Agentic Commerce Protocol (ACP) vs Google AI Mode Shopping — What Every E-Commerce Manager, CMO and Developer Needs to Know

1. Introduction

E-commerce is evolving. What once was a matter of building product pages, optimizing for search keywords and driving paid traffic is now shifting toward agentic commerce: the environment where AI‐agents act on behalf of buyers, help discover products, and complete transactions inside conversational or assisted interfaces.

For e-commerce managers, CMOs and developers, understanding these emerging platforms is becoming imperative. These aren't just incremental improvements to search or checkout — they represent new channels, new user behaviours and new integration requirements. The choices you make now about your data feeds, your checkout architecture, and your brand's presence in AI-driven discovery will impact your position in the next wave of online commerce.

In this post I compare two leading offerings: the Agentic Commerce Protocol (ACP) launched by OpenAI + Stripe, enabling checkout inside ChatGPT, and Google's AI Mode Shopping, which embeds advanced AI and agentic features (including checkout, price-tracking and local inventory calling) inside Google Search & the Gemini app. We'll examine both from three key perspectives: e-commerce managers/CMOs (strategy, brand, channel), and developers (integration, architecture, pipeline). At the end we'll draw recommendations.

2. Overview of ChatGPT's Agentic Commerce Protocol (ACP)

2.1 Purpose and Architecture

The Agentic Commerce Protocol (ACP) is an open-standard developed by OpenAI together with Stripe that enables AI agents, buyers and merchants to complete purchases through a conversational interface.

The key design goals are:

• Powerful: To connect millions of AI users (via ChatGPT) with merchants.
• Easy to adopt: If a merchant already uses Stripe (or a compatible payment processor) the integration can be relatively straightforward.
• Flexible: Works across platforms and payment processors; merchants remain the merchant of record.
• Secure: Payment tokens are delegated and encrypted; the user explicitly confirms the purchase; limited data sharing.

In effect, this means that when a user is chatting with ChatGPT and expresses a purchase intent, the system can surface a Buy button, collect minimal info, send the order to the merchant's system, process payment and fulfilment, without redirecting the user to a completely separate checkout flow. The merchant retains control of fulfilment, returns, support and customer relationship.

2.2 Components of ACP

Let's break down the main technical components of ACP for clarity:

• Product Feed Spec
  Merchants must provide a structured product feed (via CSV, TSV, XML or JSON) that includes identifiers (SKU/GTIN), prices, inventory, images, reviews, performance signals and other attributes. The feed needs to be refreshed frequently (sometimes as often as every 15 minutes) to maintain accuracy.
  Importantly, the feed is multi-dimensional: beyond basic product data it includes things like ratings, Q&A, geotagging, performance metrics of previous transactions (for ranking). A standard feed may not suffice.
• Agentic Checkout Spec
  This defines REST endpoints (e.g., /checkout_sessions) for merchants to create, update and complete checkout sessions. In addition, webhooks are used for order events (order created, fulfilled, cancelled). The merchant remains in charge of payments and fulfilment.
• Delegated Payment Spec
  The payment side may be handled through a delegated token: for example, OpenAI passes a payment token to the merchant's payment service provider (PSP), or the merchant uses Stripe's Shared Payment Token API. The merchant still gets the payment on their rails.

2.3 Instant Checkout Experience

Here is how the user experience works in practice:

• A user interacting with ChatGPT says something like: "I'd like to buy a new ergonomic office chair under €400 that ships to Spain."
• ChatGPT uses its product ranking and discovery logic (via the ACP feed) to find relevant items. Some items will show a Buy button.
• The user taps "Buy", confirms shipping details and payment method (if saved). ChatGPT then calls the merchant's checkout endpoint via the agentic checkout spec.
• The merchant validates the order (inventory, pricing, shipping region), receives the payment by normal means (or via PSP), and initiates fulfilment on their own systems.
• The merchant remains in charge of the customer relationship: returns, support, branding, etc.
• The feed ranking is organic and unsponsored, meaning merchants are not paying to appear in these listings (in early stage) — instead there is a small transaction fee for completed orders.

Currently, the offering is limited (as of this writing) to single-item purchases in the U.S. for ChatGPT Plus/Pro users (and some free users). Multi-item carts and wider region support are on the roadmap.

