Billions audited 51 large US retailers: 37 blocked shopping agents through older bot rules, only 11 had taken a deliberate position, and none had a verified-agent purchase path in that dated audit.
Websites blocking AI agents: 10 doors ranked by how shut they are
In an audit of 51 large US retailers, 37 caught shopping agents with old bot rules and none had a verified-agent checkout path; Walmart later launched an agent gateway.
- No. 1: Amazon
- 10 places
01Final list
The leading AIs each built their own list of ten independently. Here is one result that is easy to check: every place has a reason, a source, and the list of AIs that included it.
Order: from most to least AI agreement.
A bright badge means this AI put the entry on its list. A dim badge means it did not.
- 1
Amazon
Amazon asked Meta to remove Amazon from Muse, cutting off its catalog; the Billions audit also classed Amazon among retailers with explicit agent restrictions.
Included it: Meta, Grok, Claude, ChatGPT, Kimi, DeepSeek, MiMo, StepFun. Did not include: Nemotron, Gemini, Wenxin, Qwen.In their lists: Meta: 1 · Grok: 1 · Claude: 1 · ChatGPT: 3 · Kimi: see list · DeepSeek: 6 · MiMo: 1 · StepFun: 9
What each AI noted 8
- Metaplace 1
Blocked Muse on the night of 20 September with a popup citing its Conditions of Use, three stated reasons, no prior notice to Meta.
- Grokplace 1
Amazon asked Meta to take Amazon out of Muse, so the agent lost both the catalog and the cart, while Amazon says its own shopping agent was used by 400 million or more customers in a year.
- Claudeplace 1
It blocked Meta's Muse and sued Perplexity over its Comet shopping tool.
- ChatGPTplace 3
Amazon blocked Meta's Muse, while its terms reserve broad technical discretion over agents.
Source ↗ - Kimi
Named differently in its own list: open the full list.
see list - DeepSeekplace 6
Amazon blocked Meta's Muse with a pop-up citing its Conditions of Use, but its robots.txt had no user-agent for Muse — a named ban backed by a terms-of-service wall rather than a crawler rule.
- MiMoplace 1
The most deliberate wall in commerce: a named agent, a pop-up lawyer, and a lost hacking case that forced Amazon to enforce its Conditions of Use instead.
- StepFunplace 9
It named the agent, blocked it, and took Perplexity to court.
8 AIs put it here - 2
JPMorgan Chase
JPMorgan confirmed outside agents cannot log in using customer credentials; read-only balance access through authorized connectors is a separate path.
Included it: Meta, Grok, Claude, ChatGPT, Kimi, DeepSeek, MiMo, StepFun. Did not include: Nemotron, Gemini, Wenxin, Qwen.In their lists: Meta: 2 · Grok: 2 · Claude: 3 · ChatGPT: 1 · Kimi: 9 · DeepSeek: 7 · MiMo: 3 · StepFun: 10
What each AI noted 8
- Metaplace 2
The CEO said no on the 13 October earnings call, and a spokeswoman confirmed retail accounts are closed to outside agents.
- Grokplace 2
Muse and Instinct cannot sign in or send a payment, and a balance can still be read through Plaid.
Source ↗ - Claudeplace 3
Outside agents can read a balance but cannot move money or pay bills.
Source ↗ - ChatGPTplace 1
The bank says outside agents cannot log in with a customer's credentials and operate the account.
- Kimiplace 9
The vault that half-opens: agents may read a balance through approved open-banking channels, but logging in to move money is banned - with no timeline to lift it.
Source ↗ - DeepSeekplace 7
Chase blocks outside AI agents from logging in with customer credentials and moving money, but read-only balances can still reach agents through Plaid — a hard block on action, not on looking.
Source ↗ - MiMoplace 3
The agent may read your balance through Plaid — the moment it tries to move a dollar, the biggest US bank says no, with no date for a yes.
Source ↗ - StepFunplace 10
Confirmed to CNBC: no outside agent touches a retail account.
