but not decisive. A well-optimized title tag and strong backlink profile could overcome mediocre product attributes. Marketing spend could buy visibility. Brand recognition drove clicks.

AI shopping agents changed the equation. When ChatGPT recommends running shoes or Google AI Mode suggests a winter jacket, the agent evaluates structured product attributes directly. It bypasses your brand story and marketing spend, comparing data points directly: material composition, weight, dimensions, ratings, reviews, shipping speed, return policy, sustainability certifications.

The retailer with richer, more accurate, more comprehensive product data wins the recommendation, consistently.

What Is the Product Data Gap in Ecommerce?

We've analyzed catalogs across retail sectors, and the pattern is consistent. The average ecommerce product listing has 5 to 8 structured attributes: title, price, description, image URL, availability, category, maybe brand and SKU.

AI shopping agents evaluate products across 30 or more attributes when making recommendations. The gap between what retailers provide and what agents need represents lost revenue. For every missing attribute, the AI agent's confidence in recommending that product drops.

Here's what that looks like in practice:

A customer asks ChatGPT: "Find me a merino wool sweater, medium weight, machine washable, under $150, available in navy."

The AI agent needs to match on: material (merino wool), weight class (medium), care instructions (machine washable), price (under $150), color (navy), and availability. If your sweater listing only has title, price, and a paragraph that mentions "luxurious wool blend" somewhere in the description, the agent can't confidently match on material (is it 100% merino? a blend?), can't determine weight class, can't confirm care instructions, and might not parse the color options.

Meanwhile, a competitor whose listing has structured fields for material: 100% Merino Wool, weight: 280g, care: Machine wash cold, colors: [Navy, Charcoal, Oatmeal, Forest] matches perfectly. The competitor gets recommended. Your product doesn't.

What Are the Five Dimensions of AI-Ready Product Data?

Building AI-ready product data requires attention to five key dimensions. For a broader look at how this data feeds into the Agentic Commerce Optimization (ACO) discipline, see my article on ACO.

1. Attribute Depth

Go beyond the basics. For every product, you should have structured data for physical characteristics (material, dimensions, weight), usage context (occasion, activity, season, environment), care and maintenance, compatibility and requirements, certifications and standards, and comparison-relevant metrics.

The target: 30+ structured attributes per product, minimum. Category leaders in AI recommendations consistently have 40 or more.

2. Attribute Specificity

"High quality materials" is useless to an AI agent. "18/10 stainless steel, 2.5mm tri-ply construction" is actionable. Every attribute should be specific enough that an AI agent can use it for comparison filtering.

Rules of thumb:

Numbers over adjectives (280g, not "lightweight")
Specific materials over categories (100% organic cotton, not "cotton blend")
Measurable dimensions over relative sizes (41cm x 30cm x 15cm, not "medium sized")
Standards references where applicable (OEKO-TEX Standard 100, GOTS certified)

3. Structured Format

Structured product data uses machine-readable formats like JSON-LD and typed API fields rather than burying attributes in free-text descriptions. AI agents can extract some information from descriptions, but structured fields are dramatically more reliable.

For your website: JSON-LD Product schema with additionalProperty name-value pairs for every attribute beyond the basic Product fields.

For feeds: Google Merchant Center product_detail attributes, complete shipping and return information, all applicable Google product categories.

For APIs: Typed, filterable fields in your Storefront API or headless commerce layer. If you're on Shopify, Agentic Storefronts expose your Storefront API to AI agents, but only expose what's already in your catalog. Platforms like Paz.ai help retailers ensure their product data is structured and accessible to AI shopping agents across channels.

4. Accuracy and Freshness

AI agents that recommend an out-of-stock product or display a wrong price lose user trust. They learn quickly to deprioritize data sources that produce bad results. Stale inventory data can damage your AI recommendation placement significantly.

Real-time or near-real-time data syncing is becoming table stakes. Daily feed updates used to be sufficient for Google Shopping. AI agents making real-time product comparisons need current data.

5. Consistency Across Channels

Your product data appears in multiple AI systems simultaneously: Google AI Mode (via Merchant Center), ChatGPT (via web crawling and feeds), Perplexity (via web crawling), Bing Copilot (via Bing index), and emerging agents. Inconsistent data across these channels, such as different prices, availability, or attributes, reduces trust signals across all of them.

A single source of truth for product data, syndicated to all channels with real-time updates, is the operational foundation of AI commerce readiness.

What Is the ROI of Product Data Enrichment?

Quantifying the return on product data enrichment is straightforward once you track AI-originated revenue separately.

Based on early industry benchmarks, retailers who enrich their top SKUs from an average of 7 structured attributes to 35+ typically see significant increases in AI-originated orders within the first few months. In my experience, enrichment costs range from roughly $3 to $15 per SKU depending on the mix of automated extraction and manual review, and the incremental revenue from AI channels can exceed the investment within one quarter.

The economics improve at scale. Once you establish enrichment templates by product category, the per-SKU cost drops toward the lower end of that range. And the enriched data improves performance across all channels, not just AI shopping, because better structured data also improves Google Shopping performance, marketplace listings, and on-site search.

