On 6 November 2026 Amazon opened its shopping agent, Alexa for Shopping, to text messages. A customer in the United States links a phone number, messages a contact called Alexa Shopping, and asks for what they need. The agent searches, compares, texts back suggestions, adds items to the cart and checks out when told to. It started invite-only with a small group, with a wider rollout planned over the coming weeks.
The replies carry ads. When Alexa suggests options, one Sponsored Products ad can sit alongside the organic picks. Advertisers pay nothing while it is in testing, and Amazon keeps ads out of orders the agent places automatically. Amazon says more than 400 million customers used Alexa for Shopping in the past 12 months, and that use of its agentic features has grown eight times since January.
The other half of the story came earlier. Amazon blocked Meta’s shopping agent, Muse, citing unauthorised access to its platform, improper identification, and capturing customer data without permission.
What changed, and what didn’t
The excitement around this is that shopping by text with an agent is the new front door to buying. So I ask the usual question first. Is there a need? Reordering by text, building a list from a photo of a recipe, finding a gift before a deadline. These are real chores, and a text thread is a sensible place to do them. The need is there.
What changed is where the recommendation happens and how many of them there are. A results page shows dozens of listings. A text reply shows a few, and one of them is for sale. What didn’t change is who the agent works for. It is Amazon’s agent, on Amazon’s catalogue, carrying Amazon’s ads, closing in Amazon’s checkout. The buying still happens where it always did. The agent just moved the shelf into a chat.
The split I expected is now in plain view
I have said agentic commerce is 50/50 and too early to call. Some platforms will open up and welcome agents. Others will restrict access to keep their hold on the shopper. Amazon has now done both things within the same stretch: it blocked an outside agent, then shipped its own. That is the closed half taking shape.
Inside a walled platform, the agent works for the landlord. The recommendation slot is a placement, and placements get sold. The independent ad tech consultant quoted in the piece, Jonathan D’Sousa-Rauto, called it what it is: “Any new ad placement is generally seen as an interesting test.” Advertisers see a new slot to buy. Amazon sees a new thread to monetise. Neither of them is wrong.
The giants can afford to wait this out or shape how it turns out. A smaller seller can’t do either, and the time to decide where to put the effort is now.
What it means if you sell there
Take a kitchenware brand that sells on Amazon and on its own site. A customer texts Alexa for a pan that works on induction, goes in the oven, costs under sixty dollars and arrives by Friday. A few suggestions come back. One is paid. The rest are the agent’s reading of the catalogue.
The brand can buy the paid slot, and that is a normal ad decision with a normal margin question behind it. The unpaid mentions are different. There is no bid for those. The agent picks the listing it understands most clearly and can match most precisely to the way that customer phrased the question. That is the same thing I see in AI search every week. The AI doesn’t pick the biggest brand. It picks the one it can read without guessing.
And everyone asks slightly differently. Induction, oven-safe, under sixty, by Friday, non-stick, for a small flat. The listing has to carry enough specific, measurable attributes to catch each version of the question, in the words buyers actually use. Adjectives get dropped. Numbers and facts survive. For a smaller business that part of the shift is already decided, whichever way the platforms go: get data-centric.
The second thing the Amazon thread takes from you is the conversation itself. You don’t see it, you can’t follow it up, and you can’t turn it into the next order. The only channels where the conversation is yours are the ones you run: your WhatsApp number, your site chat, your Instagram inbox. That is the half I build for with AeroChat, where the customer asks in the channel they already use and the request becomes an action in the seller’s own systems, from an order status check to a reorder.
What a seller should do
Treat Amazon’s text thread as a rented shelf. Test the paid slot while it is free, and keep buying it only where the margin holds, the same way you would judge any other ad.
Put the real work into the listing data. Fill every attribute, make them specific and measurable, and write them in the phrasing customers use when they ask. Cover the long tail of ways the same question gets asked, because that is what the unpaid mention is decided on.
Keep one channel you own and make it do the work: orders, reorders and questions handled there, with the conversation and the customer record staying with you.
Watch which way the other platforms go before betting the whole plan on either half. The open half is still coming, and the data work above serves both.
Source: Digiday: Amazon’s shopping agent now takes orders by text, with Sponsored Product ads in the replies