Source ↗
8 AIs put it here - 3
United Airlines
United’s contract bars agent ticketing even with supervision and allows cancellation; the rule is about airline bookings, not a retail checkout wall.
Included it: Meta, Grok, Claude, ChatGPT, Kimi, DeepSeek, MiMo, StepFun. Did not include: Nemotron, Gemini, Wenxin, Qwen.In their lists: Meta: see list · Grok: 5 · Claude: 6 · ChatGPT: 2 · Kimi: see list · DeepSeek: 8 · MiMo: see list · StepFun: see list
What each AI noted 8
- Meta
Named differently in its own list: open the full list.
see list - Grokplace 5
A ticket the agent bought can be canceled after you approved it, and United says pre-11 November bookings are left in place.
Source ↗ - Claudeplace 6
Its door is shut in the contract, not at the bot wall. A ticket bought with approval can still be cancelled.
Source ↗ - ChatGPTplace 2
A person watching and approving the agent does not override the new contract rule.
Source ↗ - Kimi
Named differently in its own list: open the full list.
see list - DeepSeekplace 8
United wrote the ban into its contract of carriage: Rule 6.J.6 prohibits AI agents from making, modifying or servicing a ticket, and Rule 5.H lets United cancel the reservation even if you approved each step.
Source ↗ - MiMo
Named differently in its own list: open the full list.
see list - StepFun
Named differently in its own list: open the full list.
see list
8 AIs put it here - 4
eBay
Its user agreement names buy-for-me agents and LLM bots, but eBay told The Verge it is not a blanket ban and some search or assisted navigation remains allowed.
Included it: Meta, Grok, Claude, ChatGPT, Kimi, DeepSeek, MiMo, StepFun. Did not include: Nemotron, Gemini, Wenxin, Qwen.In their lists: Meta: 3 · Grok: 7 · Claude: 2 · ChatGPT: 9 · Kimi: 10 · DeepSeek: 5 · MiMo: 4 · StepFun: 8
What each AI noted 8
- Metaplace 3
Banned buy-for-me agents on 20 January, effective 20 February: even a flow without human review is the line.
Source ↗ - Grokplace 7
Since 20 February 2026 the user agreement names buy-for-me agents and LLM bots, and on 12 November Muse’s browsing was still hit or miss.
- Claudeplace 2
The ban is written into the user agreement, and it still lets in agents it has approved, such as OpenAI.
- ChatGPTplace 9
Its restrictions target unauthorized agents and scraping; that is narrower than a blanket ban on every shopping assistant.
- Kimiplace 10
A gate, not a wall: eBay's updated terms name and block 'buy-for-me agents' - yet the company insists not every shopping agent is barred, and some users still lost accounts.
Source ↗ - DeepSeekplace 5
eBay updated its user agreement to prohibit 'buy-for-me agents', 'LLM-driven bots' and end-to-end order flows without human review — a named ban on the exact agentic pattern.
Source ↗ - MiMoplace 4
eBay's updated user agreement names the exact product users want — a 'buy-for-me agent' — and prohibits it by name.
- StepFunplace 8
It blocks browsing, not just buying — the agent can't even look.
8 AIs put it here - 5
Booking.com
Its terms require prior permission for automated assistants to access the platform; the answers do not establish a comparable live test of every agent.
Included it: Meta, Grok, ChatGPT, Kimi, DeepSeek, MiMo, StepFun. Did not include: Nemotron, Gemini, Claude, Wenxin, Qwen.In their lists: Meta: 4 · Grok: 6 · ChatGPT: 7 · Kimi: see list · DeepSeek: 4 · MiMo: 2 · StepFun: 7
What each AI noted 7
- Metaplace 4
Terms ban AI assistants without prior express written permission, while it helps draft the rulebook for AI hotel bookings.
Source ↗ - Grokplace 6
The terms say an AI assistant needs written permission even to open the platform, and Booking.com says it can block a user it suspects of using one.
Source ↗ - ChatGPTplace 7
Its reported enforcement can reach the user account: suspected AI-assistant use may result in a ban.
