Your Products Have Been Defaulted to Meta Shopping Assistant Without Your Consent

On September 8, Meta’s personal AI assistant, Muse, launched in the United States, available through a standalone app, web, and WhatsApp. It can send emails, book flights, and also shop on behalf of users. On the same day, Shopify updated its changelog: Meta has entered Agentic Storefronts, becoming an AI channel. The crucial part wasn’t mentioned in the title—starting from this moment, your products are shared with Meta by default, without needing to click any buttons.

Most merchants haven’t even opened that page yet. This article breaks down what is shared, where the switches are, and what remains if you turn it off.

Two Events on the Same Day

Muse connects to email, calendar, payment, and health applications, running errands for users, with shopping being one of the tasks, requiring user approval for each expense in chat. Initially available in the U.S., it can be activated on iOS, Android, web, and WhatsApp.

Shopify’s side is worth pondering. Agentic Storefronts launched in March this year, with a dedicated page appearing in May, already featuring three AI channels: ChatGPT, Google AI Mode, and Gemini, Microsoft Copilot, with Meta being the fourth. Shopify President Harley Finkelstein stated plainly on X: “Muse didn’t exist yesterday. Today our merchants can already sell through it.”

The pipeline was laid out in advance.

In contrast to ChatGPT Ads: merchants must install the app, link accounts, and authorize synchronization for their products to enter OpenAI’s feed. Meta’s path is open by default, which is the most tangible difference between the two channels.

What Muse Can Read from Your Catalog

Shopify has officially listed the fields: title, description, price, and full inventory are all readable, along with variant options and images, totaling six categories, with inventory and price kept up to date across channels. For instance, a standalone site selling camping gear with 400 SKUs, titles, main images, variants, and stock status can be accessed by Muse at any time.

The purchasing process works like this: a user in WhatsApp expresses interest in a camping lamp that lasts for two hours, Muse searches the Shopify Catalog for the best match, provides the total price, and waits for user confirmation. If the merchant has direct checkout enabled, payment occurs without leaving Meta, and the order returns to the backend, with the channel marked as Meta, attributed like other channels. The checkout process runs on UCP, Universal Commerce Protocol, an open standard jointly released by Google and Shopify on January 11 this year, managing the entire action from product search to checkout. American Banker describes it as: Muse finds the most suitable product in Shopify’s catalog and completes the checkout via UCP. Payments go through Stripe’s Link, generating a one-time usable card number, and merchants do not receive the user’s real card number.

The progress of Shop Pay has conflicting reports: Meta’s announcement states coming soon, while Shopify’s Vanessa Lee claims it can be used in Muse now.

Specific Locations of the Two Switches

Path: Shopify backend, Sales channels, Agentic. First, you will see “Allow Shopify to manage for me,” which is enabled by default, indicating that the AI channel is managed by Shopify on your behalf. If you want to change it manually, turn it off first, then click on Meta, where there are two switches:

• Shopify Catalog access: products shared with Meta, enabled by default.
• Direct checkout: U.S. customers can pay directly on Meta without leaving, enabled by default for qualifying stores.

The third location is Auto enroll new storefronts. Turn it off, and Shopify will not automatically include you when connecting to new channels in the future; this is the only method specified in the documentation.

Direct checkout is not a complete checkout: subscription products cannot enter, bundled and customizable products are excluded, B2B exclusive products are blocked, checkout blocks will cause Meta to render the checkout page in its own way, and local delivery and pickup are not supported. Google Analytics and custom pixels do not run, the storefront does not load at all, only two server-to-server events are returned at the start and completion of checkout.

For example, a standalone site selling pour-over coffee equipment with around 100 SKUs, including a subscription plan for coffee beans. A U.S. customer buying a set of filter cup gift boxes in Muse can complete the transaction, with the order marked as Meta; subscription orders cannot enter, and such customers are redirected back to your own store. If the measurement relies on browser pixels, this part of the data will be missing.

Another pitfall: if colors, materials, and sizes are only written in the description paragraphs, the Catalog cannot capture structured values. Descriptions are meant for users browsing the page, while the agent’s filtering logic only recognizes fields.

Residuals After Turning Off

Shopify’s help documentation states: even if you exit Shopify Catalog access, your products may still be crawled and indexed like in traditional search engines, appearing in AI channels.

The crawler route cannot be controlled by the official switch.

Craftshift’s merchant guide states more directly: there is no global button to completely turn off Agentic Storefronts, nor is there a product-level opt-out; to hide a single product, you can only exclude it by tag or collection and manually mark it as unavailable in the Catalog. I searched for a single switch for merchants to “completely disappear from Meta,” but found none in the official documentation.

Enabled by default, hard to exit, and even if exited, not completely clean—this is the core of the controversy from the merchant side.

Meta’s Security Mechanism Explained in Detail, Critics Unconvinced

Consumer-side criticism is even louder. Muse needs to connect to email, calendar, and payment; TechCrunch pointed out a timeline: on August 26, Meta reached a $18 billion settlement regarding a lawsuit over social media harm to minors in 29 states, and two weeks later, it launched an agent that needs to read your email and calendar. This comparison is the main ammunition for critics.

Meta’s response is found in a security blog, authored by Tarek Sheasha, VP of Superintelligence Labs: each user has an independent virtual machine, the agent runs in an isolated container, unable to access real credentials, and external interactions go through a Sentinel check layer that the agent cannot cover; the highest bug bounty is $300,000, with the maximum reward for a successful single-user prompt injection case being $130,000. The official stance also includes: Muse conversations and virtual machine data do not enter the advertising system and can opt out of model training.

Merchant-side concerns revolve around sorting: which products are pushed to users by Muse, how the weight is calculated, whether it will be linked to advertising, and the official documentation does not clarify this nor provide specific values. Product data itself is public and visible to anyone, but the allocation rules are a black box, and this boundary is currently managed by Meta itself.

Where to Shift Advertising Budgets

Muse is only available in the U.S., and no one has data on the quality of third-party channel orders; there is currently no basis for a major shift in advertising.

This week, two things can be done. First, open the backend Agentic page to confirm whether products are indeed being shared by default, which is highly likely. Second, turn off Auto enroll, so when new channels come in, you can evaluate them yourself. Catalog access and direct checkout can be left on for two weeks to observe the conversion and returns of orders marked as Meta before deciding whether to keep them.

What is crucial to remember is: the way to access AI channels has shifted from “merchant authorization” to “platform decision-making for you.” At least ChatGPT retains that authorization door, while Meta has not set one.