And if you’re building a business, you have about 18 months to figure out which ones matter.
In September 2025, OpenAI quietly launched a feature that let ChatGPT users buy products without leaving the chat.
Most people missed it.
By December, Google, Amazon, Microsoft, and Perplexity had all launched competing versions. Walmart, Target, and Etsy had signed deals. Stripe and PayPal were fighting for the payment rails.
This isn’t a feature war. It’s a complete restructuring of how commerce works on the internet.
And almost nobody is talking about what it actually means.
The Five Platforms That Now Want to Be Your Shopping Cart
Here’s what happened in the last 90 days:
| Platform | Launch | What It Does |
| ChatGPT | Sept 2025 | Instant Checkout with Stripe |
| Perplexity | Nov 2025 | Instant Buy with PayPal |
| Nov 2025 | “Buy for Me” agentic checkout | |
| Amazon Rufus | Nov 2025 | Auto-buy at target prices |
| Microsoft Copilot | May 2025 | In-app checkout via Shopify |
Five platforms. Five different approaches. One shared assumption: the future of shopping is conversational.
Let me break down what each one actually does.
ChatGPT: The Protocol Play
OpenAI isn’t just building a shopping feature. They’re building the infrastructure layer for AI commerce.
Their approach has three components:
1. Agentic Commerce Protocol (ACP) An open-source standard (Apache 2.0) that defines how AI agents communicate with merchants. Think of it as HTTP for AI shopping. Anyone can build on it.
2. Shared Payment Tokens (SPT) Built with Stripe. These are scoped, limited-use payment tokens that let AI agents complete purchases without exposing your actual card details. The merchant never sees your card number. Neither does OpenAI.
3. Instant Checkout The user-facing feature. You’re chatting about a product, you tap “Buy,” and it’s done. No redirects. No new tabs. No entering shipping info.
Current partners: Etsy, Shopify merchants, Walmart, Target, PayPal (coming 2026)
The strategic insight: OpenAI isn’t trying to own the payment layer—they’re trying to own the protocol. If ACP becomes the standard, every AI agent in the world routes commerce through infrastructure OpenAI helped define.
Perplexity: The PayPal Bet
Perplexity took a different approach. Instead of building new payment infrastructure, they partnered with PayPal’s existing merchant network.
What this means in practice:
- 5,000+ PayPal merchants instantly available
- Checkout happens inside Perplexity
- Merchants remain merchant of record
- No new integration required for existing PayPal sellers
Live merchants: Abercrombie & Fitch, Ashley Furniture, Fabletics, Adorama, NewEgg
The killer feature: Memory. Perplexity remembers your past searches and preferences. Ask for “a jacket for my commute” and it already knows you live in San Francisco and take the ferry.
The strategic insight: Perplexity is betting that distribution matters more than infrastructure. By plugging into PayPal’s existing network, they got scale instantly. The question is whether that’s enough to compete with OpenAI’s 700 million weekly users.
Google: The Automation Play
Google’s approach is the most aggressive. They’re not just letting you buy things—they’re letting AI buy things for you.
Three features launched in November:
1. Agentic Checkout (“Buy for Me”) Set a target price. Google monitors the product. When it hits your number, Google’s AI automatically completes the purchase using Google Pay.
You wake up to a notification: “We bought those headphones you wanted. They were $149. Order arrives Thursday.”
2. AI Phone Calls (Duplex) Google will literally call physical stores on your behalf to check if something is in stock. The AI identifies itself as calling on your behalf, asks your questions, and texts you a summary.
3. Shopping Graph Integration 50 billion product listings. 2 billion updated every hour. All searchable through natural language in AI Mode or the Gemini app.
Current partners: Wayfair, Chewy, Quince, select Shopify merchants
The strategic insight: Google is betting on automation, not just assistance. They want to remove you from the purchase decision entirely. Set your preferences, and the AI handles the rest.
This is the most ambitious vision—and the most likely to face consumer resistance. Do people actually want AI spending their money autonomously?
Amazon Rufus: The Walled Garden
Amazon’s approach is characteristically Amazon: build everything in-house and block everyone else.
The stats are staggering:
- 250 million users in 2025
- Customers who use Rufus are 60% more likely to purchase
- Projected to drive $10 billion in incremental annual sales
Key features:
Account Memory Rufus remembers everything—your past purchases, your household details, your preferences. Tell it you have four kids who like board games, and that context persists.
Auto-Buy Like Google, Rufus can purchase automatically when prices hit your target. But it goes further: you can say “reorder my usual coffee” or “restock my pet supplies” and Rufus handles it.
Visual Search Snap a photo of your handwritten shopping list. Rufus adds everything to your cart.
“Buy for Me” (External) Here’s the wild part: if Amazon doesn’t carry something, Rufus will buy it from other retailers on your behalf. Amazon becomes your shopping agent even for products it doesn’t sell.
The strategic insight: Amazon blocked external AI agents (including OpenAI, Anthropic, and Google) from scraping its site—while building an internal AI that does everything those external agents would do.