How Should Retailers Start the Data Enrichment Process?

Phase 1: Audit (Week 1)

Export your full catalog. For each product category, count the average number of structured attributes. Identify the gap between what you have and what AI agents need. Prioritize your top revenue-generating 20% of SKUs.

Phase 2: Template Creation (Week 2)

For each product category, define the target attribute set. What does an AI agent need to confidently recommend a product in this category? Build enrichment templates that standardize the format for each attribute.

Phase 3: Enrichment (Weeks 3-6)

Start with the top 20% of SKUs. Use a combination of automated extraction (from existing descriptions, manufacturer data, and product pages) and manual enrichment for attributes that require human judgment.

Phase 4: Structured Data Implementation (Weeks 7-8)

Push the enriched data to all channels: JSON-LD on product pages, expanded Merchant Center feeds, Storefront API, and any third-party product data syndication platforms.

Phase 5: Monitoring (Ongoing)

Track AI recommendation rates for enriched vs. non-enriched products. Monitor data freshness and accuracy. Expand enrichment to the next 20% of the catalog.

You can achieve all of this automatically with Paz.ai - sign up for your free account today.

Why Does Product Data Enrichment Compound Over Time?

Product data enrichment compounds. Each enriched product improves your overall domain's trust signal in AI recommendation systems. As AI agents learn that your data is consistently rich, accurate, and current, they increase the frequency of your recommendations across all products, including ones not yet fully enriched.

The retailers investing in product data enrichment today are building a competitive moat. According to Juniper Research, AI in ecommerce is projected to grow from approximately $7 billion in 2024 to over $20 billion by 2028. The winners in that expanding market will be the brands whose product data was rich enough to earn recommendations from AI shopping agents from the beginning.

Product data has always mattered. In AI commerce, it's the primary factor that determines whether you get recommended or ignored.

Frequently Asked Questions

What is AI-ready product data?
AI-ready product data is structured, machine-readable catalog information with 30 or more attributes per product (material, dimensions, care, certifications) that enables AI shopping agents to accurately compare, filter, and recommend products to consumers.

How many structured attributes should each product have?
The minimum target is 30 structured attributes per product. Category leaders in AI-driven recommendations typically have 40 or more. Focus on attributes that enable direct comparison: numeric measurements, specific materials, certifications, and compatibility details.

Does product data enrichment only help with AI shopping agents?
No. Enriched structured data improves performance across all digital channels, including Google Shopping, marketplace listings, on-site search, and traditional SEO. The investment pays dividends well beyond AI commerce.

How long does it take to see results from product data enrichment?
Most retailers see measurable improvements in AI-originated traffic and orders within 60 to 90 days of enriching their top SKUs, assuming the data is properly syndicated to AI-accessible channels.

Dor Shany is the CEO of Paz.ai, an agentic commerce optimization platform that helps retailers sell through AI shopping agents. This article reflects his analysis of publicly available information. More at paz.ai/blog.

The Zero-Click Commerce Playbook: What Retailers Need to Know When Customers Stop Clicking

Dor Shany — Tue, 31 Mar 2026 06:33:24 +0000

Zero-click commerce is what happens when customers make purchase decisions inside AI interfaces - ChatGPT, Google AI Mode, Perplexity - without ever visiting a retailer's website. Research from Semrush shows over 93% of AI Mode product searches end without a single click to an external site. ChatGPT, with 900 million weekly active users, shows product cards with images, prices, and reviews inside the conversation. Perplexity does the same.

The click - the foundation of digital commerce measurement for 25 years - is becoming optional in AI-driven shopping.

How Big Is the Zero-Click Commerce Shift?

This isn't a niche behavior. Grand View Research projects the AI-powered shopping assistant market will grow from $3.36 billion to $28.54 billion by 2033 at a 27% CAGR. Shopify reports a 15x increase in AI-originated orders over the past year, per President Harley Finkelstein. Industry research suggests that roughly a quarter of Gen X and over a third of Gen Z consumers have already used AI tools for shopping tasks.

For retailers, the implication is straightforward: a growing percentage of your addressable market is making purchase decisions inside AI interfaces where your website, your brand experience, and your conversion-optimized product pages are invisible.

Why Do Traditional Ecommerce Metrics Break in Zero-Click?

The standard ecommerce funnel assumes a sequence: impression, click, page view, add to cart, purchase. Attribution models, ROAS calculations, and marketing budgets all depend on tracking this sequence.

Zero-click commerce breaks every step after the impression. When a customer asks ChatGPT "find me running shoes for flat feet under $150" and the AI shows three product cards, the retailer whose shoe gets recommended had no impression in the traditional sense. There was no ad, no organic listing, no click. The AI parsed the retailer's product data, determined it was relevant, and surfaced it.

When that customer taps "buy," the AI redirects them to the merchant's site to complete the purchase. The product discovery, comparison, and decision all happened inside the AI interface. By the time the retailer's site loads, the customer already knows exactly what they want. Traditional analytics sees a direct visit or a referral with no campaign context. No UTM parameter. No attribution trail.

