Amazon's Agentic Commerce Stack: The Complete Guide to Rufus, Alexa+, Buy for Me, and Nova

Amazon runs the largest closed-loop agentic commerce ecosystem in the world — discovery, recommendation, transaction, and fulfillment inside one wall. Most brand strategy treats it as "the Amazon channel." That's the mistake this guide fixes.

Why Amazon deserves its own guide, not just a footnote

The public conversation about agentic commerce is dominated by ChatGPT, Gemini, and Perplexity because those are the surfaces most brands understand. Amazon gets one line in most agentic commerce decks: "and Rufus."

Two problems with that.

First, Amazon is not one surface. It is a family of four merchant-facing surfaces (Rufus, Alexa+, Buy for Me, Amazon Business), each with its own ranking model, shopper segment, and monetization dynamic. Treating them as one is like treating "Google" as one channel — Search, Shopping, YouTube, and Maps are all Google but they behave nothing alike.

Second, Amazon is where most transactional agentic behavior already lives. Rufus handles more shopping conversations per day than any external agent because Amazon's app is where shoppers already are. Brands that under-index on Amazon's agentic surfaces are ceding the current volume in service of a future volume.

This post is the complete map. If you sell anywhere Amazon reaches — and for most DTC brands that is anywhere with an Amazon storefront, plus Amazon Business, plus Buy for Me participation — the map matters.

The four merchant-facing surfaces

Rufus — the in-app shopping AI

• What it is: A conversational shopping assistant inside the Amazon Shopping app and on Amazon.com. Answers product questions, compares SKUs, generates recommendations.
• Where it runs: Amazon app (iOS, Android), Amazon.com web (partial rollout), select Fire devices.
• Availability: US, UK, DE, ES, FR, IT, IN, JP (through H1 2026).
• What it reads: Your listing (title, bullets, description, A+ content), your product's structured attributes, the customer Q&A section, all reviews, and — increasingly — external content Amazon has indexed to enrich context.
• What it does not do: Send traffic off Amazon. Rufus is closed-loop. The one exception is Buy for Me (see below).
• Ranking model observed: Query-intent match, listing content depth, customer Q&A depth (heavily), review count and sentiment (recent-weighted), price competitiveness, Prime eligibility. Sponsored Products remain visible but blend into recommendation copy differently than legacy Sponsored placements.
• Attribution surface for sellers: Thin. Amazon reports Rufus-adjacent Sponsored impressions in the Ads Console but no direct per-SKU Rufus recommendation impressions to Vendor/Seller Central.
• Optimization priority: Q&A block depth, structured attribute completeness, recent review velocity. Detailed playbook in How Rufus Actually Ranks Products.

Alexa+ — the household agent

• What it is: Amazon's rebooted, agentic Alexa. Available as a subscription ($19.99/mo; free for Prime members) — a first for a household voice assistant.
• Launched: Announced Feb 2025, general rollout through 2025-2026.
• Powered by: Amazon Nova + Anthropic Claude (Amazon publicly partnered with Anthropic for Alexa+ reasoning).
• Shopping behavior: Voice-first discovery. Household categories dominate (groceries, cleaning, personal care, replenishment). Ties into Amazon Fresh, Whole Foods, and third-party retailers via Buy for Me.
• Ranking dynamic that matters: Alexa+ almost always names one product in its answer. There's no visual carousel of three cards to rank in. Top-1 wins the transaction; Top-2 and below are silent.
• What this means for brands: The bar for winning Alexa+ is meaningfully higher than the bar for winning Rufus. The signals that matter are: outcome-focused positioning ("best for babies with sensitive skin"), review depth for the exact use case being asked about, and being the safe / default choice when a shopper hasn't specified brand preference.
• Subscriber count: ~15M as of Q2 2026 (Amazon disclosed range), growing rapidly. Enough scale to matter for repeat-purchase and household consumables today.

