Amazon's Walled Garden Strategy for AI Shopping

Update, October 2026: Amazon renamed Rufus to Alexa for Shopping on May 13, 2026. On August 4, 2026, the Ninth Circuit vacated the injunction against Perplexity's Comet described below.

On March 9, a federal judge blocked Perplexity's Comet browser from accessing Amazon — Judge Maxine Chesney in the Northern District of California found "strong evidence" of unauthorized access under the CFAA. Perplexity got a 7-day stay to appeal to the Ninth Circuit. Deadline: March 16.

TL;DR: Amazon controls $830 billion in GMV and blocks every external AI agent while building its own: Rufus, used by more than 300 million customers in 2025 and credited with nearly $12 billion in incremental annualized sales (About Amazon, 2026). The rest of ecommerce can join open protocols like UCP, or accept Amazon as the default AI shopping layer. There's no third option.

That ruling isn't just about one browser. It answers a question that's been hanging over agentic commerce since I started investigating how AI agents actually shop: can a platform lock AI agents out?

Amazon's answer: yes. And not just legally — structurally, technically, and commercially. While everyone else debates open protocols, Amazon already built the walls. Here's what I found when I dug into how tall those walls actually are.

What Does Amazon Actually Control?

Amazon processed $830 billion in gross merchandise value in 2025 (Marketplace Pulse, 2026). That number alone explains the strategy. With an estimated 2 million active sellers (SalesDuo, 2026), more than 200 million Prime members at its last official count (Amazon, 2021), and over 13 billion items delivered same or next day last year (About Amazon, 2026), Amazon doesn't need to negotiate with external AI platforms. It already is the platform.

I tried mapping the full stack on a whiteboard. Rufus handles product discovery: more than 250 million customers had used it by November 2025, with monthly average users up 149% year-over-year (About Amazon, 2025). The marketplace provides supply. Amazon Pay handles money. FBA handles fulfillment. And the piece that locks the whole thing together: advertising, at $68.6 billion in 2025 (Amazon 10-K, 2026).

I couldn't find a single layer that requires an external partner. In my previous post on China's agentic commerce, I noted that Alibaba runs the exact same playbook — own the chatbot, the marketplace, the payment rail. Amazon doesn't need to worry about interoperability because there's nothing to interoperate with.

Why Did Amazon Block Every External AI Agent?

I checked Amazon's robots.txt myself. "ChatGPT-User" and "OAI-SearchBot" — both blocked since November 2025, according to Modern Retail. An earlier round of blocks had already cut ChatGPT's referral traffic to Amazon by nearly 18% month over month in August 2025 (Modern Retail, 2025). Google's AI agents got the same treatment.

Why would a retailer block free traffic? I kept turning this over.

The answer is $68.6 billion. Amazon's ad business generated $21.3 billion in Q4 2025 alone, up 23% year-over-year (PPC Land, 2026). When an AI agent shops for a user, it skips the sponsored listings entirely. As Practical Ecommerce puts it, autonomous agents bypass the ads: instead of scrolling through sponsored products, "the AI shopping agent skips to an item and initiates checkout." That threatens a US retail media search market that eMarketer expected to reach nearly $38 billion in 2025 (Digiday, 2026). Search ads make up about 60% of all retail media spend.

Here's what that means in practice: Amazon isn't anti-AI-shopping. It's anti-anyone-else's-AI-shopping. Rufus gets full access to every product listing, every review, every purchase history signal. External agents get a locked door.

Andy Jassy made the logic explicit on Amazon's Q4 2025 earnings call. Given a choice between a "horizontal agent" and a retailer's own, "a lot of customers are ultimately going to choose to use a great shopping agent from that retailer," he said, as Retail Brew reported. On the previous call, in October 2025, his critique of external agents was blunt: "there's no personalization. There's no shopping history. The delivery estimates are frequently wrong. The prices are often wrong" (Q3 2025 call transcript).

He's not entirely wrong — I tested Perplexity's shopping agent before the injunction, and the delivery estimates were frequently off by 2-3 days versus Amazon's own predictions. Third-party agents do lack purchase history and delivery precision. But framing that as a reason to block rather than a problem to solve through open data sharing tells you what the real priority is. The $68.6 billion ad business doesn't protect itself.

How Does Rufus Compare to Open Protocol Shopping?

Amazon's Rufus assistant is already one of the largest AI shopping tools in existence. More than 300 million customers used it in 2025, according to About Amazon (2026). Interactions grew 210% year-over-year (About Amazon, 2025). Across the 2025 holiday season, Black Friday included, Rufus sessions converted at 3.5 times the rate of non-Rufus sessions, according to Sensor Tower research reported by The Drum (2026).