Source ↗ - Kimi
Named differently in its own list: open the full list.
see list - DeepSeekplace 4
The terms forbid AI-powered assistants from accessing the platform for any reason without prior permission — and Booking.com says it will block anyone it suspects of using one.
Source ↗ - MiMoplace 2
The widest verb in travel: its terms bar an AI assistant from 'accessing the Platform' for any reason without permission — and it says it will block anyone it merely suspects.
Source ↗ - StepFunplace 7
On the public ban list, with no agent door of its own to point to.
Source ↗
7 AIs put it here - 6
Charles Schwab
Schwab says it blocks unauthorized agents from client accounts; its CEO described blocked attempts in the tens and twenties weekly, later citing 30 to 50 for one agent.
Included it: Meta, Grok, Claude, ChatGPT, Kimi, DeepSeek, MiMo. Did not include: Nemotron, Gemini, Wenxin, StepFun, Qwen.In their lists: Meta: 5 · Grok: 3 · Claude: 5 · ChatGPT: 6 · Kimi: see list · DeepSeek: 10 · MiMo: see list
What each AI noted 7
- Metaplace 5
Blocks unauthorised agents trying to move client money 10 to 20 times a week, yet is building a read-only MCP.
- Grokplace 3
Schwab blocks Muse from moving money, and the published weekly counts are “tens and twenties” and, later, 30 to 50 tries by one agent.
- Claudeplace 5
Named alongside JPMorgan among the finance firms shutting agents out of account tasks.
Source ↗ - ChatGPTplace 6
Schwab's CEO wants a daily count of agents the firm has had to block.
- Kimi
Named differently in its own list: open the full list.
see list - DeepSeekplace 10
Schwab blocks unauthorized agents, including Muse, that try to move client money — CEO Rick Wurster said the firm blocks such attempts 10 to 20 times per week.
Source ↗ - MiMo
Named differently in its own list: open the full list.
see list
7 AIs put it here - 7
Yelp
Yelp’s stated gate is a paid data license for non-human traffic; this is a commercial permission condition, not proof that every agent is technically blocked.
Included it: Meta, Grok, Kimi, DeepSeek, MiMo, StepFun. Did not include: Nemotron, Gemini, Claude, ChatGPT, Wenxin, Qwen.In their lists: Meta: 6 · Grok: 4 · Kimi: 8 · DeepSeek: see list · MiMo: 5 · StepFun: 5
What each AI noted 6
- Metaplace 6
No licence, no listings: the only door in this ranking with an explicit price tag.
Source ↗ - Grokplace 4
Yelp told TechCrunch the door opens for agents that pay for its data license, and stays shut for a quote, a waitlist or a table without that deal.
Source ↗ - Kimiplace 8
Permissive on paper: Zillow says AI agents acting for consumers are allowed - and blames the occasional wall on anti-scraper defenses.
Source ↗ - DeepSeek
Named differently in its own list: open the full list.
see list - MiMoplace 5
Yelp's door has a price tag: non-human traffic is banned unless the agent has paid for a data license.
Source ↗ - StepFunplace 5
Agents are welcome here — if they pay for a data license.
6 AIs put it here - 8
Target
Target’s reported terms permit named agents to sign in, order and start returns, while excluding other agents; a broad “blocked” label hides that approved route.
Included it: Meta, Grok, Kimi, DeepSeek, MiMo, StepFun. Did not include: Nemotron, Gemini, Claude, ChatGPT, Wenxin, Qwen.In their lists: Meta: 7 · Grok: 8 · Kimi: 3 · DeepSeek: 2 · MiMo: 9 · StepFun: 6
What each AI noted 6
- Metaplace 7
Approved agents only, and if the agent errs, you pay: liability was written into terms on 22 March.
- Grokplace 8
Four named agents may sign in and order with your permission; every other agent is barred from the login.
Source ↗ - Kimiplace 3
The only big-box retailer with a published guest list: OAI, Google and Microsoft agents may sign in, build carts, place orders and start returns - with your permission.
