Amazon's AI Shopping Revolution: Rufus, Alexa+, and What Conversational Discovery Means for Sellers

Key Findings

Amazon Rufus — the conversational assistant

• Launch and rollout: Rufus launched in beta in the US on 1 February 2024, expanded to all US customers on 12 July 2024 (ahead of Prime Day), reached the UK on 4 September 2024, then Germany, France, Italy and Spain in beta on 19 November 2024, plus Canada and India. By early 2026 it was effectively available across all major Amazon marketplaces.
• How it works: Built on Amazon Bedrock, Rufus routes queries across multiple large language models — Anthropic's Claude Sonnet, Amazon Nova, and a custom model trained specifically on shopping data — and uses retrieval-augmented generation (RAG), reinforcement learning from customer feedback, and AWS Trainium/Inferentia chips. Per Amazon's own engineering blog (VP/Distinguished Scientist Trishul Chilimbi), it was "trained primarily with shopping data — the entire Amazon catalogue… as well as customer reviews and information from community Q&A posts," plus curated web data, and "pulls information from sources it knows to be reliable, such as customer reviews, the product catalogue, and community questions and answers, along with calling relevant Stores APIs."
• Adoption and impact: Amazon's Q4 2025 earnings (reported February 2026) confirmed Rufus was used by more than 300 million customers in 2025 and generated "nearly $12 billion in incremental annualised sales… exceeding the $10 billion pace CEO Andy Jassy projected" at Q3. (The 250-million-user figure cited at Q3 2025 was a point-in-time number; 300M+ is the full-year total.) Monthly active users grew 149% year over year, interactions rose 210%, and customers who engage Rufus are 60% more likely to complete a purchase. Commerce columnist Kiri Masters (The Drum) reported research showing AI-assisted Amazon sessions converted at 3.5x the rate of traditional sessions during the 2025 holidays.

Alexa+ — the generative-AI voice assistant

• Announcement and pricing: Unveiled 26 February 2025, priced at $19.99/month and free for Prime members, with a limited free text-chat tier at Alexa.com for non-Prime users.
• Rollout: Early Access began March 2025 (US); Amazon opened Alexa+ to all US customers on 4 February 2026. International rollout followed — UK Early Access began 19 March 2026 (£19.99/month, free with Prime), alongside Canada, Mexico, Italy, Spain, Germany, Austria and France. The UK version was localised to feel "genuinely British" (it understands "cuppa," "knackered," "nippy").
• Capabilities: Generative, agentic, and model-agnostic (Bedrock + Nova + Anthropic). Tens of millions of users; Amazon reports customers have 2–3x more conversations than with legacy Alexa. Daniel Rausch, Amazon VP of Alexa and Echo, told GeekWire that "76% of what customers do with Alexa+ is unavailable in any other AI" — citing smart-home control, family calendars, reservations and a decade of service integrations.

The convergence: "Alexa for Shopping"

• On 13 May 2026, Amazon combined Rufus and Alexa+ into "Alexa for Shopping," rolling it out to all US customers with no Prime membership or Echo device required. It appears in the main Amazon search bar, search results, product detail pages, and cart, and brings the full Amazon store experience to Echo Show via voice and touch. Rajiv Mehta, VP of Conversational Shopping, described it as "an expert personal shopper who already knows you." The standalone Rufus chatbot is being discontinued, though its recommendation features and shopping history carry over. This is the single most important recent development for sellers: voice and on-site shopping now share one personalized assistant, with conversation context flowing in both directions.

COSMO — the AI ranking/knowledge layer

• COSMO (Common Sense Knowledge Generation and Serving System) is Amazon's AI knowledge-graph system, documented in a peer-reviewed paper at SIGMOD 2024 (lead authors Changlong Yu, Xin Liu, Zheng Li et al., ten Amazon researchers plus an HKUST academic). It mines commonsense knowledge from massive shopper behavior to interpret intent — e.g. understanding that "winter clothes" implies warmth, or "shoes for the elderly" implies slip-resistance. Per the paper, COSMO-LM expanded Amazon's knowledge graph to 18 major categories from only ~30,000 annotated instructions and is deployed in search relevance, session-based recommendation, and search navigation, with measured offline and online (A/B) improvements.
• Important nuance: The paper documents COSMO's deployment in search navigation; it does not declare A9 retired. The widely held seller-community view that COSMO directly powers Rufus is probable but not officially confirmed by Amazon. Treat A9 as a retrieval layer that still operates underneath the AI intent layer.