They’re not participating in the open agentic commerce ecosystem. They’re building a competing closed one.
Microsoft Copilot: The Enterprise Angle
Microsoft launched Copilot Shopping more quietly than the others, but their approach is distinctive.
The Merchant Program Launched April 2025. Retailers can submit product catalogs directly to Copilot, making their inventory searchable through the AI.
The Features:
- Real-time price comparisons across retailers
- In-app checkout (initially via Shopify)
- Price tracking with alerts
- Order history centralization
The claim: Copilot + Bing usage shortened the consumer purchase journey by 30%.
The strategic insight: Microsoft is positioning Copilot as the AI layer across Windows, Edge, and enterprise tools. Shopping is part of a broader play to make Copilot the default assistant for everything—including commerce.
The Notable Exception: Meta Retreated
Here’s what’s fascinating: Meta went the opposite direction.
In August 2025, Meta phased out native checkout on Facebook and Instagram Shops. All purchases now redirect to merchant websites.
They added AI features to Facebook Marketplace—suggested questions, collaborative buying, vehicle insights—but no in-app checkout.
Why?
My read: Meta is betting that their value is in discovery and targeting, not transaction completion. They’d rather use AI chat data to serve better ads than build checkout infrastructure.
This might be smart. Or it might be the mistake that costs them the commerce layer entirely.
What This Means for Businesses
If you’re selling products online, here’s the new reality:
1. Product Data Is Now Your SEO
Every one of these platforms ranks products based on structured data quality. Rich descriptions, accurate inventory, complete specifications, review data.
The businesses that treat product feeds as a strategic asset will win. The ones that upload CSVs and forget about them will become invisible.
2. You Need to Be on Multiple Platforms
No single AI will win all commerce. Different users prefer different platforms. The smart play is to optimize for ChatGPT, ensure your products are discoverable in Google’s Shopping Graph, and maintain your presence on Amazon.
3. Checkout Friction Is Being Eliminated
The platforms that remove the most friction will capture the most transactions. If your checkout process requires account creation, multiple pages, or manual entry—you’re now competing against one-tap purchases.
4. AI-Readable > Human-Readable
Your product descriptions need to work for algorithms, not just humans. This means:
- Including use cases and target customers explicitly
- Answering common questions in the description
- Using consistent, structured formatting
- Keeping inventory and pricing updated in real-time
The Bigger Picture
What’s actually happening here is a platform shift.
For 25 years, e-commerce worked the same way: search engine → website → cart → checkout.
That flow assumed humans would do the browsing, comparing, and deciding.
AI changes the assumption.
If an AI can research products, compare options, check reviews, and complete a purchase—why would a human do any of those steps manually?
The platforms launching checkout features aren’t adding a nice-to-have. They’re positioning for a world where AI agents handle the entire purchase process, and the human just approves.
The question isn’t whether this will happen.
The question is which platforms will control it—and whether your business will be visible when they do.
The 18-Month Window
Here’s what I’d tell any merchant right now:
Months 1-6:
- Apply for ChatGPT merchant program (chatgpt.com/merchants)
- Ensure your products are in Google Merchant Center
- Audit your product data for completeness
- Integrate with Shopify or a platform that supports these protocols
Months 6-12:
- Optimize product descriptions for AI comprehension
- Build review volume and quality
- Implement real-time inventory updates
- Test purchases through each platform to understand the experience
Months 12-18:
- Analyze which platforms drive actual revenue
- Double down on winners
- Prepare for international expansion (most features are US-only for now)
The businesses that move now will have 18 months of learning before their competitors even start.
In platform shifts, timing matters more than talent.
This is the window.
What Comes Next
A few predictions:
Multi-item carts become standard. Right now, most platforms only support single-item purchases. That changes in 2026.
International expansion accelerates. US-only won’t last. Canada, UK, and Australia are likely first.
Payment competition intensifies. Stripe has ACP. PayPal has Perplexity. Google has Google Pay. Visa is exploring “invisible checkout.” The payment layer is the next battleground.
Agent-to-agent commerce emerges. Today, humans approve purchases. Tomorrow, your AI might negotiate with a merchant’s AI and complete a purchase without human involvement on either side.
Someone will build the “merchant of record” layer. Right now, individual merchants handle fulfillment, returns, and customer service. Someone will aggregate that into a service layer that handles everything post-purchase for AI commerce.
The next 18 months will define the next decade of commerce.
Pay attention.
If you found this useful, I write about how technology actually changes business—not the hype, but the infrastructure shifts that matter. Follow for more.
Resources:
- OpenAI Merchant Application: chatgpt.com/merchants
- ACP Specification: agenticcommerce.dev
- Google Shopping Graph: business.google.com/us/think/search-and-video/google-shopping
- Perplexity Shopping: perplexity.ai/hub/blog/shopping-that-puts-you-first
- Amazon Rufus: aboutamazon.com/news/retail/amazon-rufus-ai-assistant