Buy for Me — the outbound bridge

• What it is: A feature inside Rufus where the agent can complete a purchase on an external retailer's site when the product isn't sold on Amazon. Amazon negotiates the checkout on the shopper's behalf using their Amazon credentials.
• Launched: Late 2024. 200+ retailers at launch; 500+ retailers as of Q1 2026.
• Retailer set: Includes Coach, GoPro, Bespoke Post, various fashion + specialty DTC brands. Steadily expanding to categories underrepresented on Amazon's own catalog.
• The strategic trade-off: Brand joins Buy for Me → gains access to Amazon's shopper base without listing on Amazon. Loses direct customer relationship (Amazon holds the customer email, order data, and future re-marketing rights).
• When to opt in: If your margin can absorb Amazon's take (~10-15% observed range) and you value reach over relationship. Categories where this makes sense: high-consideration one-time purchases (luggage, mattresses), gift purchases, discovery-heavy categories.
• When to opt out: If you have strong repeat-purchase dynamics, subscription models, or margin fragility. Every recurring transaction Amazon captures via Buy for Me is a customer relationship you've quietly outsourced.

Amazon Business + Amazon Q — the B2B agentic layer

• What it is: Amazon Business is a $50B+ GMV B2B marketplace. Amazon Q Business is the AI assistant inside procurement workflows — surfaces vendor recommendations, negotiates volume pricing, orchestrates multi-vendor orders.
• Why it matters: B2B agentic commerce is roughly two years behind consumer agentic commerce in optimization maturity. The competitive intensity for the top-3 Amazon Q Business slot in most categories is a fraction of what it is for consumer Rufus.
• Signal for brands with B2B-relevant SKUs (office, professional-grade, bulk pack sizes, industrial): This is the largest under-covered opportunity in agentic commerce today. Volume-pricing tiers, MOQ clarity, NET-30 terms, professional-tier warranty — all become ranking signals in Amazon Q Business queries.

The layer nobody thinks about: Amazon Nova

Amazon Nova is the foundation-model family Amazon rolled out in November 2024. The variants:

• Nova Pro — reasoning-heavy tasks. Powers Rufus's complex product comparisons.
• Nova Lite — fast conversation. Powers standard Rufus queries and Alexa+'s conversational layer.
• Nova Micro — edge deployment.
• Nova Canvas — image generation. Powers auto-generated Amazon Ads creative.
• Nova Reel — video generation. Same.
• Nova Sonic — voice generation. Powers Alexa+'s speaking voice.

Brands don't integrate with Nova. But Nova's behavior determines Rufus and Alexa+ ranking outcomes. Two things to know:

• Nova is trained heavily on Amazon's own product catalog. This means Nova has a "canonical" understanding of what constitutes a good SKU description, and listings that align with that canonical structure rank better.
• Nova's factual grounding is stronger than most public models for product data. It rarely hallucinates product attributes. Which means, unlike ChatGPT, Rufus is unusually good at rejecting listings with mismatched or misleading claims — a listing that says "waterproof" when reviews say "leaks in rain" gets deprioritized.

Practically: listings need to be factually consistent with reviews. Marketing copy that overpromises quietly costs you Rufus rank.

Traffic flow — what happens after a shopper asks a question

The four surfaces route traffic very differently. This matters for attribution and for margin.

Two things stand out:

1. Only Buy for Me lets the external merchant keep the customer relationship. Everything else is Amazon-mediated. If your strategic priority is customer relationship (repeat purchase, subscription, LTV expansion), Amazon's surfaces are a tax on that priority.
2. Attribution is thinnest at Rufus and Alexa+. Amazon reports very little Rufus-specific data to sellers, and Alexa+ voice queries leave no visual trace. Attribution requires proxy measurement (see section 5).

What you can measure — and what you can't

Amazon's attribution surface for agentic commerce is materially weaker than what's available for ChatGPT or Gemini. Here's what we've mapped.

What sellers can see today:

• Rufus-adjacent Sponsored placement impressions — a new placement type in the Amazon Ads Console. Not equivalent to organic Rufus impressions, but a proxy.
• Detail page views with "AI Shopping Assistant" filter — recently added in some Vendor Central dashboards, partial availability.
• Amazon Marketing Cloud (AMC) attribution paths — for advertisers with AMC access, some Rufus-adjacent path attribution is possible (requires DSP-level spend to unlock AMC).

