Consumer AI Agents Are Standing Between the Brand and the Buyer

Amazon, in a statement to GeekWire, objects on specific grounds: Meta did not notify it about Muse's access to the store, the agent does not identify itself while browsing, and apparently stores customers' credentials. Amazon can afford a wall: over $68 billion in ad revenue last year and its own Buy for Me agent with an opt-out for brands. Most companies have neither. A typical scenario is Target, where a CNN correspondent was able to browse the site through Muse, but had to enter personal details manually: the retailer blocked autonomous clicks at the checkout stage. Often such a rule is set by the security team as an anti-bot setting, and marketing only learns about it from a drop in conversion.

Is there a middle path?

The legal landscape has shifted in favor of platforms: in August, the US Ninth Circuit Court of Appeals, in the case of Amazon v. Perplexity, ruled that the "user" under the federal anti-hacking law is the one who visits the site, and the AI company acts on their behalf. However, terms of use remain a tool for restricting access.

A middle path has emerged. The Trusted Agent Protocol from Visa (together with Cloudflare) allows merchants to verify signed agents and distinguish them from malicious bots. Mastercard requires registration and verification of agents before transactions, and all European Mastercard issuers are already connected to Agent Pay. Amazon has said that third-party agents "must act openly" — and the payment networks have provided a way to demand exactly that without closing the door.

Simple arithmetic explains what blocking puts at stake: a customer who is turned away takes their agent — and everything it knows about them — to the retailer that let them in.

How should a brand prepare?

• Check your default settings: Shopify feeds products to Meta automatically via Shopify Catalog, and direct payment is enabled for eligible stores.
• Decide how far down the funnel to let verified agents go: browsing, cart, checkout — using Trusted Agent Protocol or Agent Pay signals instead of blanket blocking.
• Make your return, delivery, and warranty policies consistent across all channels and format them as machine-readable structured data.
• Make your loyalty program portable: the agent passes the loyalty ID, and bonuses are applied at payment.
• Rebuild your analytics around server-side events — before you have to explain an unexpected hole in your fourth-quarter reporting.

Consistency is no longer cosmetic. According to Accenture ("Talk to My AI Agent"), 56% of consumers are willing to tell an agent which brands to consider, but 37% of loyal customers would accept a brand substitution if the alternative is a better fit. An agent will pull up the policies of four competitors in seconds, notice that return windows differ across channels, and rank higher the one whose rules are uniform. The brand will see neither a rejection nor an abandoned cart. The sale will simply walk away.

The verdict on the situation is sober: 72% of merchants in the UK and the US, according to Checkout.com research, agree that consumers will master agentic shopping faster than retailers can prepare for it. A Deloitte survey of 13,500 Europeans shows 31% ready to trust AI with the final purchase and 47% who refused. Anyone who takes customers' current caution as a constant for years to come will repeat the story of the mobile cycle: shoppers walked around with smartphones for years before most brands had a checkout that worked on them.