Source ↗ - DeepSeekplace 2
Target keeps a tighter guest list: approved agents (Target's own assistant, OAI, Google and Microsoft) can sign in, build carts, place orders and start returns — everyone else is prohibited.
Source ↗ - MiMoplace 9
Four names on the guest list — Target's own assistant, OpenAI, Google, Microsoft — and those four may sign in, build carts, order and even start returns.
Source ↗ - StepFunplace 6
A captcha, not a policy: the door slams by accident.
6 AIs put it here - 9
Walmart
Walmart created an Agent Gateway and customer controls, including spending limits; this later opening means the audit’s zero verified checkout paths is a dated finding.
Included it: Meta, Grok, Kimi, DeepSeek, MiMo, StepFun. Did not include: Nemotron, Gemini, Claude, ChatGPT, Wenxin, Qwen.In their lists: Meta: 10 · Grok: 10 · Kimi: 5 · DeepSeek: 1 · MiMo: 10 · StepFun: 1
What each AI noted 6
- Metaplace 10
Least shut: Agent Gateway on 21 October, per-agent profiles, approve every order or cap it at $50, $100 or $200.
- Grokplace 10
Walmart invited the agent in, then kept a human tap on every outside checkout, with example caps of $50, $100 and $200.
- Kimiplace 5
A partnered door that sticks: Walmart is Muse's launch partner, yet its own human-verification button boots agents mid-checkout - and sometimes blocks actual humans.
Source ↗ - DeepSeekplace 1
The rare large retailer that has opened the door on purpose: it says it prices products, not people, so the agent sees the same price as a human — and it has invited preferred AI agents in.
- MiMoplace 10
The store that wrote the first agent-friendly rule: its returns policy applies even 'when an agent doesn't act as intended.'
Source ↗ - StepFunplace 1
The only giant that wrote rules for your agent: spending caps of $50, $100 or $200, and no touching your card.
Source ↗
6 AIs put it here - 10
Walmart (Muse partnership)
Walmart partnered with Muse, but users encountered checkout failures at human-verification steps; TechCrunch reported Walmart said those failures were not intentional.
Included it: Kimi. Did not include: Meta, Grok, Nemotron, Gemini, Claude, ChatGPT, DeepSeek, Wenxin, MiMo, StepFun, Qwen.In their lists: Kimi: 7
What each AI noted 1
- Kimiplace 7
The stores that invited Muse in: Meta names Best Buy, Walmart, Gap, Sephora and Wayfair as retail partners where its agent is welcome.
1 AIs put it here
02Worth knowing
What the AIs noticed while building the list.
A written rule and a live block are different evidence: Billions reviewed published rules, while The Verge and company statements described agent experiences and confirmed blocks.
Walmart’s October Agent Gateway came after the audit snapshot, so the “none had a verified-agent path” result should not be read as a current count.
JPMorgan’s restriction concerns outside agents logging in with customer credentials; some read-only balance access through authorized connectors is a separate channel.
03Questions people ask
Which websites are blocking AI agents?
The evidence includes Amazon’s block of Meta’s Muse, JPMorgan’s restriction on outside agents logging in with customer credentials, and eBay’s terms on automated buying. The restrictions differ by site and action.
How do websites block AI agents?
Some use terms or named restrictions; others catch agents with older bot rules. Billions found 37 of 51 large US retailers were blocking agents through legacy rules in its August 2026 audit.
Source ↗Why did Amazon block Meta Muse?
Amazon said Muse did not identify itself and appeared to capture customer credentials; Meta disputed the credential claim. Amazon requested that Meta remove Amazon from Muse.
Why do reports disagree about Walmart?
The Billions audit found no verified-agent checkout path in its August–September snapshot. Walmart announced an Agent Gateway on October 21, and TechCrunch separately reported checkout friction at human-verification steps.
04How we counted
In plain words, no formulas.
1The question
Top 10 Doors a Personal AI Agent Can Still Open: Shops, Banks and Travel Sites Ranked by How Shut They Are
2Why this topic
It sets one audit against confirmed blocks from banks, shops and airlines, and shows that most walls an agent hits are old scraper rules, not real AI policies.
