Alexa+ for brands: how voice agent commerce works

Alexa+ answers a shopping question out loud and offers to order. Amazon says more than 350 million customers used Alexa Shopping in the past 12 months. A voice answer has no carousel, so the brand that gets named gets the order and the brand in second place is not heard.

Alexa+ is Amazon's rebuilt assistant, launched 26 February 2025 at $19.99 a month or free for Prime members (Amazon, 2025). For brands it is a spoken front end to Amazon's catalog and, through Buy for Me, to a brand's own site. Amazon publishes no ranking rule; the listing is the lever.

Last updated: 13 October 2026.

What Alexa+ is, in Amazon's words

Amazon introduced Alexa+ on 26 February 2025. "Alexa+ costs $19.99 per month, but all Amazon Prime members get it for free" (Amazon, 2025). On shopping, Amazon says she "can also help you search, find or buy virtually any item online" and calls her "the world's best, most personalized AI assistant for shopping" on Echo devices and, as Alexa for Shopping, on the Amazon site and app (Amazon, 2025).

Two names, one assistant. Rufus, the shopping assistant in the Amazon app, was renamed Alexa for Shopping in 2026 (Retailgentic, 5 August 2026); voice and screen draw on the same catalog. This post covers the voice half; the screen half is in how Rufus ranks products and Amazon Rufus for brands. Where Alexa+ sits among the other assistants is in AI shopping assistants: the 2026 map.

Why voice is a different game from a screen

A screen gives the shopper a shortlist. Alexa for Shopping in the app, ChatGPT's product cards and Gemini's grid each show several products; second place still gets looked at.

A spoken answer does not work that way. In our tests, when a shopper asks Alexa+ for the best option for a stated need, the answer names a single product and offers to add it to the cart or order it. If the shopper declines, they usually revise the request rather than ask what else was considered, and the assistant names a different single product for the revised request.

So the gap between first and second is the order against nothing. You optimise not for page-one presence across many queries but for being the product the assistant will say out loud for the specific outcome the shopper described.

Amazon publishes no ranking rule for Alexa Shopping, by voice or screen; of the platform assistants only OpenAI publishes one (OpenAI Help Center, 2026). Everything below is about inputs you control, not a weighting we can verify.

What Amazon has said about Alexa Shopping

The numbers are Amazon's, from the Q2 2026 earnings call as reported by Retailgentic on 5 August 2026:

Two things follow: the assistant is already a paid surface, and Amazon measures it on order value, which fits an assistant that resolves a need in one answer.

Discovery is part of it too. In Evercore's consumer survey, 57% said Amazon's AI shopping assistant helped them purchase a product they were not previously considering, and 55% use a retailer's built-in assistant against 48% for a general AI tool (Evercore, via Retailgentic, 1 September 2026). On 2 September 2026 Alexa for Shopping added "Update Me When", a standing request that evaluates listings on a cadence while nobody is looking (Scot Wingo, 2 September 2026). Attribute completeness becomes a background job, not a launch-day one.

What the assistant reads

The inputs are the listing: title, bullets, attributes, reviews and the customer Q&A section, the same material the app assistant reads and the only material you control. Three properties matter more when the output is spoken:

• Outcome language. A shopper asks by need ("for sensitive skin", "for calls", "non-toxic"), not by attribute. If the shopper's words are not in your bullets, the assistant has to infer the match.
• Plain, verifiable claims. A certification stated in the shopper's terms (pediatrician-recommended, dermatologist-tested, BPA-free, USDA Organic) is easy to say aloud and reduces the assistant's risk in naming you.
• Fields that filter. Pack size, capacity, material and compatibility need to exist as structured attributes. A constraint the assistant cannot filter on drops you before price is compared. The same lesson came out of a 90-day on-site assistant log, where shoppers asked for TSA locks and litre capacity that were not catalog fields (Provogue case study, 2026).

What does not transfer to voice: A+ imagery, since nothing is rendered, and anything that lives only in a picture.

The voice-first playbook

1. Pick the SKUs a shopper would ask for by outcome. Replenishment, household consumables, personal care, everyday electronics and pantry are where "order me a ..." happens. Apparel and furniture rarely do; cover them on screen surfaces instead.
2. Rewrite bullets in the shopper's words. Lead with the outcome the product is for, then the certification, then the spec. Say "for sensitive skin", not "for delicate epidermis".
3. Fix the Rufus fundamentals. Q&A coverage for the use cases shoppers ask about, review coverage, backend attributes, Prime eligibility; see how Rufus ranks products.
4. Narrow the target if you are small. A smaller brand is more likely to be named for a specific outcome ("fluoride-free toothpaste for sensitive gums") than for the category ("toothpaste"). Own the narrow outcome in Q&A, reviews and off-Amazon mentions.
5. Decide on sponsored prompts deliberately. Amazon is quoting a 48% conversion lift on them (Amazon Q2 2026 call, via Retailgentic). Test them on the outcome queries you cannot win organically yet.

Buy for Me: when the order leaves Amazon

Buy for Me is the one path by which Alexa's assistant reaches a brand that does not sell on Amazon. Amazon completes the purchase "by securely providing the customer's encrypted name, address, and payment details to complete the checkout process on the brand's website", running on Amazon Bedrock with "Amazon Nova and Anthropic's Claude models"; it launched in beta "for a subset of U.S. customers" and "brands have the choice if they want to participate" (About Amazon, 2025).

For the brand, that is an order placed by an agent on your own checkout: a guest order, a payment credential you did not see entered, and a customer whose relationship Amazon holds. Whether to opt in is a margin and relationship question, worked through in Amazon Buy for Me: should your brand opt in?. Whether your checkout can be finished by an agent at all is the same question Muse, Copilot and Google's checkout now ask, covered in AI shopping agents: how they buy, what brands need.

Amazon keeps the reverse door shut. From 20 September 2026 Amazon blocked Meta's Muse from shopping its catalog, citing Conditions of Use updated 14 August 2026 that require agents to identify themselves (GeekWire, 21 September 2026). Alexa+ is the only agent Amazon lets buy on Amazon; the full picture is in Amazon's agentic commerce stack.

What not to do

• Do not optimise high-consideration SKUs for voice. Sofas and running shoes are chosen on a screen.
• Do not overclaim. The assistant repeats your words; a disputed claim is a disputed claim at volume.
• Do not skip the listing fundamentals. Voice is not a separate surface from the app assistant; it is the same program speaking.

Measurement: parallel voice-query testing

Amazon does not report Alexa+ voice impressions to sellers. The workable method is the one we use: a fixed set of outcome-shaped queries for your voice-relevant SKUs, run monthly on an Alexa+ device, logging which product is named. Imperfect, and the only first-party signal available.

For orders that leave Amazon through Buy for Me and land on your site, the attribution problem is the general one: GA4 files agent orders as Direct. Scot Wingo of ReFiBuy coined the term Dark Agentic Commerce Traffic for it. For MyMuse, GA4 showed ₹11.5K/mo of AI revenue against ₹81.2K/mo actual over a 12-month attribution window, 7× hidden (MyMuse case study); the method is in dark agentic commerce traffic.

Where Tru Commerce fits

We have run this on Gray Porter and a group of US DTC brands: we made the product pages and checkout readable and finishable by an agent, then placed a real Muse order on the store. That is the same readiness a Buy for Me order needs on your site, and agentic checkout is the product we run it on. On measurement, MyMuse is the reference: GA4 showed ₹11.5K/mo of AI revenue, the real figure was ₹81.2K/mo, 7× hidden, and the Year 1 net surfaced was +₹29.34L. To see what an agent sees on your store, book a demo.