On May 13, 2026, Amazon did something it had been building toward for two years: it stopped treating its AI shopping assistant as a feature and started treating it as the front door. Rufus, the chatbot that lived in a corner of the shopping app, was folded into Alexa+ and relaunched as Alexa for Shopping, available to every signed-in US customer, no Prime membership or Echo device required, embedded directly in the main Amazon search bar. The company’s own press release calls it “the world’s best, most personalized AI assistant for shopping.”

That claim deserves to be taken seriously, because on raw capability it is close to true. It also deserves to be read the way a merger filing gets read, because Alexa for Shopping is not just an assistant. It is the collision of four Amazon interests into a single conversational surface: the store, the ad auction, the review corpus, and the litigation strategy against every rival agent that wants to shop on the customer’s behalf. Each of those four interests is individually defensible. Stacked inside one interface with no external verification, they form the largest unexamined trust question in commerce.

What Actually Shipped

Start with the product, because the product is genuinely strong. Per Amazon’s announcement, Alexa for Shopping combines Rufus’s product expertise with Alexa+‘s personal context, and the feature list reads like a checklist of everything agentic commerce has been promising:

  • Questions in the main search bar, so the assistant is no longer a separate destination but the default way to query the store, from “What’s a good skincare routine for men?” to “Compare Kindles.”
  • AI-generated overviews at the top of search results and on product detail pages, already rolling out to all US shoppers.
  • A full year of price history on hundreds of millions of products, with price alerts and Auto Buying at a target price.
  • Scheduled Actions that add items to the cart on triggers, including conversational conditions like “add this sunscreen to my cart if the price drops to $10 and I haven’t purchased it in the last 2 months.”
  • Shop Direct and Buy for Me, an agentic checkout that completes purchases at other retailers’ stores using your saved address and card.
  • Shopping memory that flows across Echo, phone and laptop: what you asked Alexa yesterday shapes what the search bar suggests today.

The personalization thesis is stated plainly by Rajiv Mehta, Amazon’s vice president of Conversational Shopping: the assistant is “like having an expert personal shopper who already knows you and remembers your preferences, your past purchases, and your conversations, and carries that knowledge and understanding of you across your phone, laptop, and Echo devices,” so “you don’t have to start over.”

It is a compelling pitch. It is also a map of every input that now sits inside one company’s discretion.

The Numbers, Honestly Counted

The scale claims come from three different measurement regimes, and they should be kept separate, because the honest composite is more interesting than any single headline.

Amazon’s own figure, from the launch announcement: Rufus helped more than 300 million customers during 2025. On the following earnings call, per StartupHub’s reconstruction, Andy Jassy told investors monthly active users were up more than 115 percent and engagement up nearly 400 percent year over year, and Amazon has said customers who use the assistant are more than 60 percent more likely to complete a purchase.

The third-party traffic picture: Similarweb’s US desktop index of assistant-driven visits landing on Amazon product pages went from an indexed 100 in June 2025 to about 590 in May 2026, with a holiday peak near 730. That is roughly a sixfold relative move, and as MarketMaze’s careful read of the exhibit notes, it counts clicks from assistant surfaces to product pages, not headcounts, and it happened mostly before the Rufus name was retired.

The proportion, which is the number that keeps everyone honest: analysis from Am I Cited puts the assistant at roughly 13.7 percent of Amazon searches, with fewer than 3 in 100 Amazon purchases involving it today. The 60 percent conversion lift, as StartupHub correctly cautions, describes people who open an assistant mid-shopping-journey: high-intent shoppers convert, which does not prove the assistant created the intent.

So: the largest AI shopping assistant in the world, by a wide margin, still early in its takeover of the funnel. That combination is exactly why the next three sections matter now rather than later.

The Door, Closed from Inside

While Amazon builds the biggest shopping agent, it has been systematically blocking everyone else’s. Per SellerApp’s industry analysis, Amazon has blocked 47 AI crawlers and updated its terms of service to prohibit AI agent behavior on its storefront, on the explicit logic that its ads business “depends on controlling the research phase of shopping.” When buyers research in ChatGPT or Perplexity and arrive at Amazon already decided, sponsored placements stop monetizing that intent.

The legal front made the same point more forcefully. In March, a federal judge blocked Perplexity’s Comet browser from shopping on Amazon on users’ behalf, then stayed the order pending appeal, the case we examined when the Ninth Circuit took it up. Amazon’s statement on the ruling deserves preservation in amber: “an important step in maintaining a trusted shopping experience.”

The irony is structural, not incidental. A platform currently defending the integrity of third-party agents’ access to its catalog is the same platform whose assistant now narrates that catalog conversationally, reads its reviews, summarizes its product pages, and builds carts autonomously. Jassy compared third-party shopping agents to the early search engines that sent traffic to e-commerce, and set the goal as building “the best shopping assistant anywhere.” One direction of agent traffic is a trust threat. The other is the product roadmap.

StartupHub’s public Agent Readiness scan grades how hospitable a site is to outside agents: 27 checks across discoverability, content, access and machine-readable purchasing. Amazon scored 41 of 100, a D. Google scored 49, Perplexity 40, OpenAI 39. None of the companies building shopping agents make it easy for someone else’s agent to operate on their storefront, but Amazon is the only one whose storefront is the market.

The Ad Auction Moves Into the Conversation

Here is where the trust question stops being theoretical. Sponsored placements are already inside the assistant.