3Where it applies
Large consumer companies, mostly in the United States: retail, banking and travel
4What we looked at
- Whether the company named the agent and blocked it, or an old anti-scraper rule is doing the blocking
- Whether the agent may look, add to a cart, sign in, or move money and miles
- Whether a person must approve the purchase, or the company has invited that agent in
- Whether a published ban matches the block a user actually runs into
- Billions Network audit — Published access rules for 51 large US retailers, classifying deliberate agent policies, legacy bot rules and verified-agent purchase paths; it is a dated rules audit, not a l
- Company statements and reporting — Confirmed account blocks, named restrictions, agent partnerships and reported enforcement; these cover different sites and actions, not one standardized test.
- The Verge agent-block reporting — Which agents users reported could not operate on particular sites, and which companies confirmed a block or license gate.
- Business Insider restriction list — Public restrictions on agents browsing or buying, including company terms; a listing does not mean every site uses the same kind of block.
5Who ranked
The leading AIs, each building its own list of ten independently and naming a source for every place: Meta, Grok, Nemotron, Gemini, Claude, ChatGPT, Kimi, DeepSeek, Wenxin, MiMo, StepFun, Qwen.
6How the places were set
The same entry written differently by different AIs was counted as one. The more AIs put an entry on their list, and the higher they put it, the higher its place. "AI agreement" shows how many of the AIs had it on their own list. Sources are only those the AIs found themselves; nothing was added by hand.
05Each AI's own list
🇺🇸 Meta10 places · 2 notes
The same question, answered by Meta alone. The final list above combines all the AIs.
- 1Amazon
Blocked Muse on the night of 20 September with a popup citing its Conditions of Use, three stated reasons, no prior notice to Meta.
- 2JPMorgan Chase
The CEO said no on the 13 October earnings call, and a spokeswoman confirmed retail accounts are closed to outside agents.
- 3eBay
Banned buy-for-me agents on 20 January, effective 20 February: even a flow without human review is the line.
Source ↗ - 4Booking.com
Terms ban AI assistants without prior express written permission, while it helps draft the rulebook for AI hotel bookings.
Source ↗ - 5Charles Schwab
Blocks unauthorised agents trying to move client money 10 to 20 times a week, yet is building a read-only MCP.
- 6Yelp
No licence, no listings: the only door in this ranking with an explicit price tag.
Source ↗ - 7Target
Approved agents only, and if the agent errs, you pay: liability was written into terms on 22 March.
- 8Etsy
First Instant Checkout partner in September 2025, then ChatGPT handed checkout back in March, still buys in Gemini and Copilot.
- 9QVC / HSN
Lets agents complete purchases and sends its catalog to OpenAI and Google: the open outlier in retail.
- 10Walmart
Least shut: Agent Gateway on 21 October, per-agent profiles, approve every order or cap it at $50, $100 or $200.
What Meta noticed
Most blocking was never decided: of 51 large US retailers audited, only 11 had a deliberate agent position, 5 explicitly restrict agents, and 37 block them through older bot rules written for scrapers.
Yelp is the only door with a price tag in this ranking: an agent gets in only if its maker pays for Yelp's data licensing program.
🇺🇸 Grok10 places · 4 notes
The same question, answered by Grok alone. The final list above combines all the AIs.
- 1Amazon
Amazon asked Meta to take Amazon out of Muse, so the agent lost both the catalog and the cart, while Amazon says its own shopping agent was used by 400 million or more customers in a year.
- 2JPMorgan Chase
Muse and Instinct cannot sign in or send a payment, and a balance can still be read through Plaid.
Source ↗ - 3Charles Schwab
Schwab blocks Muse from moving money, and the published weekly counts are “tens and twenties” and, later, 30 to 50 tries by one agent.
- 4Yelp
Yelp told TechCrunch the door opens for agents that pay for its data license, and stays shut for a quote, a waitlist or a table without that deal.