That conversion number deserves a closer look. Rufus users are 60% more likely to complete a purchase than shoppers who don't use it, according to Yahoo Finance (2025). On the Q3 2025 earnings call, Andy Jassy said Rufus was expected to generate over $10 billion in annual incremental sales. And in October 2025, Amazon launched "Help Me Decide" — a feature that recommends products based on a shopper's browsing, searches and shopping history, not just the current search.

The UCP coalition, by contrast, has 20-plus partners including Walmart, Target, Best Buy, Macy's, and Visa. It's the open standard backed by Google and Shopify. But here's the tension: no UCP agent has access to what Rufus has. No external agent knows what you bought last month, what you returned, or when your Prime delivery window opens. The data advantage is structural.

I kept coming back to a question while researching this: is Rufus actually better, or is it just operating on a playing field nobody else can access? I tried asking Perplexity and ChatGPT to recommend running shoes while logged into Amazon with Rufus open on another tab. Rufus knew my size, my brand history, my Prime delivery window. The external agents had none of that. The conversion data is real — but it's real partly because Rufus has an information advantage that no open protocol can match without Amazon's cooperation.

Then there's Alexa+. Amazon rolled it out nationwide on February 4, 2026 (CNBC, 2026). Free for Prime members, $19.99/month standalone. The critical feature: autonomous purchasing. Alexa+ can auto-buy when price targets hit, with no confirmation required. And Amazon counts over 600 million Alexa devices already out there (About Amazon).

What Is Buy for Me Really Doing?

While Rufus keeps shoppers inside Amazon, a newer tool pushes Amazon's reach outward. Shop Direct lists products from other merchants' sites inside Amazon, and its "Buy for Me" button lets Amazon's AI agent complete the purchase on the merchant's site. Shop Direct now includes over 100 million products from more than 400,000 merchants, with tens of millions of them available through Buy for Me, according to About Amazon (March 11, 2026). Merchants can submit product feeds through datafeed partners like Feedonomics, Salsify, and CEDCommerce.

But not everyone opted in.

According to CNBC (January 2026), more than 180 businesses told Bobo Design Studio CEO Angie Chua that their products had been listed on Amazon without their permission. Chua told CNBC: "We were forced to be dropshippers on a platform that we have made a conscious decision not to be part of." Amazon's response was an opt-out email: branddirect@amazon.com. In March, Amazon said a merchant portal is coming soon.

The hypocrisy is hard to ignore. Amazon sued Perplexity for scraping Amazon product pages. Then built a tool that scrapes independent merchant sites to populate an Amazon shopping experience. Perplexity scraping Amazon? Lawsuit. Amazon scraping small retailers? Product feature.

And with the Perplexity v. Amazon injunction from March 9, Amazon now has case law on its side. Judge Chesney ruled that platform authorization matters — a user giving an AI agent permission to access their account isn't enough. The platform itself must also authorize the agent. If that reasoning holds on appeal (Ninth Circuit deadline: March 16), it hands Amazon a legal framework to permanently exclude external agents while freely extending its own reach into other merchants' sites.

Is Amazon's Closed Strategy Actually Working?

Amazon's biggest competitor in US retail chose the opposite approach. Walmart joined UCP, partnered with both Google and OpenAI, and has been vocal about it. At NRF, Walmart EVP Hari Vasudev said, as Modern Retail reported: "That approach of open partnerships is something that we strongly believe in."

The early numbers back it up. Walmart customers who use its AI assistant Sparky have an order value about 35% higher than those who don't, according to Modern Retail. Roughly half of Walmart's app users have tried it.

So which approach wins? The honest answer: I don't know yet. Amazon's walled garden got nearly $12 billion in incremental annualized sales from Rufus alone in 2025, and Rufus sessions converted at 3.5 times the rate of other sessions over the holidays. Walmart's Sparky users place orders about 35% larger, and Walmart plugs into outside AI agents: it has deals with OpenAI and Google to let shoppers buy its products inside ChatGPT and Gemini. Amazon protects its $68.6 billion ad business by keeping agents out. Walmart can afford openness because its ad business is smaller and it has less to lose from agent bypass.

Meanwhile, Amazon is quietly playing both sides through AWS. Bedrock AgentCore offers serverless hosting for AI agents and MCP servers. Visa plans to host the MCP tools for its Intelligent Commerce payments on it (AWS, 2025). According to Gartner, 40% of enterprise applications will be integrated with task-specific AI agents by the end of 2026, up from less than 5% in 2025. Amazon is happy to sell the infrastructure for open AI commerce — it just won't let those agents shop on Amazon.com.