Marketing teams measuring performance through click-based attribution will see a growing gap between actual revenue and tracked revenue. The gap is the zero-click channel.

What Determines Visibility in Zero-Click Environments?

AI shopping agents make recommendation decisions based on a different set of signals than search engines. Understanding these signals is the new competitive advantage.

Product data richness. AI agents evaluate products based on structured attributes. A product with comprehensive structured data points (material, dimensions, weight, care instructions, compatibility, certifications) consistently outperforms a product with only basic attributes, even when the basic product is cheaper. The AI can't recommend what it can't evaluate. For a deep dive on building AI-ready product data, see my article on why product data is your most valuable AI asset.

Feed quality and freshness. Google AI Mode pulls from Merchant Center feeds. ChatGPT uses its own product index built from feeds and web data. Stale prices, incorrect availability, missing attributes: any data quality issue disqualifies a product from recommendation. Retailers investing in feed optimization see measurable improvements in AI visibility.

Third-party signals. AI agents weight reviews, ratings, and editorial mentions when ranking recommendations. A product with 500 verified reviews and a 4.5 rating gets preferred over one with 10 reviews, regardless of other factors. This is similar to traditional SEO, but amplified because the AI is making the choice for the customer.

Structured data on product pages. Rich JSON-LD markup with complete Product schema (including additionalProperty, aggregateRating, offers with shipping details) gives AI crawlers the information they need. Most ecommerce sites implement the bare minimum.

The Playbook: How Should Retailers Adapt?

1. Accept the Zero-Click Reality

Stop trying to "drive traffic" from AI channels. The traffic isn't coming. Instead, optimize for the recommendation itself. Your goal is to be one of the three products the AI shows, not to get a click-through to your website.

This requires a mental model shift across marketing, analytics, and executive teams. Revenue that comes through AI channels may never show up in Google Analytics. You need new measurement approaches.

2. Treat Product Data as a Strategic Asset

If your catalog has SKUs with just a title, price, and paragraph description, you're competing with a blindfold on. AI agents need structured, machine-readable data to make recommendations.

Start with your top 20% of products (by revenue). Enrich each one with comprehensive structured attributes. Then work through the rest of the catalog. This is the highest-leverage investment you can make in AI commerce readiness.

3. Optimize Feeds, Not Just Pages

Your Google Merchant Center feed is the primary data source for Google AI Mode. Your product pages matter for ChatGPT's web crawling. Both need attention.

For Merchant Center: use product_detail attributes for every characteristic an AI might use to compare products. Don't rely on free-text descriptions for structured information.

For product pages: implement expanded JSON-LD with additionalProperty fields for each attribute. Add FAQ schema where relevant. Front-load specific, verifiable facts in the first paragraph of each product description.

4. Build Third-Party Signals

AI agents heavily weigh information from sources other than your own website. This means reviews on Google, Amazon, and niche platforms. It means editorial mentions in trusted publications. It means being discussed in relevant online communities.

If the only source of information about your products is your own website, AI agents discount it. They're trained to triangulate from multiple sources. Building a third-party presence is the single highest-impact action for zero-click visibility.

5. How Should You Monitor AI Recommendations?

Search for your products on ChatGPT, Google AI Mode, and Perplexity using the language your customers use. "Best waterproof hiking boots under $200." "Organic baby clothes that are actually soft." "Wireless headphones for working out."

Track which products appear, which competitors get recommended, and how recommendations change over time. This is the new equivalent of checking your search rankings, but for the AI channel. Paz.ai offers monitoring tools that automate this process across multiple AI platforms.

6. Implement llms.txt

Add an llms.txt file to your domain root. This emerging standard (similar to robots.txt) tells AI crawlers what your site offers, your key product categories, and where to find structured information. It's a simple file that takes an hour to create and significantly improves AI crawlability.

What Comes Next for Zero-Click Commerce?

The zero-click trend will accelerate. Voice shopping through AI assistants removes even the visual interface. Multi-agent systems that compare across retailers in milliseconds raise the bar for data quality further. Commerce protocols like Google's UCP are standardizing how AI agents discover and present products, while OpenAI's ACP handles product discovery and redirects shoppers to merchant sites for checkout.

Retailers who adapt their data, measurement, and strategy now, while most competitors are still optimizing for clicks, build the visibility advantage that compounds as this channel grows.

The sale still happens in zero-click commerce. The path to it has changed.

Frequently Asked Questions

What is zero-click commerce?

Zero-click commerce refers to the growing trend of consumers making purchase decisions entirely within AI interfaces like ChatGPT, Google AI Mode, and Perplexity, without clicking through to a retailer's website. The AI handles product discovery, comparison, and recommendation. The consumer then completes the purchase on the merchant's site, but the decision was already made inside the AI.

How do AI shopping agents decide which products to recommend?

AI agents prioritize products with rich structured data, high-quality and fresh product feeds, strong third-party signals (reviews, ratings, editorial mentions), and complete schema markup. Products lacking these signals are effectively invisible to AI recommendation engines.

Does zero-click commerce mean retailers lose control of the customer experience?