2.4 Merchant On-boarding & Options

Merchants have three basic participation options:

1. Do nothing — you can block ChatGPT crawlers or decline participation; you're not forced into it.
2. Feed only — you provide a product feed so your catalog is discoverable in ChatGPT, but you don't enable checkout. This gives exposure and traffic without deeper integration.
3. Feed + Agentic Checkout — you implement both the feed and the checkout/webhook endpoints, enabling full checkout through ChatGPT.

Integration steps for option 3 typically include:

• Apply to participate in Instant Checkout for ChatGPT.
• Upload/serve your product feed per ACP spec.
• Implement checkout endpoints (create, update, complete) and webhooks per spec.
• Integrate with delegated payment tokens (if using Stripe's Shared Payment Token API, or your PSP) and ensure security, authorisation, idempotency.
• Undergo certification or conformance review with OpenAI (and possibly Stripe) to go live.

From a data engineering perspective the feed is more complex than a typical Google Shopping feed: you'll need to include rich metadata (reviews, Q&A, geotags), ensure near-real-time inventory/price updates, and handle scale.

2.5 Implications for E-commerce Managers & CMOs

• Conversion opportunity: This is a new high-intent channel. According to some early estimates ChatGPT's user base (~700 million weekly users) represents a large pool. Early adopters may benefit from relatively lower competition compared to traditional search or paid channels.
• Data quality & freshness: Because ranking and visibility in ChatGPT depend on accurate, timely product data and clean metadata, your internal product information management (PIM) systems must be robust. Out-of-stock or incorrect pricing quickly harms experience.
• Brand visibility: One of the interesting things about ACP is that the ranking is unsponsored (at least initially) — so relevance, quality of presentation and customer reviews matter more than budget. This levels the playing field for smaller brands, potentially.
• Customer relationship: Even though the purchase happens via ChatGPT, the merchant remains the merchant of record — you fulfil, handle returns, service. That means your operations, support, and brand promise still matter.
• Channel diversification: Recognising this as a separate channel, you'll need to allocate resources (data feeds, API integration, testing) and measure performance distinctly.
• Pricing & margin: Because purchases are enabled inside a conversational interface with minimal friction, you might capture more impulse buys — yet you also face elevated expectations in terms of delivery, returns, user experience.
• Risk & compliance: With delegated payments and AI agents initiating flows, you must ensure your operations comply with privacy, payment (PCI-DSS), consumer-protection laws in your region.

2.6 Implications for Developers

From a development and architecture perspective:

• Integration complexity: You'll need to build or extend checkout endpoints (APIs) that support the ACP spec: create/update checkout sessions, handle webhooks, ensure idempotency, manage order state.
• Payment architecture: If you already use Stripe, the Shared Payment Token API may simplify things. If you use another PSP, ensure you handle delegated tokens securely, maintain audit trails.
• Feed engineering: You'll require pipelines to generate the multi-dimensional product feed: identifiers, prices, inventory, reviews, Q&A, geo-data, performance signals. Pipeline should support frequent refresh (potentially every 15 minutes).
• Testing & certification: You must operate in sandbox/test mode, handle error cases gracefully, pass conformance/certification with OpenAI/Stripe.
• Scalability & future-proofing: At launch, ACP supports single-item purchases. But you should architect with future-state in mind: multi-item carts, subscriptions, internationalization.
• Monitoring & reporting: Build dashboards for queries from ChatGPT, conversion rates, abandonment in agentic flows, error handling, inventory mismatches.
• Security & compliance: Ensure user confirmation flows are explicit, token usage is secure, implement proper authentication, webhook security, rate-limiting, failover.

3. Overview of Google's AI Mode Shopping (2025)

3.1 Purpose and Architecture

Google AI Mode (often simply "AI Mode") is the next-generation search and shopping experience introduced by Google, combining conversational/agentic capabilities (via its Gemini models) with its deeply established commerce infrastructure (its Shopping Graph). According to Google, the Shopping Graph now includes more than 50 billion product listings, refreshed hourly for many items.

The idea is that shopping becomes less about navigating category pages, filters and site after site — and more about using natural-language (or multimodal) queries, receiving curated product discovery panels and completing purchases through agentic flows (price-tracking, local store inventory checks, virtual try-on).

