Home & Living · AI

Amazon locks out AI shopping agents: what it means for Swiss SMBs

Wohnzimmer mit Smartphone auf dem Sofa

Reading time: 6 minutes · For owners of SMBs with an online shop or product catalogue in Ausserschwyz, the Linthgebiet and around Lake Zurich

Anyone who asked Meta’s new AI assistant Muse to order something on Amazon over the weekend of 20 September got a warning instead of an order confirmation: access by an unauthorised AI agent violates the terms of use. Amazon justifies the block with security. The agent does not identify itself as AI, never asked for permission and apparently handles customers’ login details, as GeekWire first reported.

These concerns are not made up. In my view, though, the real reason lies elsewhere, and it affects not only Amazon but every business that shows products online. I recorded my take on video this week:

Amazon Blocked Meta’s AI Shopper. It’s Not About Security. (6 min)

An agent does not browse

An AI agent does the shopping the way many customers would like: it finds the cheapest matching product, buys it and is gone. No scrolling, no “Frequently bought together”, no sponsored ad in second place, no spontaneous extra item in the basket.

That is exactly what a marketplace lives on. Amazon’s advertising business alone brought in more than 68 billion dollars last year, and every one of those dollars assumes that a human looks at a page. From Amazon’s point of view, an efficient agent is a customer who sees no ads and adds nothing to the basket. That is not a security problem; it is a business model problem.

Amazon has nothing against shopping agents. Amazon has something against shopping agents that Amazon does not own.

You can see this in the fact that Amazon runs such an agent itself: in May its assistant Rufus became Alexa for Shopping, which researches, recommends and, if you want, buys right away, all within Amazon. Inside its own walls, “buy this for me” is welcome.

Why this reaches the region too

You could dismiss this as a quarrel between tech giants. The Swiss figures say otherwise. According to the IGEM-Digimonitor 2026, 80 per cent of respondents use AI services, two thirds of the population use ChatGPT at least occasionally, and almost 40 per cent take advice from AI when shopping. At the same time, people in Switzerland spent CHF 15.8 billion online in 2025, and almost half of the 30 largest online retailers are marketplaces.

For a specialist retailer in Lachen, a manufacturer in the Linth plain or a furniture shop on Lake Zurich, this means: a growing share of customers asks an assistant first and only then, if at all, visits a website. If you do not show up as an answer there, you drop off the shortlist before anyone has seen your site. It is the same shift I described for local search in the post Pain points instead of keywords, just one step further.

What to do

1. Check what an assistant says about you today

Ask ChatGPT and Gemini about your product in your region, the way a customer would. Are you mentioned? Are prices, range and opening hours correct? It takes ten minutes and shows you faster than any audit where you stand.

2. Maintain product data so that a machine understands it

What do you sell, at what price, and when can you deliver? These three facts need to be on the website, clean and structured, not in a PDF price list or only in the sales team’s heads. Structured product data is now part of good search engine optimisation for SMBs, because the same data feeds Google and the AI assistants.

3. Decide deliberately which crawlers you let in

Many websites block AI crawlers by default, often without anyone having decided to. But crawlers that train models and crawlers that read your page live for a specific request are two different things. If you block everything, you also disappear from the answers. What you need to consider legally as soon as customer data is involved is summed up in our tool AI and customer data: what applies in Switzerland.

4. Give someone the job instead of letting it run on the side

In most SMBs nobody does marketing; everyone does it on the side. Topics like this get left undone because they do not feel urgent, until the enquiries dry up. I describe how this works without a dedicated department under Marketing without a marketing department, and how I use AI in practice on the page AI for marketing and sales.

Where the limits are

Not every business needs to act now. If you sell mainly through personal advice, such as a trade business that sends quotes after a site visit, you gain more from a good Google Business Profile and genuine reviews than from structured product feeds. And nobody knows today how quickly shopping agents will arrive in everyday life. My estimate is two to three years until a noticeable share of purchases happens this way. That is not a panic deadline, but it is reason enough to get your own data in order now, while the big players are still arguing about terms.

A business in Lachen or Pfäffikon does not need to think like Amazon. It just needs to make sure a machine understands what it offers.

Frequently asked questions

Why did Amazon block Meta’s AI agent Muse?

Officially for security reasons: the agent did not identify itself as AI, did not ask for permission and apparently handled customers’ login details. Economically, more is at stake: an agent buys the cheapest matching product and leaves the site again, with no browsing, no add-on purchases and no ads.

Does this also affect small online shops in Switzerland?

Yes. AI assistants read your shop before a human sees it. If you structure products, prices and delivery times cleanly and make them accessible to the relevant crawlers, you are more likely to be recommended, regardless of size.

Should I block AI crawlers on my website?

Not across the board. Crawlers for model training and crawlers that read pages live for answers and purchases are different things. If you block everything, you also disappear from the answers of the assistants your customers already use.

What should I do first?

Check whether an assistant understands your offer: ask ChatGPT or Gemini about your product in your region and see what comes back. Then maintain product data, prices and delivery information in a structured way.

Want to know what an AI assistant says about your business today?

Book a free 30-minute initial call. Together we look at how ChatGPT and Gemini present your offer today, and what to do first. Prefer to read on your own first? In the tools you will find checklists, no sign-up needed.

Newsletter

One lever every one to two weeks

Short, actionable insights. No spam, unsubscribe at any time.

Conversation

Let's talk directly

30 minutes by phone or video, free and without a pitch. Or send a short note about what you need.

remco@marketing-march.ch · +41 78 951 66 90

Personal reply within 24 hours (working days).