Rows of small robot icons on glass tiles representing AI agents trying to access websites — Personal AI agents are running into blocks on retail and travel websites. Image: Getty Images via TechCrunch.

Personal AI agents can now shop, book flights and reserve tables for people. There is one big catch. A lot of websites will not let them in, and the shopper usually has no idea why.

TechCrunch's Sarah Perez reported on October 6, 2026 that websites are blocking personal AI agents. Some blocks are deliberate and some are mistakes. The same day, Meta and Sierra announced an open standard meant to fix it.

In this guide, I cover what is going on and why agents get blocked. I also share what agencies in India and the USA should check on client websites right now.

What Is Happening With AI Agents and Websites?

Personal agents such as Meta's Muse, Instinct and ChatGPT's Dots now do real tasks for people. They book flights, reserve tables and order groceries. The user does not need any technical setup.

Trouble starts when an agent reaches a site that does not want it. It also starts when the site cannot tell it apart from a spam bot. According to TechCrunch, users have complained about blocks on sites including Yelp, eBay, Zillow, Pizza Hut, Adidas and several airlines.

Here is what the companies themselves have said:

Amazon began blocking Muse from its retail site, so the agent could no longer browse or buy from the catalog. CNBC reported that Amazon cited worries about scraping.
Walmart is actually a Muse partner. It told TechCrunch the failures were not intentional. A "verify you are human" button was breaking the agent's session.
Delta said it has no partnership that lets a third-party agent book its flights. It is still evaluating how external AI agents could fit in.
Yelp said it does not allow non-human traffic unless the agent pays for access through its data licensing program.
eBay said it does not ban all shopping agents. Its rules restrict unauthorized agents and actions like scraping and model training.

So there are two very different kinds of blocks. Some are business decisions. Others are old anti-bot tools doing exactly what they were built to do, just against the wrong visitor.

Why AI Agents Get Blocked, Even by Accident

Most bot protection was built for a simple world. Humans were good, and automated traffic was suspicious. Personal agents break that rule because they are automated but act for a real customer.

Accidental blocks usually come from a few places:

Human checks. Click-to-verify buttons and puzzles assume a person is at the screen. If the agent's session is interrupted, the check fails.
Rate limits and WAF rules. Agents move fast and often run from cloud IP ranges. Firewall rules tuned for scrapers can flag them.
CDN default settings. Many sites never touch their bot settings. When the provider changes a default, behavior changes overnight.
Login and fraud systems. TechCrunch found users claiming eBay suspended their accounts over agent use. Fraud tools often treat unusual sessions as account takeover.

The tricky part is that the shopper cannot tell which kind of block they hit. They just see a failed task and blame the brand, the agent or both.

The Cloudflare factor

Cloudflare sits in front of a huge share of business websites, so its settings matter here. In a September 15 post, Cloudflare explained how it now sorts bots into search, training and agent behavior. "Agent" covers user-directed agents visiting a page on behalf of a human.

Cloudflare also said it will deprecate its single "Block AI Bots" switch in favor of these separate controls. New domains get preset configurations, and the presets for ad-supported sites are stricter. Its reasoning is simple: agents fetch the page with nobody there to see the ads.

TechCrunch noted that some people suspect these crawler default changes are disrupting Muse traffic. When asked, Cloudflare said it had no specific data to share and pointed to its public Radar data hub. So treat that theory as unconfirmed.

Cloudflare blog graphic about staying discoverable in search while disallowing AI training — Cloudflare's September 2026 update split bot controls into search, training and agent behavior. Image: Cloudflare.

Personal Agent Protocol: The Proposed Fix for AI Agents

On October 6, Sierra published its announcement of Personal Agent Protocol. It is an open standard that Meta and Sierra are developing with Genesys, Instinct, Rocket, Shopify, Stripe and Walmart. The goal is to define how personal agents interact with businesses.

Here is how it is meant to work, based on Sierra's post:

Discovery starts on the website. An agent can find out what the company offers and how to reach it.
Sessions can begin as a guest. That may be enough to check stock or ask about returns.
Sign-in is controlled by the customer. They decide whether the agent gets read-only or write access.
It runs on OAuth. This is the same well-known standard behind many "sign in with" buttons.
The business picks the route. Agents can use the regular website, APIs built on standards such as MCP and OpenAPI, or the company's own agent.

