AI shopping agents can only recommend what they can see. For most merchants, that means their products are invisible.

Recent research from Kaiser and Schulze analyzed 973 e-commerce websites with $20 billion in combined annual revenue. Their findings challenge the hype: ChatGPT referrals account for less than 0.2% of all e-commerce sessions, and when traffic does arrive, it underperforms every major channel except paid social. Affiliate links convert 86% better than ChatGPT referrals. Organic search outperforms ChatGPT by 13%.

The problem isn't that consumers don't want AI-assisted shopping. It's that most product catalogs aren't built for agents to read.

This gap represents both a warning and an opportunity. The merchants who fix their data infrastructure now will capture disproportionate share as agentic commerce scales. Those who wait will find their products permanently excluded from the fastest-growing discovery channel in retail.

Why AI shopping agents can't find most products

AI shopping agents don't browse your website like humans do. They don't scroll through category pages, read marketing copy, or respond to hero banners. Instead, they query structured data feeds, parse machine-readable markup, and make decisions based on attributes they can programmatically evaluate.

This creates a fundamental visibility problem. Most e-commerce infrastructure was built to persuade humans, not inform algorithms.

When a customer asks ChatGPT to "find waterproof hiking boots under $200 that ship by Friday," the agent needs to evaluate:

• Product identifiers (GTINs) to match against its database
• Structured attributes (waterproof rating, price, shipping speed)
• Real-time availability (is it actually in stock?)
• Fulfillment data (can it arrive by Friday?)

If any of these data points are missing, incomplete, or inconsistent, the agent skips your product entirely. There's no second chance. No "close enough." The agent moves to a competitor with cleaner data.

The Kaiser and Schulze research confirms this pattern. Despite 50,000+ transactions from ChatGPT referrals in their dataset, the conversion rates lagged traditional channels—evidence that even when agents do recommend products, the underlying data quality creates friction that kills conversions.

The specific data problems merchants face are well-documented:

• Missing GTINs: Without valid Global Trade Item Numbers, agents can't reliably match products against their database
• Incomplete descriptions: Thin or generic product copy forces agents to skip rather than guess
• Stale inventory: Products showing as available in agent recommendations but out of stock at checkout
• Variant confusion: Size and color options that aren't properly represented in feeds
• Price mismatches: Different prices across channels that trigger transaction failures

One merchant resource noted that "if your data fill rate is 95%+, an agent is less likely to skip your item. If it's low, you risk the AI overlooking or misrepresenting your offerings." The bar is high because agents have no tolerance for ambiguity.

What AI agents look for in product data

The Agentic Commerce Protocol (ACP) published by OpenAI defines exactly what product data agents need. Understanding these requirements is the first step to visibility.

Required fields

Every product in your feed must include:

‍

Missing any required field means the product won't be indexed.

Recommended fields that drive AI product discovery

Beyond the minimum, these fields determine whether agents recommend your products over competitors:

‍

The GTIN requirement deserves emphasis. Research on product feed optimization found that a single missing GTIN is enough for AI agents to skip a product entirely. If you sell products without GTINs (custom items, handmade goods), you need alternative identification strategies—but for standard retail inventory, GTINs are non-negotiable.

Schema.org markup

Beyond feed data, agents also parse Schema.org Product markup directly from your product pages. This structured data helps agents understand:

• Product type and category
• Aggregate ratings and review counts
• Price validity periods
• Availability at specific locations

Merchants with complete Schema.org implementation get indexed more accurately. Those without it force agents to guess—and agents don't guess in your favor.

Product feed checklist for AI agent visibility

Use this checklist to audit your product feed readiness:

Tier 1 mandatory fields (products without these won't appear)

[ ] Every product has a unique, descriptive title

[ ] Descriptions include key attributes (material, dimensions, compatibility)

[ ] Prices are accurate and sync with your live site

[ ] Availability reflects real-time inventory

[ ] Images are high-resolution (minimum 800x800) and properly formatted

[ ] URLs resolve to live product pages (no redirects, no 404s)

Tier 2 discovery fields (products without these get deprioritized)

[ ] GTINs assigned to all standard retail products

[ ] Brand names standardized (no variations like "Nike" vs "NIKE" vs "nike")

[ ] Shipping information includes cost, speed, and geographic coverage

[ ] Return policies are machine-readable (days, conditions)

[ ] Product variants (size, color) have separate entries with proper parent relationships

Tier 3 competitive advantage fields

[ ] Customer review data included (rating, count)

[ ] Promotional pricing with validity dates

[ ] Cross-sell and compatibility data

[ ] Sustainability and certification attributes

[ ] Loyalty program benefits visible to agents

Industry analysis suggests that merchants with 95%+ fill rates on Tier 1 and Tier 2 fields see significantly better agent discovery. Below 80%, products are routinely skipped.