Partially, yes. The discovery and comparison phases move inside AI interfaces that retailers don't control. However, retailers still own the checkout experience, post-purchase communication, and brand relationship. The key shift is that winning the recommendation inside the AI becomes as important as ranking in traditional search results.

How can retailers measure revenue from AI channels?

Traditional click-based attribution often misses AI-originated sales. Retailers should monitor referral traffic from AI platforms (chat.openai.com, gemini.google.com, perplexity.ai), track direct visits that follow AI interaction patterns, and use tools designed for AI commerce analytics to close the measurement gap.

Why Your SEO Team Needs to Learn Agentic Commerce Optimization (ACO)

Dor Shany — Tue, 31 Mar 2026 06:32:09 +0000

Agentic Commerce Optimization (ACO) is the practice of optimizing product data, feeds, and digital presence so AI shopping agents - ChatGPT, Google AI Mode, Perplexity - recommend and sell your products. It extends traditional SEO to cover the AI commerce channel where, according to research from Semrush, over 93% of AI Mode searches end without a click.

The biggest shopping channel in a generation is growing under your nose, and your SEO team's playbook doesn't cover it.

ChatGPT hit 900 million weekly active users as of early 2026 (per DemandSage's tracking of OpenAI disclosures). Morgan Stanley surveyed U.S. consumers and found that 16% of all Americans have already made a purchase influenced by ChatGPT. Their bull case projects agentic commerce reaching $385 billion in U.S. online spending by 2030. Walmart reported that customers using its AI shopping assistant have a 35% higher average order value (per its Q4 2025 earnings call).

These aren't projections about some distant future. This is happening right now, and there's an emerging discipline designed to address it. I call it Agentic Commerce Optimization, or ACO.

How Did We Get from SEO to ACO?

Every major shift in how people find and buy things online has produced a corresponding optimization discipline. Understanding the progression helps clarify why ACO is different and why it matters.

SEO (Search Engine Optimization) has been the backbone of digital marketing since the late 1990s. You optimize for keywords, backlinks, page speed, and structured data. The goal is traffic. Users search, click a blue link, land on your site, and hopefully buy something. After 25 years, the playbook is mature and well understood.

AEO (Answer Engine Optimization) emerged as Google started answering questions directly in search results. Featured snippets, knowledge panels, voice search results - these changed the game because users could get answers without clicking through. If you optimized for AEO, you were fighting for that answer box at the top of the page. The goal shifted from "get them to my site" to "be the answer Google displays."

GEO (Generative Engine Optimization) showed up alongside ChatGPT and other large language models. When someone asks an AI "what's the best running shoe for flat feet," the AI generates a response that may or may not mention your brand. GEO is about influencing those AI-generated responses through content strategy, brand mentions across the web, and structured information that LLMs can pick up during training or retrieval.

ACO (Agentic Commerce Optimization) goes a critical step further. It optimizes for the entire purchase flow within AI-powered experiences. The user asks ChatGPT for a running shoe recommendation, the AI surfaces product cards with prices and specs, and the user clicks through to purchase. The goal with ACO is not traffic or mentions - it's getting your product surfaced, recommended, and selected by AI agents as the best match for the user's intent.

Where GEO focuses on earning brand mentions in AI responses, ACO targets the full purchase flow - getting your product surfaced, compared, and selected. For a deeper look at the protocols that power these AI shopping experiences, see my guide to ACP, UCP, and MCP.

Why Does Your SEO Playbook Fall Short?

The instinct most marketing teams have is to assume their existing SEO work translates. Some of it does. A lot of it doesn't.

What still matters: Product schema markup, clean site architecture, fast page loads, and accurate structured data remain important. AI agents still crawl the web, and they use the same underlying data that search engines use. If your product pages have solid Schema.org implementation, that's a foundation you can build on.

What doesn't translate: Keyword density, meta tag optimization, internal linking strategies, and most of the tactical SEO work your team does every week have minimal impact on how AI agents evaluate and recommend products. An AI agent deciding whether to recommend your running shoe doesn't care about your title tag or your H1 hierarchy. It cares about whether your product data answers the user's question completely enough to make a confident recommendation.

Consider a concrete example. A traditional SEO-optimized product page might have a title like "Men's Running Shoes - Brand X CloudRunner - Free Shipping." That's great for Google's ranking algorithm. But when an AI agent is comparing your shoe against four competitors to answer "what's the best cushioned running shoe under $150 for someone who runs 30 miles a week," it needs to know the cushioning technology, the intended weekly mileage range, the surface type it's designed for, the weight, the drop, and whether it works with orthotics. Most product pages don't have that information in a structured, parseable format.

AI agents parse product data differently than search crawlers. A Googlebot is looking at HTML structure, links, and keyword relevance. An AI shopping agent is trying to build a complete understanding of your product so it can compare it meaningfully against alternatives and make a recommendation it's confident in. The depth and specificity of your product information becomes the competitive advantage, not your keyword strategy.

What Numbers Should Worry SEO Teams?

