Amazon Rufus Now Shops on Its Own. What Sellers Should Do.

Amazon expanded Rufus Scheduled Actions to all U.S. customers this week, letting the assistant restock pet food, queue gift ideas, and execute price-triggered buys with zero prompt from the shopper. Rufus already touched 300 million customers in 2025, and those users were 60% more likely to complete a purchase. The walled garden just got autonomous.

TL;DR: Amazon Rufus rolled out Scheduled Actions to all U.S. shoppers in April 2026, turning a chat assistant into a recurring auto-buyer for 300M+ customers. For sellers, this is the first agentic surface where discovery, decision, and checkout all happen inside one platform's rails. The brands that win Rufus visibility will not be the same ones winning ChatGPT or Gemini, because Amazon optimization runs on a different stack. Sellers need a parallel walled-garden plan now.

What Did Amazon Just Ship With Scheduled Actions?

Scheduled Actions lets Rufus execute recurring shopping tasks without a fresh prompt. While chatting, a customer taps "+" to create a task such as "restock dog food monthly" or "alert me when this face cream hits $75." Rufus runs the research, picks the product, and either notifies the shopper or drops items in cart for one-click checkout. Amazon's own product page calls this "agentic auto-buy."

Amazon disclosed the underlying scale in its April 21 update: roughly 300 million U.S. customers used Rufus in 2025, and those users converted 60% more often than non-Rufus shoppers. CEO Andy Jassy framed Scheduled Actions as the next phase, and analysts immediately read it as "an early version of fully agent-driven commerce." See Amazon's own announcement of the feature set for the full list, which now also includes price-history tracking, price-alert auto-buy, and handwritten grocery list ingestion.

Key stat callout: 300 million U.S. customers used Rufus in 2025; those customers were 60% more likely to complete a purchase than non-Rufus shoppers (source: Andy Jassy via Yahoo Finance, Apr 21, 2026).

The strategic move is what matters. Amazon is suing Perplexity for trying to do agentic shopping from outside Amazon's walls, while simultaneously normalizing autonomous purchases inside its own app. That is not a contradiction; it is a positioning statement. Amazon wants the agentic loop to close on Amazon. Discovery, comparison, decision, transaction, fulfillment, returns, all on the same rails.

Why This Is Different From ChatGPT or Gemini Shopping

Outside Amazon, agentic commerce runs on open protocols. ACP powers ChatGPT and Microsoft Copilot discovery, then redirects shoppers to the merchant's own checkout page. UCP powers Google AI Mode and Gemini through Shopify-syndicated catalogs. Both protocols treat the merchant site as the destination. We covered the open-protocol landscape in detail in the 2026 protocol guide.

Rufus does not work that way. Rufus is a single-vendor AI shopping agent that retrieves from Amazon's own catalog, ranks with Amazon's own signals, and checks out on Amazon's payment rails. There is no merchant redirect. The shopper never leaves the app. That has three consequences for sellers.

First, your Rufus visibility is decoupled from your visibility on ChatGPT or Gemini. Different retrieval engine, different ranking signals, different feed format. Winning one does not mean winning the other.

Second, Rufus rewards the same fundamentals that win Amazon search: complete A+ content, accurate attributes, depth of customer reviews, Subscribe & Save eligibility for replenishment categories. But it adds a layer on top. Rufus reads listings to answer specific questions ("is this gluten-free", "does this fit a 6-month-old"). Listings written for keyword stuffing will be skipped. Listings written for human readability and attribute completeness will be read.

Third, recurring-purchase categories are about to consolidate fast. Once a shopper sets up a Scheduled Action for a household-goods SKU, the friction to switch brands collapses. The brand that gets selected on month one keeps the slot until something breaks.

How Sellers Should Read the Walled-Garden Split

The agentic stack is splitting cleanly into two surfaces, and brands need a strategy for each. The table below maps the differences.

This is the same split that already exists in retail media (Amazon Ads vs. Google Shopping Ads) and the same split that already exists in marketplaces (Amazon Seller Central vs. your own Shopify store). Agentic commerce just inherited it. The brands that already have a sophisticated Amazon team and a separate DTC team will adapt fastest. The brands that treat Amazon as a copy-paste of their DTC catalog will lose visibility on both surfaces.

There is also a reason this matters more for CPG and household goods than for fashion or electronics. Replenishment is the use case Amazon highlighted first, and it maps to categories where Subscribe & Save already does heavy lifting. CPG brands need to assume Rufus Scheduled Actions becomes a meaningful share of their Amazon revenue within 12 months and plan accordingly.

