AI agents are already shopping on e-shops: Meta Muse makes up 72% of their requests

AI agents are already shopping on e-shops: Meta Muse makes up 72% of their requests

Peter Gáborík
Published
5 min read
AI Agent
Share this article:Add on Google

E-shops no longer get visits only from people and search crawlers. More and more AI agents search for products, compare prices and complete purchases on a user's behalf. Since September the biggest of them is Meta's Muse. In your analytics such a visit can look like a customer when it is not one.

What HUMAN found

Security company HUMAN, which tracks automated traffic on websites, published its first data on Muse on 28 September 2026. According to its measurements:

  • Muse accounts for an average of 72% of daily AI agent requests to websites.
  • Its daily volume is 5.2 times that of the next largest agent, ChatGPT Agent.
  • 84% of Muse requests target product and search pages.
  • In a recent sample, Muse was 25% more likely to reach checkout pages than other AI agents.
  • Muse reached in 2 days a request volume that took ChatGPT 11 months.

Read the data with care. PPC Land notes that HUMAN did not publish absolute counts, sample size or a definition of a request. HUMAN also sells the agent detection product it recommends.

What Meta Muse is

Muse is a personal AI agent Meta launched in the US on 8 September 2026. According to PPC Land, it handles tasks across email, calendar, payments, bookings and shopping. It runs in iOS and Android apps, on muse.ai and in WhatsApp. The basic tier is free.

Meta says Muse runs on a secure virtual machine with its own browser, and a monitoring agent called Sentinel approves everything it sends to the internet. Payments go through Link by Stripe. For sensitive steps such as signing in, booking or buying, Muse asks the user for explicit approval.

Its strength is distribution. HUMAN, citing Apptopia, says 95% of Muse users also use Facebook and 63% use Instagram. No other agent has that kind of reach among people who already shop through social media.

Amazon blocked the agent

On 20 September Amazon stopped Muse from shopping on Amazon.com. Users see a message saying that access by an unauthorised AI agent violates the conditions of use. According to PPC Land, Amazon says it never agreed to Muse operating in its store, the agent does not identify itself while browsing, and it appears to store customer credentials. Meta says Muse has no visibility into passwords or payment methods.

The dispute shows that every store will have to decide whether to let agents in, on what terms and how to recognise them.

Meta is drafting rules for agents

On 6 October Meta previewed the Personal Agent Protocol, an open standard it is developing with Sierra and partners such as Shopify, Stripe and Walmart. It is meant to handle authentication, secure passing of user identity, and visibility into what personal agents do on company websites and APIs. According to Social Media Today, anyone will be able to implement it.

Businesses can already offer Muse a connector with direct access to their API through the muse.ai platform. Other features, such as steering agents toward preferred paths through a site, are still only planned.

What it means for e-shops in Central Europe

Muse is available only in the US for now. Agents such as ChatGPT Agent already browse European websites, though, and Meta has Facebook and Instagram users in every country. For an e-shop in Slovakia or Czechia this has three practical effects:

  • Distorted analytics. Agent visits can show up as regular sessions. Your conversion rate drops, funnel and attribution shift, and you end up optimising on the wrong numbers.
  • Fraud risk. An agent that does not identify itself is hard to tell apart from a malicious bot. Blocking everything, however, means losing purchases an agent makes for a real customer.
  • A new shop window. An agent does not see banners or design. It reads product pages, prices, availability and specifications. If yours are incomplete or unclear, you lose the comparison.

What to do now

  1. 1Check your server logs or bot protection tool to see how much traffic comes from AI agents and which pages it hits.
  2. 2Decide what agents may do. Browsing products is usually fine, while login, cart and payment are where you want control.
  3. 3Separate agent traffic in analytics so it does not skew your conversion rate and campaign reporting.
  4. 4Review product data: full names, specifications, prices including VAT, availability and structured data. It helps agents and search engines alike.
  5. 5Follow how the Personal Agent Protocol develops. If your platform (Shopify, for example) adds support, switching it on will be easier.

If you are wondering whether your platform can handle changes like this, read how to choose the right e-commerce solution.

Want an e-shop ready for customers and their agents? See how we build e-shops and book a consultation.

Frequently asked questions

What is an AI agent on an e-shop?
A program that browses a website for a user, searches and compares products and can complete a purchase. Unlike a search crawler, it acts on behalf of a specific person.
Is Meta Muse available in Europe?
Not yet. Meta launched it on 8 September 2026 in the US only. Other agents, such as ChatGPT Agent, already browse European websites.
Should I block AI agents on my e-shop?
We do not recommend blanket blocking, since an agent may be buying for a real customer. It is better to recognise agents, allow product browsing and keep sensitive steps such as login and payment under control.
How will agents affect my analytics?
If you do not separate their visits, they count as regular sessions. Conversion rate, funnel and campaign reporting then show distorted numbers.
Peter Gáborík

Written by

Peter Gáborík

Project Manager – AI, Automation & SEO

Project manager and founder of Weboptim. Builds growth strategies around SEO, automation and AI; in online marketing since 2017.

Related service

E-Commerce

View service

Tips that save time – once a month

One e-mail a month: a new guide, the automation of the month and one number from practice. No spam.