AI shopping agents are getting blocked: is your website turning customers away?

Reading time: 5 minutes · For owners of SMBs with an online shop or online booking in Ausserschwyz, the Linth region and around Lake Zurich

You would never turn someone away at the shop door just because they’re shopping on behalf of a customer. On many websites, that is exactly what’s happening right now, and usually the business has no idea. The customer sends her AI assistant off to order something or book an appointment. The website takes it for a robot and won’t let it in. The purchase then happens somewhere else.

Meta summed it up this month, quoted by TechCrunch: turning away a personal agent means turning away the customer behind it. I explain what this means for your website in the video:

Why this is happening now

AI assistants that shop and book for their users are now arriving in everyday life. Meta’s Muse is one of the first for a broad audience; others will follow. At the same time, websites are getting worse at telling whether a human or a machine is in front of them. I’ve already described how Amazon deliberately locks such agents out. What’s new is that businesses that don’t want to are turning them away too.

Walmart provides the clearest example. The retailer works with Meta so that Muse can shop there. Even so, users reported that orders failed. Walmart says this wasn’t intended. The typical stumbling block, according to TechCrunch: the button you’re supposed to click to confirm you’re human. It’s no different for an SMB. Bot protection at the hosting provider, a security plugin in the shop, a captcha on the contact form, switched on after a wave of spam and forgotten ever since.

We’ve been here before. Anyone who didn’t have a mobile website a good ten years ago lost sales. In April 2015 Google started favouring mobile-friendly pages in mobile search. I’m convinced: if you don’t build your website today so that an agent can find the data it needs, order and leave again, you’ll miss out on a lot of sales. Over the next twelve to 24 months this will become decisive.

What to do

There’s no finished standard yet. That’s no reason to wait. Five steps I’ve been working on with clients for about a year:

1. Try it yourself

Give an AI assistant with a browser function a real task on your website: find a specific product, put it in the basket, request an appointment. Where does it get stuck? At a captcha, a pop-up, a page that only works by clicking on images? Much of what helps the agent also makes the site easier for people.

2. Set your bot protection deliberately

Check with your web developer or hosting provider which bot protection is active and what it blocks. One distinction helps: you’re welcome to lock out crawlers that collect content for training AI models. An agent that wants to order for a customer is something else. How to separate the two cleanly is now part of being found in search engines and AI assistants.

3. Structured data and few clicks

An agent doesn’t read mood boards, it reads data: price, availability, delivery time, delivery area, opening hours, free appointments. This information belongs on the page as structured data, not just in an image or a PDF. And the product mustn’t sit ten or twenty clicks deep. A flat structure where everything is openly accessible helps people just as much.

4. Make the agent’s job easy

The less computing effort an agent needs to complete a purchase with you, the more likely it is to choose your products and services. It’s a technical topic, but above all it’s common sense: ask your tech team, or ask Claude or ChatGPT directly how Shopify or large retailers solve it. At the end of September Shopify opened its checkout to AI agents.

5. Serve two audiences

Images, stories, inspiring content: all of that stays important, for people. Alongside it you need a second, plain layer for everyone who simply wants to buy, whether themselves or through an agent. If you only believe in the brand story, you forget the customers who already know what they want. In many SMBs exactly this work gets left undone because the website runs on the side. That’s what marketing without your own marketing department is for.

A standard is on its way. Meta and Sierra have introduced the Personal Agent Protocol, an open standard that lets businesses recognise that an agent is acting on behalf of a real customer and define what it is allowed to do. Shopify, Stripe and Walmart are on board as partners, and a first version of the specification has been announced for the end of October.

Where the limits are

If you sell through consultation and quotes, such as natural stone work or built-in furniture, no agent will be ordering from you any time soon. There it’s enough for assistants to find and describe your offer correctly. It’s different as soon as people book or order online, for example at a dental or dental hygiene practice with online appointment booking. And bot protection exists for a reason: spam, load peaks, abuse. It’s not about opening every door, but the right one. Which data an agent may see on behalf of your customers is something I covered in my post on access and trust with AI agents. For the legal side, the checklist on AI and customer data in Switzerland helps.

Frequently asked questions

Why are AI shopping agents being blocked on websites?

Some retailers lock them out deliberately, Amazon for example. Often, though, it’s unintentional: bot protection, captchas and security settings can’t tell an agent from a harmful bot and turn it away, even when it’s shopping for a real customer.

How can I tell whether my website is turning agents away?

The easiest way is to test it yourself: give an AI assistant with a browser function a concrete task on your website and see where it gets stuck. Then check with your web developer or hosting provider which bot protection is active.

Do I have to allow all bots for this?

No. You can keep locking out crawlers that collect content for training AI models. What matters is that agents ordering or booking on behalf of your customers don’t fail at the same protection.

What is the Personal Agent Protocol?

An open standard being developed by Meta and Sierra, with partners such as Shopify, Stripe and Walmart. It is meant to let businesses recognise an agent acting on behalf of a real customer and define what it is allowed to do. A first version of the specification has been announced for October 2026.