Research from Semrush analyzing tens of millions of AI search sessions found over 93% of AI Mode searches end without a single click. That finding was published in early 2026 and it has significant implications for anyone whose strategy depends on organic click-through rates.

Meanwhile, the AI shopping channel is growing fast. Amazon has said Rufus is driving billions in incremental annualized sales, with over 300 million customer interactions. Google's Universal Commerce Protocol is already live with Etsy and Wayfair, with Shopify, Target, and Walmart coming soon. Perplexity launched integrated shopping with PayPal checkout. Every major platform is building commerce directly into the AI experience.

McKinsey estimates generative AI could add $2.6 to $4.4 trillion annually to the global economy, with commerce as one of the largest verticals affected. Gartner forecasts that agentic AI will overtake chatbot spending by 2027.

The shift is accelerating faster than most teams realize. What took Amazon years to build in terms of merchant adoption, AI commerce platforms are compressing into months.

What Should SEO Teams Start Doing Now?

I've talked to dozens of retailers about this over the past year. The ones who are moving fastest tend to start with these practical steps:

1. Audit your AI bot access. Check your robots.txt file. Many sites inadvertently block GPTBot, ClaudeBot, PerplexityBot, and other AI crawlers. If AI agents can't access your product pages, you're invisible in AI shopping results. This is a 15-minute fix that most teams haven't done.

2. Assess your product data depth. Pull up your top 20 products and ask yourself: does the structured data on this page contain enough information for an AI to confidently recommend this product over a competitor's? If the answer is no, you've found your first project. Think about use cases, compatibility, materials, performance specs, and anything a knowledgeable salesperson would mention when helping a customer in a store. Tools like Paz.ai's AI readiness assessment can help benchmark where your catalog stands.

3. Implement llms.txt. This is an emerging standard (similar to robots.txt) that tells AI agents how to interact with your site. It specifies what content is available, how to access product data, and what actions are supported. Early adoption signals to AI platforms that you're ready for agent-driven commerce.

4. Enrich your catalog for conversational queries. People don't ask AI agents the way they type into Google. Nobody says "best running shoes 2026" to ChatGPT. They say "I need a shoe for long runs on pavement, I want something under $150." Your product data needs to support that kind of natural-language matching. Add FAQ content, use-case descriptions, and comparison attributes at the product level.

5. Monitor your AI visibility. Sign up for a free account at Paz.ai to start tracking. Start tracking whether your products appear in ChatGPT, Perplexity, and Google AI Mode responses for relevant queries. This is the ACO equivalent of checking your search rankings. If you're not showing up, you know where to focus. You can start with manual testing today.

6. Join the commerce protocol programs. Google's Universal Commerce Protocol has an early access program and is already live with major retailers. OpenAI's commerce protocol focuses on product discovery and recommendations within ChatGPT, directing users to retailer sites for checkout. The retailers joining these programs now will have a significant head start as AI-driven shopping scales.

Does ACO Replace SEO?

I'm not saying fire your SEO team. Organic search still drives enormous volume, and it will for years. But the growth is happening elsewhere, and the skills that make someone great at SEO - analytical thinking, structured data expertise, understanding how algorithms evaluate content - translate well to ACO work.

The smartest move for most organizations is to expand the SEO team's mandate. Give them ownership of AI visibility alongside search visibility. The data infrastructure work (Schema.org, product feeds, structured content) is similar enough that the same team can own both. The strategic thinking is different, but SEO professionals who've navigated algorithm updates for a decade tend to adapt quickly.

What you can't afford to do is wait. The zero-click rates in AI Mode mean the window where traditional SEO carries the full load is closing. Building ACO capabilities now creates a head start that's hard to close later - similar to how companies that invested in SEO by 2005 shaped the playbook everyone else had to follow.

Frequently Asked Questions

What is Agentic Commerce Optimization (ACO)?
ACO is the practice of optimizing product data, feeds, and digital presence so AI shopping agents like ChatGPT, Google AI Mode, and Perplexity recommend your products. It extends traditional SEO to cover AI-driven commerce channels where the majority of searches now end without a click to any website.

How is ACO different from SEO?
SEO optimizes for search engine rankings and click-through rates. ACO optimizes for AI agent recommendations and product selection. While both rely on structured data, ACO requires deeper product information - use cases, comparison attributes, conversational-query matching - because AI agents evaluate products differently than search crawlers.

Do I need a separate team for ACO?
Not necessarily. The skills that make someone great at SEO (structured data, analytical thinking, algorithm adaptation) translate well to ACO. Most organizations can expand their existing SEO team's mandate to include AI visibility. The key is starting now while the discipline is still emerging.

Which AI shopping platforms should I prioritize?
Focus on the three with the largest user bases: ChatGPT (900 million weekly active users), Google AI Mode (integrated into the dominant search engine), and Amazon Rufus (300 million+ customer interactions). Perplexity is smaller but growing quickly and worth monitoring. Make sure your robots.txt isn't blocking any of their crawlers.