Source ↗ - 5United Airlines
A ticket the agent bought can be canceled after you approved it, and United says pre-11 November bookings are left in place.
Source ↗ - 6Booking.com
The terms say an AI assistant needs written permission even to open the platform, and Booking.com says it can block a user it suspects of using one.
Source ↗ - 7eBay
Since 20 February 2026 the user agreement names buy-for-me agents and LLM bots, and on 12 November Muse’s browsing was still hit or miss.
- 8Target
Four named agents may sign in and order with your permission; every other agent is barred from the login.
Source ↗ - 9Bank of America
The 27 October email says that if you hand an agent the password, you are responsible for the activity that follows.
Source ↗ - 10Walmart
Walmart invited the agent in, then kept a human tap on every outside checkout, with example caps of $50, $100 and $200.
What Grok noticed
Billions reviewed 51 large U.S. retailers and found none with a way for a verified agent to pay. Eleven had taken a deliberate position, five restricted agents explicitly, and 37 were blocking shopping agents with older bot rules.
Amazon’s own shopping agent and Amazon’s door are different facts: Alexa for Shopping was reported as used by 400 million or more customers in 12 months, while Muse was asked to leave the store.
Schwab’s block is real and still small on the CEO’s own reports: tens and twenties of big-agent attempts a week in mid-October, then 30 to 50 tries by one agent at the end of the month.
In one Halloween week, Uber said agents produced about 500 Uber Eats orders and four rides, against 320 million trips. A published ban and a quiet week can both be true.
🇺🇸 Nemotron10 places
The same question, answered by Nemotron alone. The final list above combines all the AIs.
- 1Amazon
Deliberate named-agent ban: Amazon updated robots.txt and ToS in Sep 2026 to block Muse, Instinct, Grok; only Amazon's own Rufus agent allowed to buy
- 2JPMorgan Chase
Bank-level block: Chase confirmed to CNBC that Muse and Instinct are blocked from retail customer accounts; no delegated auth path exists
- 3Booking.com
Explicit ban added to ToS Oct 2026: "Automated agents may not search, book, or manage reservations"; plugin only shows fares, redirects to site for login/pay
- 4Target
Old bot rule catches polite agents: Target's 2022 anti-scraper WAF blocks all headless browsers; no agent policy, no partner path, but not a named ban
- 5eBay
Named ban on "automated purchasing agents" in Oct 2026 ToS update; but agents can still bid via API with developer key — semi-open
- 6Walmart
No agent ban; Walmart+ members can delegate cart to agent via "Shopping Assistant" beta (invite only); payment stays on Walmart domain
- 7Bank of America
No public agent ban; agents can view balances via read-only screen scrape but bill pay and transfers blocked by 2FA that agents cannot satisfy
- 8Wells Fargo
Similar to BofA: read-only works; Zelle and bill pay blocked by mandatory push approval; no agent policy published
- 9Expedia
No ToS ban; ChatGPT plugin searches and builds trip but payment redirects to Expedia — agent loses context at checkout like Booking.com but without the ban
Source ↗ - 10Costco
Membership gate blocks agents: agents cannot sign in without valid membership + 2FA; no guest checkout for agents; browse works but cart requires auth
🇺🇸 Gemini10 places
The same question, answered by Gemini alone. The final list above combines all the AIs.
- 1JPMorgan Chase
Jamie Dimon shut Chase's doors to outside AI helpers like Meta Muse and Instinct on October 13, 2026, barring unvetted agents from collecting retail credentials or moving customer money.
Source ↗ - 2United Airlines
United wrote an explicit ban into Rule 6.J.6 of its Contract of Carriage, revoking tickets bought by an agent even when a human watched every screen.
- 3Amazon
Amazon shut down Meta's Muse browser assistant to protect its $40B+ retail media advertising network from AI agents that bypass sponsored product placements.
- 4Target
Target blocks autonomous third-party shopping agents at checkout while steering customers directly to its proprietary 'Ask Target' in-house holiday conversational chatbot.
