← All articles

Mikhail Polyakov

Websites are closing their doors to AI agents: who is keeping them from buying and booking, why, and what a shop owner should decide

Level: BeginnerReading time: 7 min

Topics: AI, Small business

A small brass courier robot with an empty basket stands at the locked glass door of a shop with shelves of goods visible inside

The short answer: AI agents that OpenAI, Meta, Google and other companies release to buy and book on a person's behalf are running into closed doors. According to The Verge, in recent weeks users of such agents have been reporting blocks on popular sites. Yelp, by its own account, lets agents in only through a paid programme. Delta Air Lines does not let them buy tickets, the banks JPMorgan Chase and Charles Schwab have confirmed that they do not let agents into accounts, and Amazon is suing Perplexity, the maker of one such agent. Some agents, meanwhile, are cut off by protections set up against bots in general. Below: who does what, what the agent makers reply and what a website owner should decide from this.

Where the news comes from

I read The Verge article (by Emma Roth, 17 October 2026). It brings together company statements, user reports and an expert comment. I did not open the TechCrunch piece, the Meta post, the Reddit discussions or the court documents: everything said about them below is as reported by The Verge.

What an AI agent is in this story

It is a program based on a neural network that a person gives a task in words. According to the promises of its makers, as retold by The Verge, an agent can find and book a restaurant table on its own, sort through email or buy new shoes. The article names these agents: Muse from Meta, Grok from xAI, Dots from OpenAI and an agent from the company Instinct.

Who is closing the doors and how

According to The Verge:

  • Yelp, as the company told TechCrunch, lets agents in if their makers pay for Yelp's data licensing programme;
  • Delta Air Lines does not let agents buy tickets on users' behalf;
  • JPMorgan Chase and Charles Schwab confirmed that they do not let agents into customer accounts for security reasons;
  • Cars.com and Amtrak, according to users, stopped opening for their agents, although they used to. The Verge suggests that other companies may be quietly blocking agents too, but that is the outlet's assumption;
  • Amazon said that Muse's "continued access" violates the site's rules. There are also reports that Grok, the Instinct agent and Dots cannot use Amazon.

The Amazon and Perplexity dispute

Amazon already has a court case over agents. According to The Verge, last year it banned Perplexity's shopping agent and sued the company. The parties' positions are as follows.

  • Amazon: it "repeatedly asked" Perplexity to stop agent purchases on its site; companies that buy on behalf of customers should act openly and respect a service's decision whether to take part or not.
  • Perplexity: Amazon does not want agents because then fewer people see its advertising. Perplexity compares its agent to an extension of the user: in its words, Amazon does not recognise a person's right to hire an assistant or an employee who acts on their behalf.

A judge ordered Perplexity to stop agent purchases on Amazon, but in August an appeals court overturned that decision. After that, Amazon filed an amended lawsuit. How the dispute will end is not yet known.

When an agent is cut off by bot protection

Not all blocks are aimed at agents. According to Chris Canal, head of EquiStamp, a company that evaluates AI safety, many protections are set up against bots in general: against programs that try to commit fraud or break in. AI agents trigger the same protections. Examples from the article:

  • according to posts on Reddit, Google is blocking some accounts that use Muse for "suspicious activity";
  • some users cannot complete payment on Walmart, even though Walmart has an arrangement with Meta.

The Verge also mentions Cloudflare, a company that helps sites protect themselves from unwanted AI visitors. This is about other programs, crawlers: they collect site pages for search and for AI training rather than buying on a user's behalf. In 2025 Cloudflare introduced one-click blocking of AI crawlers and later started blocking crawlers that collect sites for both search and AI training at the same time.

What the agent makers reply

  • Meta wrote in a post earlier in October that refusing a personal agent means refusing the customer behind it, and that the company prefers to work with businesses on their concerns. According to The Verge, instead of fighting Amazon, Meta made deals with Walmart, Gap, Sephora, Best Buy and other retailers so that Muse can buy from them.
  • Shopify opened its network of shops to agents. More on this in the piece Shopify opens its catalogue to AI agents.
  • OpenAI and Google did not immediately respond to The Verge's questions, and Instinct declined to comment.

Chris Canal believes that if companies obstruct agents, the agents can do less that is useful, and that this affects how companies like OpenAI and Anthropic are valued. He also expects the web to become "annoying": "prove you are not a robot" checks may appear even on a simple grocery store site. This is his opinion and forecast, not data.

What a site owner and someone who uses an agent should decide

This is my conclusion from the article, not advice from its author.

For the owner of an online shop or booking service. The article shows four ways companies deal with agents:

  • let them in by agreement with the agent's maker, like Meta's partner retailers;
  • let them in for a fee, like Yelp;
  • keep them away from specific actions, like Delta with ticket purchases and the banks with accounts;
  • fight in court, like Amazon.

Four things are worth weighing: sales that may come through agents; visitors who see your site and advertising themselves; security; and whether your shop can correctly accept and fulfil an order placed by a program. The banks in the article closed accounts to agents precisely for security reasons, and bot protection is needed against fraudsters.

The order I would suggest:

  1. Find out what the site does now. Ask the developer or contractor who set up the site's protection whether it lets automated visitors through to checkout and payment. According to Canal, bot protection can cut off agents even without a separate decision by the owner.
  2. Test it with a single order. The most direct way: place one order for an inexpensive item on the live site through an AI agent, the way a customer would, and watch the agent while it works. Do not touch the protection for this. Two things need checking. First: whether the agent got as far as payment or was stopped by the protection. Second, if the order went through: whether the employee who processes orders sees it with the correct item, quantity, price, address and phone number, and can handle it in the normal way, that is, pick and ship it or cancel it with a refund. After the check, cancel the order and make sure the money has come back.
  3. If the agent did not get to the end, your protection does not currently let agents through; what to do about that is in point 4. If the agent got through but the order was recorded with errors or the employee could not process it, pass this on to your contractor and do not count on orders from agents until a repeat test order goes through without errors.
  4. If you want to let agents in but the protection blocks them, it is not worth weakening it for all bots for this purpose: the existing protection against fraudsters matters more than sales through agents. In the article, Meta negotiates access with retailers, and Shopify itself opened its network of shops to agents, that is, access is granted by the shop or its platform. So if your shop runs on an e-commerce platform, first find out from it whether it has such a programme. If there is no programme or you do not have a platform of your own, leave the protection as it is and come back to the question when such a programme appears.

For someone who hands purchases to an agent. According to the reports in the article, on some large sites the agent does not cope: it is not let in, the account gets blocked or payment fails. The banks in the article do not let agents into accounts for security reasons. Hence three simple rules. Before paying, check yourself exactly what the agent put in the basket and at what price. Do not give the agent access to your bank account, since the banks themselves do not let it in. If the agent could not place the order, place it yourself on the shop's site.

Summary

The makers of AI agents promise that an agent will buy and book for you, but, according to The Verge, some large sites do not let them in. Some charge for access, others close specific actions such as buying tickets or access to bank accounts, Amazon is suing Perplexity, and bot protections cut off agents along the way. Meta's response is to negotiate with retailers directly, and Shopify has opened its network of shops to agents. A site owner should know what their protection does with agents and decide whether they need such customers without sacrificing protection against fraudsters.

I work on AI agents and automation. If you would like to see my projects or discuss your own task, take a look at my portfolio.

Sources