A Retailer's Guide to AI Shopping Protocols: ACP, UCP, and MCP Explained

Dor Shany — Tue, 31 Mar 2026 05:50:14 +0000

ACP (Agentic Commerce Protocol) is OpenAI and Stripe's standard for product discovery and commerce integration with ChatGPT. UCP (Universal Commerce Protocol) is Google's open standard for purchases inside AI Mode and Gemini, backed by over 60 organizations. MCP (Model Context Protocol), originally from Anthropic and now governed by the Linux Foundation, is the infrastructure layer that lets AI agents connect to any commerce system. Together, these three protocols define how products get discovered, carted, and purchased through AI assistants.

Three protocols are going to determine whether your products can be sold through ChatGPT, Google AI Mode, and Perplexity over the next few years. Most retailers I talk to haven't heard of any of them.

One thing I learned from building commerce infrastructure: when new protocols emerge, the companies that implement early get structural advantages that are very hard to close later. We're at that moment in commerce right now.

What Problem Do These Protocols Solve?

Right now, if you want to sell through an AI assistant, there's no standard way to do it. Every platform has built its own approach. ChatGPT has one way of handling product discovery and checkout. Google has another. Perplexity has a third.

For retailers, this creates an ugly situation. Imagine if, in 2005, every search engine required a completely different format for your website. That's roughly where AI commerce is today. Companies like Paz.ai are working on unified integration layers to help retailers avoid building separate implementations for each protocol, but the underlying standards are what make any of this possible.

These three protocols are the industry's attempts to standardize how AI agents discover products, manage carts, process payments, and complete purchases. They come from different companies with different philosophies, and they overlap in some areas while being complementary in others.

What Is ACP (Agentic Commerce Protocol)?

ACP (Agentic Commerce Protocol) is the standard created by OpenAI and Stripe for enabling AI-assisted product discovery, recommendations, and commerce through ChatGPT.

OpenAI and Stripe built ACP, with Shopify as a key launch partner. The protocol was originally designed to let ChatGPT handle the entire checkout process inside the conversation, including payment through Stripe, so the shopper would never visit a website.

In March 2026, OpenAI reversed course. Reporting from The Information revealed that OpenAI discontinued its Instant Checkout feature, moving purchases back to merchant websites and apps. An OpenAI spokesperson said they are "evolving our commerce strategy to better meet merchants and users where they are." Clicking "Buy" in ChatGPT now redirects to the merchant's site or app to complete the purchase.

What still works: ChatGPT shopping is still active and growing. The product discovery and recommendation experience remains strong, with 900 million weekly ChatGPT users (per OpenAI's confirmed figures) getting AI-powered shopping suggestions. ACP as a protocol also still exists and is open-sourced at agenticcommerce.dev. What changed is where the transaction actually happens: on the merchant's site, not inside the chat.

Why this matters: OpenAI discovered that building a full checkout experience inside a chatbot was harder than expected. Payment processing, shipping logistics, returns, customer service disputes, regulatory compliance across jurisdictions - these are hard problems that existing merchant infrastructure already handles well. The pivot validates something important: the real value in AI commerce isn't replacing checkout, it's driving discovery and intent. The infrastructure that connects AI agents to merchant systems (product data, inventory, pricing) matters more than where the "Pay Now" button lives.

Who's integrated with ACP:

Shopify has ACP enabled across its platform, meaning over a million merchants can surface products in ChatGPT. Purchases redirect to the merchant's Shopify storefront to complete
Instacart was one of the first ACP partners, with grocery product discovery in ChatGPT directing users to Instacart's app to order
Target and Etsy have ACP integrations that surface products in ChatGPT conversations, with checkout on their own sites
PayPal announced ACP integration rolling out in 2026

What it means for you: ACP is now primarily a product discovery and recommendation channel. Your products can appear in ChatGPT shopping conversations, but customers complete purchases on your site. This actually simplifies the retailer's side: you don't need to support a separate anonymous checkout flow. You do need rich product data so ChatGPT can recommend your products confidently, and you need your site's checkout experience to be seamless when that traffic arrives.

What Is UCP (Universal Commerce Protocol)?

UCP (Universal Commerce Protocol) is Google's open standard that enables purchases directly inside Google AI Mode and Gemini search results.

Google built UCP with backing from over 60 organizations including major payment networks, commerce platforms, and global retailers. When someone uses Google's AI search to research a product and the AI shows a "Buy" button right in the results, UCP is the protocol handling that transaction behind the scenes.

The defining feature of UCP is discovery-first commerce. It's designed to work within the context of search and research, turning product exploration into direct purchasing. Google announced UCP supports six core capabilities: product discovery, cart management, identity linking (via OAuth 2.0), checkout, order management, and extensible vertical-specific modules.

Who's using it right now:

Etsy and Wayfair were the first live UCP partners
Shopify, Target, and Walmart are in the pipeline for UCP integration (per Google's announcements at their digital advertising events in early 2026)
Google has also launched "Direct Offers" in AI Mode, which lets merchants surface promotions and deals directly in AI-generated search results

What it means for you: If organic search through Google is a significant channel for your business (and for most retailers, it is), UCP is how you maintain relevance as Google shifts from blue links to AI-generated shopping experiences. Research from Semrush analyzing tens of millions of AI search sessions found that over 93% of AI-generated responses produce zero clicks to external sites. If your products can't be purchased inside AI Mode, they may simply not get purchased at all.