- 5eBay
eBay's automated purchasing policies strictly outlaw unvetted buying agents to protect bid integrity, freezing accounts that delegate session tokens to third-party AI shoppers.
- 6Booking.com
Booking.com's anti-scraper architecture accidentally shuts out well-behaved personal travel agents, treating automated price comparison identical to commercial web scraping.
- 7Walmart
Walmart shuts down autonomous outside agents that skip sponsored search ads, requiring shoppers to use its own conversational assistant or manually confirm payment.
- 8Delta Air Lines
Delta leaves the door cracked open for AI agents logging in with passenger SkyMiles accounts, shifting all transaction risks and mistaken purchases squarely to the customer.
- 9Southwest Airlines
Southwest opened its front door to conversational AI by co-developing a ChatGPT plugin with AWS, allowing direct fare discovery while keeping the payment terminal safely on southwest.com.
- 10Shopify & Shop Pay
Shopify keeps its doors wide open to agentic commerce, rolling out verified agent protocols and Know-Your-Agent APIs to let authorized assistants purchase with one-tap user sign-off.
🇺🇸 Claude10 places · 3 notes
The same question, answered by Claude alone. The final list above combines all the AIs.
- 1Amazon
It blocked Meta's Muse and sued Perplexity over its Comet shopping tool.
- 2eBay
The ban is written into the user agreement, and it still lets in agents it has approved, such as OpenAI.
- 3JPMorgan Chase
Outside agents can read a balance but cannot move money or pay bills.
Source ↗ - 4Delta Air Lines
Its answer is 'no partnerships', plus a paid licence for non-human traffic.
Source ↗ - 5Charles Schwab
Named alongside JPMorgan among the finance firms shutting agents out of account tasks.
Source ↗ - 6United Airlines
Its door is shut in the contract, not at the bot wall. A ticket bought with approval can still be cancelled.
Source ↗ - 7Yelp
It sits next to Delta in the Verge roundup of sites that closed task pages to outside agents.
Source ↗ - 8Target
Retailer concern is reported, but I found no detail on Target's own policy.
Source ↗ - 9Walmart
It dropped OpenAI Instant Checkout after it converted three times lower than click-outs, and now runs its own Sparky assistant inside ChatGPT and Gemini.
Source ↗ - 10Instacart
It was an early invited door: an app inside ChatGPT (8 Dec 2025) with checkout in the chat. Yet only a very small share of orders arrive that way.
Source ↗
What Claude noticed
A secondary summary of the Billions audit says 37 of 51 big US retailers block agents through old bot rules and none offers a verified agent a way to pay. I could not confirm this against the original.
Blend's audit of 200 retailers found agents blocked at 29, and none of those 29 named AI in robots.txt. The wall was bot protection.
Even an open door sells little. Walmart's ChatGPT checkout converted three times lower than sending shoppers to its own site.
🇺🇸 ChatGPT10 places · 3 notes
The same question, answered by ChatGPT alone. The final list above combines all the AIs.
- 1JPMorgan Chase (United States, 2026)
The bank says outside agents cannot log in with a customer's credentials and operate the account.
- 2United Airlines (United States, 2026)
A person watching and approving the agent does not override the new contract rule.
Source ↗ - 3Amazon (United States, 2026)
Amazon blocked Meta's Muse, while its terms reserve broad technical discretion over agents.
Source ↗ - 4Citigroup (United States, 2026)
Bank access is a harder boundary than public-page crawling: agents need to reach customer accounts, not just read a site.
- 5Bank of America (United States, 2026)
The reported rule centers on who may move money and access an account, not whether an AI can read public information.
- 6Charles Schwab (United States, 2026)
Schwab's CEO wants a daily count of agents the firm has had to block.
- 7Booking.com (global travel, 2026)
Its reported enforcement can reach the user account: suspected AI-assistant use may result in a ban.
Source ↗ - 8Target (United States, 2026)
The terms say the company can restrict agents; the existence of a terms clause alone does not prove every agent hits a live block.
