ChatGPT processed 800 million weekly conversations in late 2024, but only 2.1% involved purchasable products. When OpenAI launched Instant Checkout in October 2024, that changed, giving over one million Shopify merchants automatic integration overnight.

For VPs of marketing and commerce who've already committed to agentic commerce, the challenge isn't whether to participate but ensuring your infrastructure can actually support it. The merchants struggling right now share a common trait: they treated this as a marketing channel when it's actually an infrastructure project. Your data quality, fraud detection, and attribution capabilities determine whether you succeed or spend months firefighting problems that could have been prevented.

This article breaks down what you need in place before launch, drawn from merchant forums, early implementations, and the technical requirements that determine whether your products even appear in search results.

The infrastructure you need for ChatGPT shopping

When Shopify flipped the switch on ChatGPT integration, merchants discovered catalog problems they'd never noticed before. Missing Global Trade Item Numbers (GTINs) meant AI agents couldn't match products, thin descriptions caused agents to skip items entirely, and variant confusion showed customers partial size ranges.

ChatGPT relies entirely on your product feed to decide what to show, and missing required fields disqualify entire product lines immediately. You need unique product IDs, GTINs or manufacturer part numbers, complete titles and descriptions, real-time availability, images, variant information, delivery options, and return parameters. This reveals problems that existed across your operation but never created friction until now. Your product databases have inconsistencies between systems, inventory updates run in batches instead of real-time, pricing gets cached, and description quality varies by whoever entered the data. Traditional e-commerce tolerated these issues, but agentic commerce doesn't.

Someone on your team needs to own continuous feed management, with feeds refreshing every 15 minutes to reflect current inventory and pricing. If you run multiple locations, franchise networks, or distributed fulfillment, coordinating accurate data across hundreds of points becomes exponentially harder, and one outdated price at a single location breaks the experience.

Research tracking 973 e-commerce sites found catalogs with incomplete information, wrong prices, and broken links. Prices shown in ChatGPT don't sync with live inventory, and products appear available when they're actually out of stock. These aren't edge cases but systemic quality issues you haven't prioritized fixing because traditional e-commerce worked around them.

Fraud detection in ChatGPT shopping

Your fraud systems catch suspicious human behavior like rapid purchases across unrelated categories, unusual locations, and velocity spikes. AI agents behave completely differently, placing many small orders quickly, buying across categories simultaneously, and triggering patterns your systems flag as fraud when they're actually legitimate customer transactions.

You can't block all agent traffic without losing the channel, but accepting all agent behavior creates exposure. How do you distinguish legitimate AI agents from actual fraud?

The industry hasn't solved this yet, though payment networks including Mastercard and Visa are developing Know Your Agent frameworks with cryptographic verification. Google's Agent Payments Protocol uses signed mandates that prove what users authorized, but these standards are nascent and you can't rely on them today.

You need agent-specific fraud rules separate from human transaction rules, which typically means working with fraud vendors who understand agentic commerce or building custom risk models. Your payments team, fraud operations, and payment processor all need involvement, and if you don't have dedicated fraud vendors, this becomes a procurement requirement. PCI scope narrows with payment tokenization since raw card numbers never reach your systems, but you still need to verify agent identity and manage authorization proof in ways your compliance team hasn't addressed.

Attribution challenges in ChatGPT shopping

Traditional e-commerce shows you who clicked ads, browsed your site, abandoned carts, and eventually converted. You build customer profiles, predict churn, and segment for loyalty programs. In agentic commerce, that entire journey disappears because the transaction happens inside ChatGPT.

You get order data but not the context around it. You don't see that your customer considered three competitors, visited your website first, or is making their fifth purchase with you. You just see the final order, stripped of the behavioral signals you've relied on for years. Your attribution is already messy, with online discussions showing Klaviyo and Shopify reporting different revenue values for the same campaign. Agentic commerce makes this worse by removing pre-purchase visibility entirely, breaking your last-click and multi-touch models when there are no clicks to track.

For retail media networks, this threatens your entire business model since media revenue depends on proving ads influenced purchases. When discovery and consideration happen in a black box you can't access, how do you prove performance? Walmart generates $4.4 billion annually from retail media, and losing visibility into what drives purchases puts that revenue at risk. ChatGPT doesn't tell you whether this customer is new or returning, what alternatives they considered, or why they chose your product. You learn about purchases after they happen with limited information about triggers.

Some merchants are testing workarounds like subscription models where relationships transcend individual transactions, loyalty programs that incentivize customers to declare preferences directly, and value-added services requiring ongoing interaction. These require new infrastructure and won't work for all product categories, but they represent attempts to maintain customer relationships when the platform controls discovery.

Your analytics and business intelligence team needs involvement alongside marketing operations to build models that account for invisible touchpoints, focus on incrementality rather than channel-specific metrics, and establish relationships with AI platforms to access whatever data they'll share.