What Stays The Same Across Both Surfaces

The discouraging read is that sellers now have two optimization tracks. The encouraging read is that the underlying input, clean and structured product data, feeds both. Structured product data with rich attributes makes a listing more readable to Rufus and more retrievable on ChatGPT. The same enrichment that wins one surface tends to help the other.

This is the core of agentic commerce optimization. Treat your catalog as the product, not as marketing copy. A SKU that has 30 well-defined attributes (material, dimensions, fit, ingredient list, certifications, use cases, allergens, compatible accessories) gets parsed correctly by every retrieval system, walled or open. A SKU that has 5 attributes and a pretty hero image only gets parsed by the systems already built around guessing.

The other thing that travels is the digital shelf discipline. Brands that already monitor share-of-voice on Amazon search are halfway to monitoring share-of-voice in Rufus answers. Brands that already monitor share-of-voice in ChatGPT shopping results are halfway to monitoring it in Gemini results. The mental model transfers.

What to Do This Week

1. Audit your top 25 Amazon SKUs by replenishment frequency. These are the listings most likely to land in Scheduled Actions first. Confirm Subscribe & Save is on, attributes are complete, A+ content is current, and review velocity is healthy.
2. Run the same 25 SKUs through Rufus directly. Ask product-aware questions ("which of these face creams is fragrance-free", "which dog food is good for senior small breeds"). Note where Rufus picks a competitor and why the listing won.
3. Separate your AI optimization roadmap into two tracks. Track A is open agents (ChatGPT, Gemini, Perplexity, Copilot) fed via your product feed and on-site schema. Track B is walled gardens (Rufus, eventually Walmart's assistant) fed via marketplace listings. Same underlying data, two delivery formats.
4. Stand up monitoring for both tracks. Tools like our AI Readiness Report cover open agents; combine that with weekly Rufus prompt tests for the walled-garden side. If your products show up in 2 of 25 prompts, that is your baseline.
5. Brief leadership on the recurring-purchase risk. Once a Scheduled Action locks in a competitor, switching costs are real. The window to defend a slot is the next two quarters, not the next two years.

Frequently Asked Questions

What is Amazon Rufus Scheduled Actions?

Scheduled Actions is a Rufus feature that lets shoppers automate recurring purchases or alerts. Examples include monthly restock of pet food, gift ideas tied to upcoming birthdays, or buying a product once it drops below a price threshold. Amazon expanded it to all U.S. customers in April 2026.

Does Rufus use the ACP or UCP protocol?

No. Rufus retrieves from Amazon's own catalog using internal systems. ACP and UCP are open protocols that power discovery on ChatGPT, Microsoft Copilot, Google AI Mode, and Gemini, where the merchant site is the destination. Rufus is a closed walled-garden agent and uses neither.

How does Rufus rank products?

Rufus combines classic Amazon relevance signals (search performance, reviews, A+ content, in-stock status, Prime eligibility) with NLP-driven listing comprehension that reads attributes and copy to answer specific questions. Listings written for keyword stuffing perform worse than listings written for human readability and attribute completeness.

Should I optimize for Rufus the same way I optimize for ChatGPT?

Partly. Both reward complete, structured product data and accurate attributes. They differ on delivery: ChatGPT pulls from your product feed and merchant site, while Rufus pulls from your Amazon listing. The underlying enrichment work is shared, but the channels need separate publishing pipelines.

What categories will Scheduled Actions hit hardest first?

Replenishment-heavy categories: pet food, household goods, paper products, supplements, baby and kids consumables, beauty staples. Anything that already has high Subscribe & Save attach is the first target. Gift-adjacent categories tied to recurring calendar events (birthdays, holidays) come next.

Does this change anything for my Shopify or DTC site?

Not directly, but it raises the importance of your open-agent visibility. As Amazon closes its loop on Amazon shoppers, your DTC traffic will increasingly come from open agents like ChatGPT and Gemini, where the customer still lands on your checkout page. The walled-garden split makes both tracks more strategic, not less.

Amazon shipping autonomous purchases at 300M-customer scale is the clearest signal yet that agentic commerce is no longer a feature on a roadmap. It is the default behavior on the largest commerce platform in the country. Sellers that treat their Amazon catalog and their open-web product feed as one shared source of truth, optimized for retrieval rather than for human browsing, will keep their slots in both Scheduled Action queues and ChatGPT product cards. The rest will spend the next two quarters wondering why their replenishment revenue stopped growing.