AI agents in the customer journey: what you can still control

An AI agent has already bought something in place of one of your customers, and you don't know it yet. Muse, Meta's new agent, passed 2.5 million downloads in two weeks: faster than ChatGPT climbing to the top of Apple's App Store.

Does this really change anything for my company, or is it still a big-retailer thing?

It changes things for anyone who sells online or manages a digital customer journey. AI agents are no longer an enterprise tool: they're in consumers' pockets.

Meta's Muse, OpenAI's Dots, Anthropic's Cowork: these tools, known as consumer AI agents, meaning software that acts autonomously on behalf of a person, don't wait for the customer to visit your site. They compare products, read return policies, and complete the purchase on infrastructure the brand never touches.

The Italian e-commerce market is worth €66.6 billion in 2026, up 6% from the previous year, according to the Netcomm B2c eCommerce Observatory at the Politecnico di Milano. A growing share of those purchases is about to flow through agents the brand doesn't control. This also concerns an industrial parts shop, a consulting firm with online quotes, or a network of stores with digital booking.

What exactly happened in just a few weeks?

In less than a month, three major players made opposite decisions about the same AI agent. The market has split in two, and every brand now has to choose a side.

On September 20, 2026, Amazon showed a pop-up to Muse users browsing its store: unauthorized access, violation of the terms of use. Two days later, PayPal announced that its customers can shop and check out through Muse at PayPal merchants worldwide. The day before that, Shopify had enabled Shop Pay for Muse across all its stores, and did so by default, with a manual opt-out for merchants.

According to Shopify, merchants who are enrolled automatically must turn the feature off in their admin settings if they don't want to participate. On September 28, Shopify extended support to browser agents with three new tools that let them read, update and complete transactions, always with the buyer's explicit authorization, as reported by TechCrunch.

What happens to my analytics data when an agent buys instead of the customer?

Agent checkout doesn't fire your tracking pixels. The order arrives, but the behavioral history that led to the purchase disappears.

When Muse completes a purchase on Shopify, Google Analytics and custom pixels don't fire. Shopify passes only server-to-server events for “checkout started” and “checkout completed.” The order shows up as the merchant's, but the path that led the customer to choose that product (the pages visited, the comparisons made, the time spent) never reaches your analytics tools.

If you've built a CDP (Customer Data Platform, a centralized system that collects and unifies customer data) over the years, that infrastructure captures nothing from an agent checkout. It's as if the customer walked in through the back of the store, picked up the product and left without passing the visible checkout counter.

This topic ties into the growing tracking challenges we covered in our article on Google Site Kit and automatic tracking on WordPress: data gaps keep multiplying, and every new channel adds another blind spot.

Is blocking AI agents the right move, or does it backfire?

Blocking agents only keeps the customers who give up their agent on your site. But those who keep using their agent take themselves, and all their data, to a competitor who accepts it.

Amazon can afford to block Muse: it generates over $68 billion in advertising revenue a year from people browsing its pages, and it has its own shopping agent (Buy for Me) that identifies itself and lets brands opt out. Most companies have neither a comparable advertising business nor a proprietary agent to protect.

The logic is simple: a block only works if the customer abandons the agent and goes back to browsing manually. If the customer prefers the agent, they take it to the competitor who accepts it. And they take all the context the agent has stored: preferences, dietary restrictions, sizes, purchase history. Data the brand will never see.

On the legal front, in August 2026 the Ninth Circuit (a US federal court) ruled that the "user" accessing a website is the actual person, and the AI agent acts on their behalf. This limits the ability to use anti-hacking laws to block agents operating on a customer's behalf.

A year ago, AI traffic converted worse. Today it converts 42% better. Blocking this traffic in 2026 means blocking the customers most ready to buy. The full data is available in the Adobe Digital Insights report.

Is there a middle ground between blocking everything and opening everything?

Yes: let verified agents in and block malicious bots. Payment networks have already built the infrastructure to do it.

Visa developed the Trusted Agent Protocol in collaboration with Cloudflare: it lets merchants verify an agent's digital signature and tell it apart from a scraper (a program that copies data without permission). Mastercard requires agents to be registered and verified before they can transact on its network, and every Mastercard issuer in Europe is already enabled for Agent Pay.

In practice: a brand can configure its system to accept verified agents at checkout and block unidentified ones. It's not a binary choice between "open to everything" and "closed to everything."

Three concrete scenarios: what changes for an e-commerce, a manufacturing company, and a professional firm?

The impact varies by industry, but the logic is the same: customers who arrive through an agent have already decided, and your brand must be readable and trustworthy before the agent chooses.

Illustrative examples: numbers are meant to show the reasoning, they are not real customer data.

Which marketing levers does the brand really still control?