What sellers cannot see:

• Organic Rufus impressions per SKU.
• Rufus recommendation position (were you Top-1? Top-3? Not shown?).
• Alexa+ voice-shopping impressions.
• Buy for Me referral source (external retailers see the order but not the fact that Rufus recommended it).

Our workaround: rolling parallel-query testing. We maintain a corpus of ~100 seed queries per customer per month across their category. Each month we run those queries against Rufus manually and log which SKUs appear in Top-3 and Top-1 recommendations. Compared week-over-week, this is a stable, though imperfect, proxy for Rufus visibility movement. Not sufficient for financial attribution but sufficient for tactical optimization.

If you're serious about Amazon agentic commerce, this parallel-testing infrastructure is a required build (or a required vendor capability).

Recent Amazon moves (2025 → H1 2026)

The chronology matters because merchant strategy has to react to it.

Two observations. First, the cadence has accelerated — six meaningful moves in the last 18 months versus three in the two years prior. Second, Amazon is cautiously opening its wall (UCP endorsement, Buy for Me expansion) while still preferring closed-loop. Both directions matter for brand strategy.

The decision framework — Amazon as channel to double down or route around

Every brand we work with faces the same strategic question. There is no universal right answer; there is a category-specific right answer.

Double down on Amazon if:

• Your category has high Amazon-native shopper intent (majority of category-related searches begin on Amazon).
• Your margin structure can absorb Amazon's take without pain.
• Your product depends on reviews and social proof (Amazon's review depth is unbeatable).
• You are OK with Amazon holding the customer relationship.

Route around Amazon (invest asymmetrically off-Amazon) if:

• Your category has strong DTC / brand.com precedent (specialty DTC categories).
• Your margin structure requires holding onto the direct customer relationship.
• Your product benefits from brand storytelling and content that Amazon flattens.
• You have subscription / repeat-purchase dynamics you want to protect.

Hybrid (participate on Amazon but with a boundary):

• List on Amazon for discovery, but drive high-intent shoppers to brand.com via off-Amazon marketing.
• Opt out of Buy for Me but stay listed on Amazon.
• Use Amazon for one product line (introductory / gateway) and hold premium SKUs on brand.com only.

Amazon India did the double-down math. Their +0.98% Share of Voice move in 6 weeks required intensive listing + Q&A investment specifically because they concluded Amazon was the majority discovery surface for their categories.

What to do this quarter

For a brand that hasn't yet built an explicit Amazon agentic strategy:

1. Audit visibility on Rufus for your top-20 SKUs and top-50 shopping queries in your category. Use the parallel-query approach described in section 5, or run our free Citation Rank scan — the scan includes an Amazon Rufus dimension.
2. Deepen the Q&A section on your top-20 SKUs. Target 15-25 Q&As per SKU. This is the single-highest ROI move in Amazon agentic commerce today.
3. Reallocate 20-30% of Sponsored Products budget to Rufus-adjacent placements for one category as a test. Measure over 4-6 weeks.
4. Make the Buy for Me decision consciously. Opt in for one product line, hold out others, or opt out entirely — but don't drift.
5. If your SKU mix has B2B relevance, claim your Amazon Business presence and optimize for Amazon Q Business queries. This is the least-contested slot in agentic commerce today.
6. Track Alexa+ visibility for household-consumable SKUs. Even at 15M subs, the top-1-winner dynamic makes this asymmetrically valuable for brands that can win the "safe default" slot.

Beyond that: track the surface as a full category, not as a single Rufus lever. Rebalance quarterly as Amazon evolves.

CTA

If you want to see your baseline visibility across Amazon Rufus, ChatGPT, Gemini, Perplexity, Copilot, and Claude — plus a category-specific recommendation for whether Amazon is a channel to double down on or route around for your brand — start with a free Citation Rank scan. 24-hour turnaround. No credit card.

If you're ready to close the loop across all six surfaces — Rufus included, with our parallel-query attribution stack — book a demo. Pricing is on the public tiers page.

Amazon is not the future of agentic commerce. It is the present of agentic commerce, running at scale, in production, right now. Under-covering it is a strategic error most decks are quietly making.