What Is MCP (Model Context Protocol)?

MCP (Model Context Protocol) is a platform-neutral communication standard that enables AI agents to connect to external commerce systems, inventory databases, and payment processors.

Originally created by Anthropic (the company behind Claude), MCP is now donated to the Linux Foundation's Agentic AI Foundation. It was co-founded by Anthropic, Block, and OpenAI, with support from Google, Microsoft, AWS, Cloudflare, and Bloomberg.

MCP is the connectivity layer that lets AI agents talk to external tools and services. It standardizes how AI agents connect to external services the way REST APIs standardized web service communication a decade ago. It doesn't handle the shopping experience directly, but it provides the plumbing that lets an AI agent connect to your inventory system, check real-time pricing, verify stock levels, or process a return.

Where ACP and UCP handle the consumer-facing shopping experience, MCP operates at the infrastructure level. It's what enables an AI agent to pull your product catalog, check whether an item is in stock at a specific warehouse, or initiate a payment through a connected service.

The defining feature is that MCP is platform-neutral. Because it sits at the Linux Foundation, no single company controls it. MCP adoption is accelerating rapidly, with major cloud providers (AWS, Google Cloud, Azure) all offering managed MCP hosting.

Who's using it right now:

Visa deployed MCP tools on AWS as part of their Intelligent Commerce partnership, handling identity verification, fraud detection, and settlement
Worldpay launched MCP servers for payment integration with AI agents
Google launched managed MCP servers for Maps, BigQuery, and other services (TechCrunch described it as Google going "all-in on MCP")
Shopify stores can connect to AI agents through MCP for inventory, pricing, and fulfillment data

What it means for you: MCP is the foundational layer. Even if you're not implementing ACP or UCP today, ensuring your commerce infrastructure can communicate with AI agents through MCP is how you stay accessible as the ecosystem evolves. It's also particularly relevant for B2B. Gartner predicts that by 2028, AI agents will handle a significant majority of B2B purchasing workflows, representing trillions in commerce value, and MCP is the protocol that makes that machine-to-machine commerce possible.

How the Three Protocols Work Together

These aren't competing standards in the way that VHS competed with Betamax. They operate at different layers:

Consumer-facing discovery + commerce:  ACP (ChatGPT)  |  UCP (Google)
Infrastructure/connectivity:                   MCP (universal)

MCP is the foundation. ACP and UCP sit on top, handling commerce-specific workflows for their respective platforms. A retailer that wants full coverage across AI shopping channels will eventually need presence in both ACP and UCP ecosystems, with MCP providing the data connectivity underneath. Platforms like Paz.ai are building unified integration layers across all three protocols so retailers don't have to manage each one separately.

It's worth noting that ACP and UCP are evolving differently. UCP is pushing toward full in-platform checkout within Google AI Mode. ACP, after OpenAI's March 2026 pivot, now focuses on product discovery and recommendation within ChatGPT while routing purchases to merchant sites. This difference actually reinforces why MCP matters most at the infrastructure level: regardless of where checkout happens, the AI agent still needs to access your product data, inventory, and pricing.

PayPal's approach is instructive here. They're supporting all three: ACP integration for ChatGPT, UCP membership in the Google coalition, and their own MCP Agent Toolkit for developers. Major payment processors are hedging across all protocols because they expect all three to matter.

Which Protocol Matters Most for Your Platform?

Your existing commerce platform shapes which protocol you should prioritize:

If you're on Shopify: You're in the best position right now. Shopify has integrated with ACP (giving you ChatGPT product discovery), is joining UCP (Google AI Mode access), and supports MCP connections. Your main action item is ensuring your product data is rich enough to perform well when AI agents evaluate it. Shopify's "agentic storefronts" feature makes your store accessible through ChatGPT, Copilot, and Gemini.

If you're on Magento/Adobe Commerce: Adobe is part of the UCP coalition, so Google AI Mode integration should come through that channel. ACP integration will likely require middleware or a third-party connector. Focus on MCP readiness through your product data feeds and APIs, and ensure your Merchant Center feed is complete.

If you're on Salesforce Commerce Cloud (SFCC): Salesforce offers ACP support through its agent commerce capabilities, which gives you a path to ChatGPT product visibility. UCP integration will depend on Salesforce's rollout timeline. The platform also positions SFCC for MCP-based agent interactions.

If you're on a custom or headless platform: You have the most flexibility but also the most work. You'll need to build protocol integrations yourself or through partners like Paz.ai. The upside is you can implement exactly what you need. Start with MCP (broadest applicability) and add ACP or UCP based on where your customers are shopping with AI.

How Should Retailers Prepare for AI Commerce Protocols?

You don't need to implement all three protocols tomorrow. But you do need to start preparing, because the retailers moving now will have meaningful advantages by the time these protocols reach mainstream adoption.