Broader context: AI is becoming the discovery interface

• US: Adobe Analytics (Q1 2026 AI Traffic Report, released 16 April 2026, covering 1 trillion+ US retail visits) found AI-sourced traffic to US retail sites grew 393% year over year in Q1 2026, and in March 2026 AI traffic converted 42% better than non-AI channels — a new record. This is a stark reversal: in March 2025 AI traffic converted 38% worse than non-AI, roughly an 80-percentage-point swing in 12 months. AI-referred shoppers also engaged ~12% more and spent ~48% longer on site. Adobe's Vivek Pandya summarized it: "AI is quickly becoming the primary interface between consumers and their favorite brands."
• UK: Adobe (via Marketing Week) reported UK generative-AI retail traffic grew 180% year over year in March 2026, with AI-referred shoppers engaging 11% more than other channels; a year earlier UK AI visits had converted 80% worse than non-AI traffic, a gap that has since narrowed sharply. 63% of surveyed UK shoppers said they now use AI assistants at least weekly, and 66% trust that AI tools provide accurate results.
• GEO: "Generative Engine Optimization" — optimizing product content to be discovered and recommended by AI answer engines (Rufus/Alexa, ChatGPT, Google AI Overviews, Perplexity) — has emerged as a discipline distinct from, and layered on top of, traditional SEO.

A closer look

What Rufus/Alexa for Shopping is and how it works

Rufus is Amazon's generative-AI conversational shopping assistant, embedded directly in the Amazon Shopping app and website. Shoppers type or speak natural-language questions and receive conversational answers plus a curated set of recommendations — often making a decision inside the conversation without clicking into a traditional listing. Behind the scenes, a real-time router decides whether a query needs broad general knowledge, deep product detail, or fast search, balancing capability, latency and quality; RAG then pulls current, well-sourced evidence (reviews, catalog, Q&A, Stores APIs).

The kinds of questions shoppers ask, per Amazon's own examples, span:

• Broad/early-journey research: "What do I need for cold-weather golf?", "What to consider when buying running shoes?"
• Category comparisons: "What's the difference between lip gloss and lip oil?", "Compare drip to pour-over coffee makers."
• Specific product questions: "Is this pickleball paddle good for beginners?", "Is this jacket machine washable?"
• Order/account inquiries: "When did I last order AA batteries?", "Where is my order?"

Rufus has appeared in the search bar, on product detail pages, in the Amazon Lens Live visual-search experience, and across the homepage. Under "Alexa for Shopping," Amazon added AI overviews at the top of search results and on detail pages, side-by-side product comparison from search results, a full year of price history, Scheduled Actions (condition-based automations such as "add this sunscreen to my cart if the price drops to $10"), and cross-web shopping via "Shop Direct" and the agentic "Buy for Me" checkout.

Evolution timeline (key milestones): AI Shopping Guides (100+ categories, October 2024); Interests (26 March 2025); audio review/product summaries (2025); Help Me Decide (23 October 2025), which recommends a single product when shoppers are torn; 50+ technical upgrades (18 November 2025) adding account memory, price tracking and agentic auto-buy on price drops; 365-day price history (May 2026); and the Alexa for Shopping merger (13 May 2026).

How Alexa+ relates to and now converges with Rufus

Before the merger, the simplest framing was: COSMO is the brain (intent interpretation in search), Rufus is the on-site shopping voice, and Alexa+ is the cross-device, agentic assistant (smart home, calendars, bookings, plus general AI tasks). The May 2026 merger fuses Rufus's product expertise and shopping history with Alexa+'s personalization and agentic actions. For product discovery this matters because: (1) what a shopper tells Alexa on an Echo (e.g. brainstorming a science-fair project) now informs recommendations in the Amazon app, and vice versa; and (2) the assistant is no longer confined to a chat window — it's woven into the search bar, results, detail pages and cart where decisions actually happen.

Voice commerce reality check: Voice-driven discovery is growing but full voice-only purchasing remains modest. Roughly half of US consumers (~49.6%, ~154 million) use voice search for shopping-related activity, and ~74% have completed some part of a buying journey by voice — but only a small minority complete an entire purchase by voice (one eMarketer-cited estimate puts voice-only completed purchases around 11%; treat the exact figure as directional). Voice's sweet spot is replenishment and low-consideration reorders (groceries, household supplies, pet food), with average voice order values well below screen-based ecommerce. The strategic shift is that conversational, question-shaped queries (voice or typed) are 3–5x longer than keyword searches and overwhelmingly phrased as questions — which is exactly what AI-era listings must answer.