Amazon began testing sponsored ads within Rufus conversations in September 2024, confirmed via its own ads API changelog, per Zonguru’s guide. Since then the monetization surface has formalized: SellerApp documents Sponsored Prompts becoming a billable CPC placement, and Pacvue’s holiday guidance lists what advertisers can now buy: Sponsored Products and Sponsored Brands prompts, Alexa+ Conversational Ads, and Prime Video Sponsored Tiles.

So the interface that answers “what’s the best protein powder for me” is operated by the same company that sells placement to protein powder brands, in the same session. Whether a recommendation is earned or paid is invisible to the shopper by design, and the FTC does not consider Amazon a disinterested party on this exact question: on August 31, 2026, the Commission and 22 states sued the company over an alleged secret ad surcharge scheme in its auctions. The defendant in that suit is the referee in this one.

To be fair, conversational advertising is not inherently illegitimate; it is undisclosed influence inside a personalized recommendation that is. A search results page at least visually separates sponsored tiles. A conversational answer mixes its inputs into prose, and shopping memory makes every answer different per user, so no two shoppers can compare what they were told.

What the Assistant Reads

Underneath the assistant sits the corpus, and the corpus has a documented contamination rate.

Amazon’s retrieval layer, per the Am I Cited teardown, draws on five source categories: descriptions and bullet points, A+ content, customer reviews, Q&A, and product images. Reviews supply “the parts no seller writes,” which is precisely why purchased reviews are written to imitate them.

The fake review economy this corpus filters is industrial in scale. Amazon’s own Trustworthy Shopping accounting says it proactively blocked more than 275 million suspected fake reviews in 2024 alone. Pangram’s detector analysis of 30,000 front-page Amazon reviews found 3 percent now AI-generated, even after the FTC’s rule making fake AI reviews expressly illegal. The FTC’s August consumer alert on brushing scams, seed packages used to manufacture verified-looking purchase histories, shows the supply side adapting faster than enforcement.

Amazon’s own detection writeup contains the detail that should anchor this whole debate: its models analyze “whether the seller has invested in ads (which may be driving additional reviews).” Amazon’s engineers know that advertising spend and review inflation travel together. That signal is used to protect a ranking input. It is not surfaced to the shopper, and it does not currently appear anywhere in the assistant’s conversational answers.

Amazon’s AI review summaries state they “use only our trusted review corpus from verified purchases.” Trusted by whom is the operative question: the entity that monetizes the listings, sells the ads, and is being sued over how it ranks them.

Memory as Moat

The final layer is the flywheel. Shopping memory spans purchases, browsing, review history, Prime Video viewing and Alexa interactions, and it cuts in one direction: toward what you already buy. Similarweb’s most-cited warning about the assistant, relayed in the StartupHub analysis, is of a two-tier dynamic inside the interface, where past purchase memory creates a retention engine for incumbent products, and skipping Sponsored Products risks exclusion from the consideration set.

For a shopper, memory is convenience. For an incumbent brand, memory is an annuity. For a better product with no purchase history and no ad budget, the conversational shelf is the hardest shelf in retail to reach, and no one outside Amazon can measure the bias, because no one outside Amazon can see the ranking.

This is the Google AI Mode problem at greater scale: recall the Productrise finding we covered last week, matched products surfacing 21.6 percent pricier inside Google’s AI answers than in classic search. At Google, at least, the Shopping Graph and the assistant are separable systems a third party could audit. At Amazon, store, ads, corpus, assistant and payments are one legal entity, and the interface is a black box with a purchase button.

What Independent Verification Looks Like

None of this argues against AI shopping assistance; it argues for a layer whose incentives do not run through the transaction. The design requirements are now easy to state:

  • Filter before scoring. Smart Score runs 0 to 100 on review quality and authenticity, computed after manipulated and low-information reviews are removed, so purchased rating volume stops being an input to trust, whether the reader is a human or a model.
  • Score persistence over spikes. GoBuy Verified requires holding a filtered score of 80 or above for 90 days, which is precisely the test that separates stable product quality from a purchased window engineered around a shopping event.
  • Curation over auction. Top seven verified products per category, not a consideration set that paid ads can enter invisibly.
  • Machine-native delivery. Any agent queries the filtered layer over MCP at gobuy.ai/api/mcp before recommending or buying, and the Chrome extension injects the same trust panel directly onto the Amazon pages the house assistant narrates. When the referee owns a team, the box score has to come from outside the stadium.

What to Watch

Four signals over the next two quarters. First, disclosure inside conversational answers: whether sponsored placements in Alexa for Shopping carry labeling a court would accept as clear and conspicuous, or whether FTC v. Amazon’s discovery reaches assistant ranking. Second, the Ninth Circuit appeal in the Comet case, which will decide whether third-party agents may ever shop on Amazon lawfully, and thus whether any external check on the house assistant survives contact with terms of service. Third, whether Amazon publishes any audit methodology for AI overviews and review summaries, or whether “trusted corpus” remains a self-certification. Fourth, the proportion number: when assistant-mediated purchases cross from 3 in 100 toward 3 in 10, every bias above stops being an academic concern and becomes the default consumer experience of the largest store on earth.

The assistant will be convenient, and it will be everywhere. Verify before you, or anything acting for you, buy: gobuy.ai, with agent integration docs at gobuy.ai/agent-docs.