This week:

Check your Google Merchant Center feed. Make sure it's complete, accurate, and includes the native_commerce attribute that UCP will require. If you don't have a Merchant Center account, set one up. This is table stakes for UCP.
Whitelist AI crawlers. Make sure GPTBot, ClaudeBot, and PerplexityBot aren't blocked in your robots.txt. This takes 15 minutes.
Audit your product data depth. AI agents make recommendations based on how completely they understand your products. Pull up your top sellers and ask: does this product listing contain enough detail for an AI to recommend it with confidence over a competitor? If not, start enriching.

This month:

Join the waitlists. Google's UCP has early access programs. OpenAI's ACP program is open to merchants who want their products surfaced in ChatGPT shopping. Getting on them now means you'll have integration support when you're ready to implement.
If you're on Shopify, activate agentic features. Shopify has made agent-readiness a default capability. Make sure it's turned on and that your product data is optimized for conversational queries, not just keyword searches.
Assess your API readiness for MCP. Can an external system query your product catalog, check inventory, and get pricing through an API? If not, that's your infrastructure gap. MCP connectivity depends on having accessible, well-documented APIs.

This quarter:

Run a protocol readiness assessment. Map your current capabilities against what each protocol requires. Identify the gaps. Build a roadmap. Paz.ai offers protocol readiness assessments for retailers evaluating their AI commerce strategy. The retailers who treat this as a strategic initiative (rather than a tactical IT project) will move faster.
Start measuring AI visibility. Track whether your products appear in ChatGPT, Perplexity, and Google AI Mode for relevant queries. This is the baseline you'll improve against as you implement protocol support.

Frequently Asked Questions

What is the difference between ACP and UCP?

ACP (Agentic Commerce Protocol) powers product discovery and shopping recommendations within ChatGPT, built by OpenAI and Stripe. UCP (Universal Commerce Protocol) handles purchases within Google AI Mode and Gemini, backed by Google and 60+ organizations. After OpenAI discontinued Instant Checkout in March 2026, the key difference is that UCP aims to keep the full transaction inside Google's platform, while ACP now focuses on discovery and recommendations in ChatGPT before redirecting to merchant sites for checkout. Both protocols solve the same underlying problem - connecting AI assistants to commerce - but they take different approaches to where the purchase actually completes. For a deeper look at how to optimize for these AI shopping channels, see my article on Agentic Commerce Optimization (ACO).

Do I need all three protocols?

MCP is the foundational layer that all AI agents will use for data connectivity. ACP and UCP are channel-specific. You need ACP if your customers use ChatGPT for shopping discovery, UCP if they use Google AI Mode. Most major retailers will eventually implement all three, but start with whichever aligns with your highest-traffic AI channel and build MCP connectivity as your infrastructure baseline.

How do I get started with AI commerce protocols?

Start with the basics: ensure your Google Merchant Center feed is complete, whitelist AI crawlers in your robots.txt, and audit your product data for depth and accuracy. Then join the early access programs for ACP and UCP. If you want to move faster, platforms like Paz.ai provide multi-protocol integration so you don't have to build each connection from scratch.

What does MCP mean for B2B commerce?

MCP is especially significant for B2B because it enables AI agents to access complex product catalogs, negotiate pricing, check inventory across warehouses, and manage procurement workflows. Gartner predicts that by 2028, AI agents will handle a significant majority of B2B purchasing workflows, representing trillions in commerce value, and MCP is the protocol that makes that machine-to-machine commerce possible.

Did OpenAI kill AI shopping in ChatGPT?

No. OpenAI discontinued the Instant Checkout feature (where you could complete a purchase without leaving ChatGPT), but ChatGPT shopping is still very much alive. Product discovery, recommendations, and comparison shopping all still work through ACP. The change is that when you're ready to buy, you're redirected to the merchant's website or app. This works the way Google Shopping already does: you see products and prices, but you buy on the retailer's site. The discovery layer is often the most valuable part of the funnel anyway. For more on how the shift to AI-mediated shopping affects retailers, see my zero-click commerce playbook.

Why Early Protocol Adoption Matters

Morgan Stanley research found that 16% of American consumers have made a purchase directly influenced by ChatGPT, with broader AI-assisted shopping reaching roughly a quarter of consumers. That's not early adopters anymore. Industry studies indicate a majority of consumers now use AI tools at some point in their shopping journey. The question for retailers has shifted from "will AI shopping matter?" to "will we be ready when our customers expect it?"

The protocol landscape will continue evolving. OpenAI's March 2026 decision to move checkout back to merchant sites is a perfect example. New capabilities will be added, standards will converge in some areas, and the technical requirements will change. But the retailers who build their foundation now (clean data, protocol awareness, API readiness) will adapt to those changes much more easily than those starting from scratch.

I saw this dynamic play out with OAuth adoption in payments. The companies that implemented early influenced how the spec evolved and built integrations that became the reference implementations. The ones that waited inherited someone else's decisions and spent twice as long catching up. The same pattern is forming around ACP, UCP, and MCP right now.

Dor Shany is the CEO of Paz.ai, an agentic commerce platform that helps retailers sell through AI shopping agents. This article reflects his analysis of publicly available information. More at paz.ai/blog.