The underlying architecture:

• Shopping Graph: a massive database of products from global retailers and local stores, with rich metadata (reviews, inventory, price, images).
• Generative & multimodal AI (Gemini) that can interpret conversational queries, images or uploads, and do "query fan-out" (i.e., run multiple internal searches in parallel to find relevant criteria) to refine recommendations.
• Agentic features: price-tracking with auto-purchase ("buy for me"), phone calls to local stores using Duplex tech for inventory checks, virtual try-on for apparel.

3.2 Key Features

3.2.1 Conversational Discovery & Product Curation

In AI Mode, users can talk or type in natural language ("I want a travel bag under €80, light, with wheels, that fits a cabin flight") and the system uses internal "query fan-out" to figure out relevant criteria (e.g., wheels, cabin size, weight, durability, waterproofing). According to Google's announcement:

"Say you tell AI Mode you're looking for a cute travel bag. It understands that you're looking for visual inspiration … it will run several simultaneous searches to figure out what makes a bag good for rainy weather and long journeys…"

Results appear in a dynamic panel: images, price ranges, colour options, user reviews, inventory availability. The visuals matter: high-quality photography, clean product presentation and rich attributes matter more than ever.

3.2.2 Agentic Checkout & Price Tracking

Google's AI Mode design includes "agentic checkout" capabilities: users can set a target price, size or colour and ask the system to notify when the criteria are met — then optionally authorise Google to complete the purchase via Google Pay and the merchant website. Example quote:

"Our new agentic checkout will help you easily buy at a price that fits your budget … just tap 'track price' … when the item drops to your budget … and … tap 'buy for me'. … We'll add the item to the merchant's cart and securely complete the checkout on your behalf with Google Pay."

3.2.3 Local Inventory and Duplex Calling

For local purchases, AI Mode offers a "Let Google Call" feature: Google's AI (via Duplex) calls nearby stores, checks their stock, prices and promotions, then emails/texts a summary to the user. This supports "near me" inventory checks at scale.

This matters for physical-inventory retailers, local stores and omnichannel strategy. If your store data is inaccurate or phone numbers/hours are wrong, you risk being bypassed or mis-represented.

3.2.4 Virtual Try-On

Google also launched a virtual try-on tool (initially as an experiment in Search Labs) that lets users upload a full-length photo of themselves, and the system will show how shirts, pants, skirts or dresses look on them — using a custom image-generation model that understands human poses and fabric behaviour.

For apparel brands this is a major opportunity (and challenge): ensure images, 3D or at least multi-angle views are high-quality, and the model has enough product metadata to offer realistic output.

3.3 Implications for E-commerce Managers & CMOs

• Feed readiness: Success in AI Mode depends heavily on your product feed quality and freshness — because Google's Shopping Graph refreshes billions of listings hourly. If your price/inventory is stale you risk mis-representation.
• Visual merchandising: Because discovery is more visual (image panels, dynamic filters, try-on), you need high-quality photography, realistic product views, strong review signals, accurate colour/size variants, and rich attribute data.
• Price-tracking strategy: With Google's price tracking and agentic checkout features, users may wait for deals or set alerts. As a brand you may face margin pressure: if you wait too long or discount too much, you risk commoditisation.
• Local inventory & omnichannel: The "call store" feature means your local inventory data must be accurate, staff must be ready for AI-powered calls, and phone number/hours/promotions must be correct.
• Brand visibility & trust: Google's AI Mode will likely integrate ads or sponsored placements in future. Brands need to monitor how conversational responses, shopping panels and ad placements interplay.

3.4 Implications for Developers

• Merchant Center integration: Ensure your product feed meets Google's detailed specifications (GTINs, price, availability, variants) and update it frequently.
• Agentic checkout implementation: To support features like "buy for me", your cart/checkout must integrate with Google Pay and your backend must handle redirected traffic or API-driven initiation from Google.
• Duplex call support for stores: If you have local stores participating, ensure your telephony, inventory systems and staff are ready to respond to AI-initiated phone calls.
• Compliance & trust: AI-driven purchases raise additional governance questions: user auth, payment security, elevated expectations for fulfilment, handling of returns, data flow compliance.
• Future-proofing: AI Mode is still evolving — multi-item carts, subscription flows, voice/multimodal queries, geographic expansion. Developers should build modular, scalable systems.