Who owns ChatGPT shopping

ChatGPT Shopping touches marketing, payments, and engineering simultaneously, which makes ownership unclear. Different teams have essential responsibilities, but without coordination, your initiative stalls between departments.

Product and merchandising

Your product team owns feed accuracy, auditing current data, identifying gaps, and establishing governance for continuous updates. They also optimize information specifically for AI interpretation, which means rethinking how you structure titles, descriptions, and metadata.

Engineering

Your engineers own checkout endpoints, and if you're on Shopify, this is largely automatic. On custom platforms, you're looking at 4-10 weeks implementing API endpoints for checkout creation, order status webhooks, and payment token integration. You need coordination with payment processors and thorough edge case testing.

Payments and finance

This team handles fraud detection, PCI compliance verification, and payment processing rules. They work with payment service providers on tokenization, implement agent-specific fraud rules, and prepare for validating mandates as standards mature. They also manage financial reconciliation of transactions occurring outside traditional checkout flows.

Marketing operations

Marketing operations rethinks attribution from scratch since your traditional tools don't measure ChatGPT's contribution accurately. You need multi-touch models, relationships with AI platforms to access available data, and a shift from channel-specific metrics to incrementality. Your team needs to accept that some customer journey data simply won't exist.

Customer service

Your service team prepares for new inquiry types, developing playbooks for when AI agents make purchase errors or customers ask why products didn't appear in search results. You potentially need tools that access chatbot conversation logs when troubleshooting orders, which raises privacy and access control questions.

Compliance and legal

Your legal team evaluates data privacy implications since customers delegating purchases to ChatGPT share addresses, payment details, and preferences with OpenAI. Your privacy policies need updates, and GDPR and CCPA compliance shifts when customer data flows through third-party AI platforms.

Executive leadership

You need executive sponsorship because teams have conflicting priorities, with product potentially delaying feed work while engineering focuses on other roadmap items. Someone needs authority to align priorities around agentic commerce infrastructure, and without clear ownership at the executive level, your initiative stalls with each team waiting for another to move first.

How to launch your ChatGPT shopping initiative

Treat this as infrastructure work that opens a new channel once complete, not as a marketing launch. Breaking work into phases helps you manage complexity and validate each component before moving forward.

Phase 1: Data Foundation (3-6 weeks)

• Audit product data against the OpenAI Product Feed Spec
• Identify gaps in variants, descriptions, and pricing consistency
• Establish governance for continuous feed updates
• Set up real-time data sync if inventory changes frequently

Phase 2: Fraud and Compliance (2-4 weeks)

• Verify PCI compliance scope with your payment service provider
• Implement agent-aware fraud detection rules distinct from human transaction rules
• Establish or verify Stripe account integration
• Document how you'll handle agent authentication signals as standards mature

Phase 3: Checkout Integration (timeline varies by platform)

• Shopify merchants: Enable through Instant Checkout app with minimal lift
• Custom platform merchants: Implement checkout endpoints, webhooks, and order status tracking (4-10 weeks with engineering resources)

Phase 4: Attribution and Analytics Setup (2-4 weeks)

• Configure multi-touch attribution models
• Establish baseline metrics for ChatGPT traffic
• Set up dashboards for ChatGPT-attributed orders
• Plan for data limitations upfront and acknowledge what you won't see

Phase 5: Organizational Alignment (ongoing)

• Establish cross-functional teams with clear accountability
• Define escalation paths for support issues
• Plan customer communication strategy for the new channel
• Set realistic expectations about visibility and control you'll actually have

Infrastructure to enable the future of agentic commerce

You're facing pressure to enable ChatGPT Shopping while your infrastructure has gaps, and the technology is advancing faster than most organizations can adapt. The merchants succeeding right now aren't the ones who moved fastest but the ones who addressed three critical areas first: data quality, attribution modeling, and compliance frameworks.

Your infrastructure needs to deliver clean, structured product data that updates in real-time, models that account for invisible pre-purchase behavior and stitch customer identity across sessions, and frameworks that handle data sharing with AI platforms while meeting regulatory requirements. These capabilities aren't optional extras but fundamental prerequisites for participating in this channel.

When evaluating solutions, look for platforms that capture customer intent before transactions occur, embed compliance controls at the event level, and normalize data across both agent and traditional channels. These capabilities close the measurement, attribution, and compliance gaps that conversational commerce platforms leave unsolved. Companies building first-mile data infrastructure, like MetaRouter, focus specifically on these challenges.

Start with an infrastructure audit, mapping your current data quality, fraud detection capabilities, and attribution models against the requirements outlined in this article. Identify gaps, prioritize the ones that would cause immediate problems at launch, and build cross-functional alignment before you flip the switch. The capability gaps exist today, and the question is whether you address them proactively or discover them when customers start complaining.

‍