What this means for sellers

1. AI radically compresses the consideration set. Traditional search returned ~50 products and let the shopper choose; Rufus typically surfaces roughly five named products in a conversational answer. A listing that ranked on page one under A9 does not automatically earn an AI recommendation. Being in the answer is now the whole game.
2. Intent and clarity beat keyword density. Amazon's systems now reward contextually rich, natural-language content that explains who the product is for, what problems it solves, and in what real-world situations it belongs. Keyword stuffing actively hurts: "any language model, Rufus included, evaluates semantic coherence," and incoherent keyword strings provide near-zero useful signal.
3. Structured backend data feeds COSMO. Completing every relevant attribute field in Seller Central (subject matter, intended use, material, specific uses, granular item-type keyword) hard-codes products into the right "intent buckets" in the knowledge graph. Practitioners note LLMs prefer clean, labeled structured fields over prose, treating verified attributes as higher-confidence than "marketing fluff."
4. Reviews and Q&A are first-class discovery inputs. Rufus draws answers from reviews and community Q&A; review quality (helpful-vote rate, reviewer credibility) is weighted, not just count. Reviews that mention specific use cases and personas (e.g. "perfect for nurses on long shifts") can cause Rufus to surface a product for adjacent queries.
5. Off-Amazon authority matters. Amazon confirms Rufus is trained on information "from across the web," so a brand's editorial coverage, expert mentions and Reddit/review-site presence influence how the AI contextualizes products — the basis of Amazon-flavored GEO.
6. Paid is entering the conversation. Amazon introduced Sponsored Products Prompts and Sponsored Brands Prompts in open beta in November 2025; they moved to general availability with CPC billing on 25 March 2026. These auto-generated prompts (e.g. "Why choose [brand]?") are drawn from a seller's detail page, Brand Store and campaign data, are opt-out (not freely authored), and surface inside Rufus conversations and on detail pages. The Sponsored Products Prompts report shows exactly which conversational questions Amazon associates with a product — valuable intelligence for listing optimization.

Recommendations

Stage 1 — Foundational (do now, low cost)

• Rewrite titles, bullets and descriptions in benefit-led natural language that connects features to outcomes and names use cases, occasions and target personas. Replace "dog bed large washable orthopedic" strings with "Orthopedic dog bed for large breeds with joint pain — washable memory foam that supports hips and spine."
• Fill every relevant backend attribute field; use the most granular item-type keyword available (e.g. "running-shoes," not "shoes").
• Keep A9 keyword fundamentals intact (high-intent keyword in the first ~80 characters, backend search terms) — traditional keyword search still drives the majority of purchases.

Stage 2 — Differentiation

• Seed comprehensive Q&A and cultivate detailed, specific reviews that reference real use cases and buyer personas.
• Invest in A+ Content and at least five images including infographics and lifestyle shots, since Rufus reads images via OCR and favors complete, story-driven modules.
• Build off-Amazon authority (editorial mentions, expert content) to strengthen GEO across Rufus, ChatGPT and Google AI.

Stage 3 — Paid and measurement

• Pull the Sponsored Products Prompts report; study which conversational questions Amazon ties to your products, and reinforce those topics in your detail pages. Build relevance now while prompt CPCs are nascent.
• Allow 7–14 days after listing changes before judging Rufus/COSMO impact — the semantic layer updates more slowly than A9 keyword ranking.

Benchmarks that should change your strategy: If Rufus/AI-assisted sessions move from a minority to a majority of your category's traffic, shift budget decisively from keyword-volume tactics to intent/GEO optimization. If Sponsored Prompts CPCs begin clearing at meaningful rates (post-March 2026 monetization), treat that above-the-fold conversational placement as premium inventory and bid accordingly. If your category shows the Adobe pattern (AI traffic converting at or above non-AI), prioritize machine-readability of your listings and brand content.

What we still don't know

Two honest caveats before you act on any of this. First, Amazon has published no official Rufus or Alexa for Shopping optimisation playbook. Its newsroom and science blog describe what the assistant draws on — the catalogue, reviews, Q&A and the wider web — but the tactical advice in this guide, like everyone else's, is informed inference from those sources rather than Amazon doctrine. Second, AI-mediated sessions are still a minority of total Amazon shopping activity as of early 2026, even if they're growing fast. Traditional keyword search still drives most purchases, so treat the steps above as how to get ahead of the shift — not a licence to abandon the A9 fundamentals that pay the bills today.

Charlie Banks

Founder of CVR Studios and an active Amazon seller. CVR Studios has launched and scaled 125+ brands on Amazon. More about CVR →