4. Comparative Analysis: ChatGPT ACP vs Google AI Mode Shopping

4.1 Scope and Audience

4.2 Strategic Considerations for E-commerce Managers & CMOs

• Channel mix: Use both platforms to diversify reach. ChatGPT offers an early-adopter competitive advantage and less saturation, but a smaller user base compared to Google.
• Data strategy: With ACP you'll need additional metadata (reviews, Q&A, geotags, performance signals) and frequent feed refreshes; for Google you need to ensure your Merchant Center feed meets high standards.
• Pricing & promotions: On Google AI Mode, price-tracking may lead to more consumers waiting for deals and triggering "buy for me" flows when the price drops — your margin strategy must account for this.
• Customer experience: Since both channels blur discovery and purchase, your post-purchase experience (delivery, returns, support) must be seamless.
• Brand visibility & trust: On ChatGPT, because the checkout is embedded in chat, your brand presentation matters a lot. On Google AI Mode you must ensure your brand's narrative, imagery, review signals are strong.
• Risk & compliance: With AI-initiated or AI-assisted purchases the governance risks increase — you must ensure compliance with payment standards, consumer rights, data protection.
• Competitive timing: Because both platforms are relatively new early adopters may gain advantage. If you wait until they are saturated you may lose the lower-hanging fruit.

4.3 Considerations for Developers

• Implementation effort: ACP integration is heavier in terms of new APIs, payment token flows, webhooks, certification. Google integration may leverage existing Merchant Center and Google Pay flows.
• Scalability: For ACP, feed generation must support frequent refresh and rich metadata; for Google you may already have feed infrastructure but need to ramp up imaging, variant data, local store systems.
• Testing & certifications: Both platforms demand rigorous testing. ACP requires certification with OpenAI/Stripe. Google will require you to meet its validation rules.
• Future-proofing: Both platforms are evolving quickly. Building modular, extensible architecture is wise.
• Data pipelines & monitoring: You should build dashboards for each platform: number of impressions, conversion rate from the AI channel, average order value, error rates.
• Security & compliance engineering: You'll need robust handling of delegated payment tokens, encryption, user consent flows, authorised merchant endpoints.

5. Conclusion and Recommendations

5.1 Summary

We've looked at two major moves in agentic commerce: the Agentic Commerce Protocol (ACP) via ChatGPT/OpenAI/Stripe, and Google's AI Mode Shopping experience. Both represent significant shifts in how consumers discover, evaluate and purchase products — from keyword-based search and separate checkout flows, to conversational interfaces and embedded checkout.

5.2 Recommendations

Here are actionable steps you should consider:

• Adopt a dual-platform strategy: Don't pick only one. Enable your catalogue for ChatGPT/ACP and prepare for Google AI Mode. While resources may limit full rollout initially, at least plan how you will engage both channels.
• Invest in product data quality: This is foundational. For both ACP and Google AI Mode you'll need accurate prices, inventory, variants, images, reviews, Q&A, local availability (if applicable).
• Align customer experience with AI expectations: These channels lower friction and raise expectations. Make sure your fulfilment, returns, customer support, local store readiness are up to speed.
• Monitor regulatory and compliance issues: With delegated payments, AI-initiated flows, global expansion, you must keep an eye on payment regulations, data protection, consumer rights.
• Prepare your teams and budgets: You'll need collaboration across marketing, operations, engineering. Allocate time and budget now.
• Measure channel performance separately: Build dashboards to track AI-agentic channel performance: traffic from ChatGPT, conversion rate, AOV, cost per transaction.
• Optimize for early movers: Because both platforms are still early in many markets, getting onboard early may yield advantage.
• Stay agile: Both ACP and Google AI Mode are evolving quickly. Build your architecture and processes to be modular and adaptive rather than rigid.
• Review brand & pricing strategy: With price-tracking and agentic flows, you may face new buyer behaviour. Consider how your pricing cadence, promotions and inventory strategy align.
• Educate stakeholders: Ensure executives, marketing, product teams understand what "agentic commerce" means. You may need new KPIs, new workflows and new skillsets.

5.3 Final Thought

We are entering a phase where the interface for commerce is shifting: fewer websites, less manual browsing, more conversational, intelligent agent-mediated flows. The way a user asks "help me buy X" may even output the purchase as the result. For e-commerce managers, CMOs and developers, this means revisiting not just how you sell but where and how seamlessly you enable a purchase.

If you treat these platforms as incremental add-ons to your existing channels, you might miss the strategic shift. If you build your feed/data, checkout/connectivity, and brand presence to support conversational commerce, you stand a chance to lead.