What Does This Mean for Everyone Outside Amazon?

For the sellers on Amazon's marketplace, the path is already set. Rufus sells for you. Alexa+ sells for you. Your job is product quality, pricing, and Amazon's ad flywheel. Nothing changes — which is both the comfort and the trap.

For merchants outside Amazon, the choice is more complicated. The open protocols — UCP, ACP — represent the alternative. McKinsey estimates agentic commerce could reach up to $1 trillion in US retail by 2030. Morgan Stanley projects AI shopping agents could account for $190 billion to $385 billion of US e-commerce spending by then. That's an enormous market forming outside Amazon's walls.

But it's forming slowly. And in the gap, Amazon launched Shop Direct — bringing 400,000 merchants inside its ecosystem whether they chose it or not. The legal precedent from the Perplexity injunction makes this asymmetry harder to challenge.

Here's what keeps bugging me about the whole picture. Amazon doesn't need to win the protocol war. It doesn't need UCP or ACP. It just needs external agents to remain worse than Rufus — worse data, worse delivery estimates, worse personalization. As long as the infrastructure gap stays open, Amazon's walled garden looks less like a prison and more like a luxury hotel.

Frequently Asked Questions

Does Amazon support UCP or ACP protocols?

No. Amazon is absent from both major agentic commerce protocols. UCP, backed by Google and Shopify, has 20-plus partners including Walmart, Target, Best Buy, Macy's, and Visa. ACP, backed by OpenAI and Stripe, also excludes Amazon. Instead, Amazon built its own closed AI shopping tools — Rufus, Alexa+, and Shop Direct — that operate entirely within its ecosystem.

Why did Amazon block AI agents from crawling its site?

Amazon blocked Perplexity's and Google's AI agents in its robots.txt in mid-2025 and added two more OpenAI crawlers, ChatGPT-User and OAI-SearchBot, in November 2025 (Modern Retail). The primary motivation appears to be ad revenue protection. Amazon's advertising business generated $68.6 billion in 2025 (Amazon 10-K, 2026). AI agents bypass sponsored product listings entirely — when an agent shops for a user, no one sees the ads. That threatens Amazon's fastest-growing revenue stream.

How does the Perplexity v. Amazon ruling affect AI shopping?

Judge Maxine Chesney's March 2026 preliminary injunction blocked Perplexity's Comet browser from accessing password-protected Amazon sections under the CFAA. The key precedent: a platform must authorize the AI agent, not just the user. This potentially gives any major retailer legal grounds to block external AI shopping agents. Perplexity is appealing to the Ninth Circuit with a March 16 deadline.

Can merchants opt out of Amazon's Shop Direct?

Yes, but only through an opt-out process. According to CNBC (January 2026), more than 180 businesses reported finding their products listed on Amazon without their permission. Amazon's current opt-out method is emailing branddirect@amazon.com. Amazon says a merchant portal is coming soon. The default is opt-in — your products may already be listed.

Is Walmart's open approach better than Amazon's closed one?

They solve different problems. Walmart's open approach — joining UCP, partnering with Google and OpenAI — comes with an AI assistant, Sparky, whose users place orders about 35% larger than other customers (Modern Retail). Amazon's closed approach drove nearly $12 billion in incremental annualized sales through Rufus in 2025, and Rufus sessions converted at 3.5 times the rate of other sessions over the holidays. Walmart can afford openness because its ad business is smaller. Amazon's $68.6 billion in ad revenue creates a structural incentive to keep agents out.

Conclusion

Amazon's walled garden isn't a temporary position — it's a strategy backed by $830 billion in GMV, $68.6 billion in ad revenue, and now federal case law. The walls are structural, not ideological.

But the rest of ecommerce doesn't have to accept it. Open protocols exist, Walmart is proving the alternative works, and McKinsey projects up to $1 trillion in US agentic commerce by 2030 — far more than any single company can capture.

The question isn't whether Amazon's walled garden will hold. It's whether everything outside those walls can organize fast enough to matter before Amazon's Shop Direct absorbs them anyway. The stay on the Perplexity injunction runs out on March 16, and the Ninth Circuit's ruling could reshape who gets to build AI shopping and who gets locked out.

Part 4 of "The Agentic Commerce Investigation" series. Previously: Which Ecommerce Stores Show Up When AI Goes Shopping?, The Agentic Commerce Infrastructure Gap, China Is Already Living in the Agentic Commerce Future.