Sierra plans to publish a v0.1 specification later this month, along with design workshops and a reference implementation. Future ideas include finer permissions, push notifications and payments that do not expose card details.

Sierra co-founder Bret Taylor is leading the effort. He put it bluntly to CNBC: "It is kind of chaos until such a standard exists." CNBC also reported that OpenAI and Anthropic have not joined yet, though Taylor expects other AI companies to take part.

Personal Agent Protocol announcement graphic with logos of Meta, Sierra, Genesys, Instinct, Rocket, Shopify, Stripe and Walmart — Partners named in the Personal Agent Protocol announcement. Image: Sierra.

What This Means for Agencies and Their Clients

If you build or run websites for clients, this lands on your desk. The client will not ask "should we support AI agents?" They will ask why sales dropped, or why support tickets mention a failed booking.

Meta made the business case in its own words: "Turning away a personal agent means turning away the customer behind it." That is a fair point for many retailers. It is not true for every business, though.

A publisher that lives on ad views has a real reason to be careful. A lead-gen site might worry about fake form fills. A SaaS login page has security concerns. The right answer depends on how the client makes money.

Client type	Starting stance on agents	Main thing to watch
Ecommerce and D2C stores	Allow on product and checkout pages	Payment fraud and inventory scraping
Travel and booking	Allow with tight rate limits	Fare scraping and fake reservations
Ad-supported publishers	Restrict or allow selectively	Lost ad views and content reuse
SaaS and member portals	Allow only after sign-in	Account takeover and data access
Lead generation sites	Allow browsing, guard forms	Spam and fake leads

We covered the shopping side earlier in our Meta Muse for small business checklist. The short version is that agent traffic is becoming normal customer traffic for many stores.

A Practical Checklist for Making Sites Ready for AI Agents

You do not need to wait for the spec to act. Here is the checklist I would run on client sites this month:

Write down a policy first. Agree with the client on which pages agents may use and which they may not. Put it in the terms of use too.
Audit your bot settings. Check the CDN, WAF and any bot management tool. Look for defaults that changed in the last few weeks.
Review human checks on key flows. Find every CAPTCHA or verify button on login, cart and checkout. Ask if each one is still needed.
Separate agents from scrapers in logs. Tag traffic by behavior, not just by user agent. Track failed checkouts and abandoned sessions from automated visitors.
Keep markup clean. Agents read pages like a careful user. Clear labels, proper form fields and structured product data help them finish tasks.
Plan an API path. For high-volume clients, a documented API or MCP server may serve agents better than the website.
Track the Personal Agent Protocol. Read the v0.1 spec when it ships. Decide which clients should be early adopters.

Do not throw away your defenses, though. The OWASP Automated Threats to Web Applications project lists real attacks like scalping, credential stuffing and carding. Bad bots are still out there, and some will pretend to be friendly agents.

Researchers recently tracked a large Chinese AI agent fleet. We covered it in our post on AI bot traffic and the Tencent agent fleet. The goal is to let in good agents without opening the door to everyone.

Notes for teams in India and the USA

Many Indian agencies build stores for US brands, so both markets matter at once. Muse already has millions of US users, according to Meta's David Singleton in the CNBC interview. US clients will feel this first.

For Indian D2C brands, the timing is a little different. Agent shopping is newer there, but Shopify and Stripe are both protocol partners. If your clients use those platforms, support may arrive through updates you do not control. Keep an eye on release notes.

Key Takeaways

Personal AI agents like Muse, Instinct and Dots are being blocked by many sites, both on purpose and by accident.
Amazon and Yelp block by choice. Walmart's failures came from a human-verification check, even though it partners with Muse.
Cloudflare now separates search, training and agent bot controls, with stricter presets for ad-supported sites.
Personal Agent Protocol, from Meta, Sierra and partners, uses OAuth to let businesses see and control agent access. A v0.1 spec is due this month.
Agencies should set a per-client agent policy, audit bot settings and human checks, and keep defenses against bad bots.

Need help reviewing how your site handles automated traffic, or planning an API for agents? Our team offers software development consulting for agencies and growing businesses in India and the USA.

About the Author

Nitin Kumar is Lead Project Coordinator at Debugged Pro. He manages web, ecommerce and AI projects for startups and growing businesses in India and the USA. He writes about practical fixes that web and product teams can ship quickly.