Real-time inventory sync for agentic commerce

Static feeds updated once daily won't cut it. Agents make purchasing decisions in real-time, and nothing kills a transaction faster than discovering an item is out of stock at checkout.

The technical requirements:

Update Frequency: Your feed should reflect inventory changes within minutes, not hours. For high-velocity SKUs, near-real-time sync is essential.

Variant Accuracy: When you sell "blue running shoes in sizes 7-13," each variant needs accurate availability. Showing "in stock" when size 10 is sold out creates failed transactions and damages your reliability score with agents.

Price Consistency: If your website shows $49.99 but your feed shows $54.99, agents may reject the transaction or deprioritize future recommendations. Price sync must be bulletproof.

Geographic Availability: Products available only in certain regions need location-specific availability data. An agent recommending a product that can't ship to the customer's address is a failed experience.

The infrastructure challenge here is significant. Most merchants run inventory systems that weren't designed for real-time external sync. Bridging this gap requires either upgrading your inventory management or implementing a data layer that normalizes and syncs inventory data to external feeds.

How to test your product visibility to AI agents

Before assuming your products are discoverable, verify it:

Manual testing

1. Ask ChatGPT to find products in your category with specific attributes
2. Note whether your products appear, and if so, how they're described
3. Check if pricing, availability, and shipping information are accurate
4. Test edge cases: specific sizes, colors, or configurations

Feed validation

Use Google Merchant Center's feed diagnostics to identify:

• Missing required attributes
• Data quality warnings
• Disapproved items
• Improvement opportunities

Many of the same data quality issues that hurt Google Shopping performance will hurt agent visibility.

Schema markup testing

Use Google's Rich Results Test to verify your Schema.org Product markup is valid and complete.

Competitive benchmarking

Search for your competitors' products through AI agents. Compare:

• Do they appear more frequently?
• Is their data more complete?
• Are their prices and availability more accurate?

The answers reveal where your data infrastructure falls short.

Which e-commerce platforms support AI shopping agents

Major commerce platforms are racing to enable agent visibility. Understanding what's already available helps prioritize your infrastructure investments.

Shopify launched Agentic Storefronts in 2025, allowing merchants to expose their catalogs to AI agents through standardized APIs. For Shopify merchants, the basic connectivity is handled—but data quality remains the merchant's responsibility. A Shopify store with incomplete product data will be just as invisible to agents as any other poorly-maintained catalog.

Etsy integrated with ChatGPT's shopping features, enabling direct discovery of handmade and vintage items. The integration exposed a challenge unique to marketplace sellers: when you don't control the platform's feed infrastructure, your visibility depends entirely on how well you've populated your individual listing data.

PayPal announced automatic ACP support for merchants in its network starting in 2026, meaning payment connectivity won't be the blocker. But PayPal can't fix your product data—that infrastructure work falls to you.

The pattern across all platforms: connectivity is becoming commoditized, but data quality remains the differentiator. Platforms can route agent requests to your store, but they can't make your products discoverable if the underlying data is incomplete.

How product data quality improves all commerce channels

Here's what makes this investment worthwhile beyond agent commerce: the same data infrastructure that makes products visible to AI agents improves performance across every channel.

Clean, structured product data with accurate GTINs, complete attributes, and real-time inventory sync benefits:

• Google Shopping campaigns (better quality scores, lower CPCs)
• Marketplace listings (Amazon, Walmart, Target)
• Social commerce (Instagram, TikTok Shop)
• Affiliate networks (more accurate product matching)
• Your own site search (better relevance, fewer dead ends)

The merchants who treat data infrastructure as a strategic investment—rather than a compliance checkbox—see compounding returns across channels.

This is the cross-surface reality of modern commerce. The same product data flows to web, app, marketplaces, social platforms, and now AI agents. Inconsistencies in one channel create friction in all of them. Excellence in data infrastructure creates advantage everywhere.

Building first-mile data infrastructure for AI commerce

Product visibility to AI agents isn't a feature you toggle on. It's the result of data infrastructure that captures, normalizes, and syncs product information at the source—what we call the first mile.

First-mile data capture ensures that product data is correct from the moment it enters your ecosystem. GTINs assigned at product creation. Attributes validated against schema requirements. Inventory synced the moment it changes. Prices updated in real-time across every downstream channel.

When data quality is enforced at the first mile, you don't need to fix problems downstream. You don't have feed-specific cleanup processes. You don't have channels showing conflicting information. Every surface—agents included—receives the same accurate, complete data.

The merchants who win in agentic commerce won't be those who scramble to optimize for ChatGPT after the fact. They'll be those who built data infrastructure that works across all surfaces from the start.

The agents are already shopping. The question is whether they can see your products.

‍

MetaRouter provides the first-mile data infrastructure that enterprise retailers use to capture, normalize, and route product and customer data across all commerce surfaces. Learn how server-side data collection improves data quality at the source.‍

‍