Three levers remain in your control: the moment before the agent acts, the quality of the data the agent reads, and the direct relationship with the customer after the purchase.

The AI agent acts on the customer's behalf, but the customer has already formed their preferences before delegating to the agent. If your brand is on the customer's mind when they set up the agent, you've already won half the battle. This is the moment for brand awareness and pre-purchase content: articles, videos and reviews the customer read before opening the agent's app.

The second lever is the quality of your data catalog. The agent reads your site like a machine: if information on price, availability, return policy and technical specs isn't structured and up to date, the agent ignores it or misreads it. Those who have invested in technical SEO and schema markup, as we explain in our article on AI agents in media buying, have a head start.

The third lever is the post-purchase relationship. The order arrives anyway: the customer is yours. Communication after the purchase, such as email, support and loyalty programs, is still the brand's territory, not the agent's.

Why is everyone talking about this now: is the market truly ready, or is it still hype?

The numbers say it's not hype. AI traffic to retail sites is growing at triple-digit rates, and in Italy the AI market reached €1.8 billion in 2025.

According to the Artificial Intelligence Observatory at Politecnico di Milano, the Italian AI market reached €1.8 billion in 2025, up 50% from the previous year. 71% of large Italian companies have launched at least one AI project. SMEs are still behind: fewer than 10% have an active project. But consumer AI agents won't wait for SMEs to be ready: they come from customers, not from companies.

In Europe, the IAB Europe report on the impact of AI on digital advertising (September 2026) shows that 86% of European organizations already use AI for marketing, and more than half of the industry expects agentic ad buying to reach operational scale within a year.

This shift is part of a broader transformation in paid media: anyone running campaigns already has to deal with agents that intermediate purchasing decisions, as we analyzed in our article on how to scale campaigns with AI agents.

What are the most common mistakes companies are already making?

The most common mistake is letting the IT security team make the call on AI agents while marketing has no idea what's going on.

The original MarTech source describes a telling case: Target blocked automated checkout clicks, but that decision probably started as an anti-bot rule from the security team. Marketing found out when the conversion numbers changed. This pattern repeats in many companies.

A second common mistake is thinking the issue only concerns checkout. AI agents collect data throughout the entire browsing journey: if your catalog isn't structured, the agent rules you out at the comparison stage, before it even reaches the cart. The challenge of a machine-readable catalog is the same one we covered when discussing AI agent governance in media buying.

How do you get started: what are the steps to adapt your strategy?

Start with a quick audit of where you stand today: what agents see when they visit your site, and what happens to your data when they complete a purchase.

You don't need to wait for the problem to show up in your numbers. Three initial checks are enough to see where you stand.

How can I tell if my strategy is working with AI agents?

The signals to track are different from the classic ones: looking at total traffic isn't enough, you need to separate AI traffic and measure its quality.

So, is it worth adapting now, or can it wait?

It's worth it if you have an active e-commerce store, an online product catalog, or a digital purchasing process. Every week, a growing share of your highest-intent customers uses an agent. Adapting now costs less than catching up later.

A middle ground: you mainly sell offline but have a digital presence. You can wait a few more months, but use that time to structure your catalog and define your policy: doing it in an emergency costs far more.

Not yet if you sell exclusively through direct offline channels, with no digital catalog and no online purchasing process. The impact is still indirect, but keep an eye on how your competitors are moving.

How we see it at SHM

For us, the arrival of consumer AI agents isn't a threat to the relationship between brand and customer: it's a test of the quality of that relationship. An AI agent chooses based on the data it finds. If your catalog is well structured, your policies are clear, and your brand is already in the customer's mind before they delegate to the agent, you've already done the right work.

We believe AI should be integrated to free companies from routine and give time back — not to replace the ingenuity of the people who build customer relationships. Consumer AI agents shift the battleground, they don't end the game. Those who own the pre-agent moment with quality content, and those who structure their data so machines read it correctly, will gain a real advantage.

At SHM Studio, we work on this front with our clients: from the strategic integration of AI into marketing processes, to SEO optimized to be found by AI agents, to the overhaul of paid campaigns in an ecosystem where traffic is changing nature. Think of a manufacturing company with a complex technical catalog: structuring that data so AI agents can read it is exactly the kind of work we've already done in international SEO, and it's now becoming urgent even for those who have never thought of agents as counterparts.

You can explore the whole topic in our Marketing and performance ADV section or contact us directly to figure out where to start.

You'll find more insights on how AI agents are changing campaigns in our article on agentic buying and automated media buying and in the one on creator marketing, where the fragmentation of attention follows a similar logic.

Frequently Asked Questions

The real competitive advantage isn't blocking AI agents. It's being the brand your customer wants their agent to find.