AI agents face barriers when trying to automate online shopping and bookings

AI assistants promise to save users from browsing websites, comparing products, and filling out forms. To deliver on this promise, they need access to the very pages they aim to replace. However, an increasing number of platforms are blocking their entry.

Major US retailers, airlines, and banks have started restricting access for programs that browse on behalf of third parties. According to The Verge, Delta prevents agents from booking tickets, while JPMorgan Chase and Charles Schwab have confirmed they block automated access to accounts for security reasons.

This resistance highlights a key distinction: demonstrating that an AI can operate a browser is different from allowing it to perform tasks online. Even with user authorization, a website may reject the chosen assistant, as the decision to allow or deny interaction rests with the platform.

Conflict between platforms and AI developers

Amazon made this conflict explicit by demanding that Perplexity stop making purchases through its Comet browser on the platform. In a November 2025 statement, the retailer said that third-party applications must operate transparently and respect service rules about whether or not to participate in such operations.

The company accused Comet of accessing its store without properly identifying itself as an automated agent. Perplexity disputes this interpretation and argues that consumers should be able to choose an assistant to act on their behalf. The dispute has reached the courts, with a lawsuit filed by Amazon against the startup.

In addition to security concerns, control over the customer experience is also at stake. If an agent selects a product and completes a transaction, the platform may lose opportunities to display ads and influence the buyer’s decision. For AI companies, taking on this role means becoming the starting point—and possibly the final destination—of the purchase.

Security as an additional barrier

Not all blocks are specific decisions against AI agents. Websites already use mechanisms to detect automated access, prevent fraud, and protect accounts. A tool that performs a legitimate task may encounter these same barriers.

This explains why a page works when a user browses directly but presents a verification challenge or denies access when visited by an assistant. It also means that an isolated block does not prove that a company has adopted a general ban against AI.

These restrictions add to the technical challenges of making an AI-controlled browser recognize buttons, interpret pages, and execute sequences of actions without meeting the service’s identification and authorization requirements.

Negotiations and selective partnerships

Some companies see agents as a new sales channel. Meta has moved in this direction with Muse, establishing integrations with retailers such as Walmart, Sephora, and Best Buy. These partnerships are part of the company’s expansion into AI services for consumers and businesses.

The landscape is heterogeneous: one store may accept certain partners and reject others, while a financial service imposes its own restrictions on account access. The reach of an assistant now depends on the conditions negotiated with each platform.

In the case of Yelp, access also involves payment for content. According to The Verge, the company allows agents to participate in its paid data licensing program.

Different uses: search, training, and navigation

The discussion becomes more complex because not all bots aim to make purchases. Some browse pages to build search indexes; others collect material to train models; agents may consult the same address to fulfill an immediate user request.

A website may want to appear in search results but refuse other uses of its content. Cloudflare, a provider of infrastructure and security, has begun offering separate controls for these activities, allowing for more specific choices than simply blocking all artificial intelligence.

The company argues that advertising, subscriptions, and direct relationships with visitors depend on people arriving at the pages. Automated access can extract information without generating the visit that supports this revenue.

“These AI agents trigger the same protections that already existed against malicious programs,” said Chris Canal, CEO of EquiStamp, an AI security assessment company, in an interview with The Verge.

Frequently Asked Questions

• Why do websites block AI agents? Reasons include security, access control, content use, and revenue preservation. Each service sets its own conditions.
• If I authorize an agent, can it access any website? No. User authorization does not eliminate technical restrictions and the rules of the visited platform.
• Does every block mean a ban against AI? No. Protection systems against automated access can also interrupt agents performing legitimate tasks.
• Are navigation agents and training bots the same? No. Agents can act in response to a user request, while training collectors gather material to develop or adjust models.
• Are all companies rejecting these assistants? No. Some block access; others allow integrations or negotiate partnerships and licensing.