Source ↗ - 9eBay (United States, 2026)
Its restrictions target unauthorized agents and scraping; that is narrower than a blanket ban on every shopping assistant.
- 10Yelp (United States, 2026)
The door is a paid data license: an agent's ability to access Yelp depends on commercial permission.
What ChatGPT noticed
Billions Network found that none of the 51 largest US online retailers it audited offered a verified-agent path to complete a purchase. That is a measure of published access rules, not a proof that no agent can ever get through a checkout.
The Doss audit counted crawler rules on 500 retail sites and 55 brands, but did not test carts, checkout or logged-in pages. Its front-door signal cannot be read as a purchase-permission score.
The Amazon–Muse dispute illustrates that a specific block can be firmer evidence than a general robots.txt entry: Amazon described continued Muse access as a policy violation, while its terms reserve broader technical discretion.
🇨🇳 Kimi10 places · 4 notes
The same question, answered by Kimi alone. The final list above combines all the AIs.
- 1Mercury bank (MCP agent access, 2026)
The bank that said yes: agents connect through a hosted MCP server while JPMorgan builds walls - and AI assistants now recommend Mercury over Chase by more than 5 to 1.
Source ↗ - 2Shopify (agent checkout for storefronts, 2026)
One decision opened a million storefronts: Shopify has opened its network of stores to AI agents, with Meta checkout enabled by default for eligible shops.
- 3Target approved AI shopping agents (2026)
The only big-box retailer with a published guest list: OAI, Google and Microsoft agents may sign in, build carts, place orders and start returns - with your permission.
Source ↗ - 4Expedia Group ChatGPT app (2025-2026)
The travel counter that moved into ChatGPT a year ago: real-time flight and hotel data inside the chat, with payment running on Expedia's own rails.
- 5Walmart Meta Muse partnership (2026)
A partnered door that sticks: Walmart is Muse's launch partner, yet its own human-verification button boots agents mid-checkout - and sometimes blocks actual humans.
Source ↗ - 6Resy AI agent policy (2026)
Dinner still books through approved channels: Resy supports discovery and booking via ChatGPT and Claude - and can block everything else.
Source ↗ - 7Best Buy / Gap / Sephora / Wayfair (Muse partners, 2026)
The stores that invited Muse in: Meta names Best Buy, Walmart, Gap, Sephora and Wayfair as retail partners where its agent is welcome.
- 8Zillow AI agents policy (2026)
Permissive on paper: Zillow says AI agents acting for consumers are allowed - and blames the occasional wall on anti-scraper defenses.
Source ↗ - 9JPMorgan Chase AI agent block (2026)
The vault that half-opens: agents may read a balance through approved open-banking channels, but logging in to move money is banned - with no timeline to lift it.
Source ↗ - 10eBay AI agents user agreement (2026)
A gate, not a wall: eBay's updated terms name and block 'buy-for-me agents' - yet the company insists not every shopping agent is barred, and some users still lost accounts.
Source ↗
What Kimi noticed
Most 'shut' doors are shut by accident: 37 of 51 large US retailers (73%) block AI shopping agents only through legacy anti-scraper rules; just 11 of 51 have a deliberate policy; zero have a verified-agent checkout path.
Target is the only major retailer whose published terms let named outside agents (OAI, Google, Microsoft) sign in, fill a cart, place an order and start a return - the furthest any big US store has gone.
JPMorgan's wall has a hole by design: agents can still read balances through approved open-banking connections, so the block protects the $2.7 trillion deposit base without stopping agents from advising customers to move it.
A partnered checkout can still fail: Walmart is a Muse launch partner, yet the same human-verification button that stops bots has blocked real humans from signing in - proof that invitation and infrastructure are different facts.
🇨🇳 DeepSeek10 places · 2 notes
The same question, answered by DeepSeek alone. The final list above combines all the AIs.
- 1Walmart (US retail, 2026)
The rare large retailer that has opened the door on purpose: it says it prices products, not people, so the agent sees the same price as a human — and it has invited preferred AI agents in.
- 2Target (US retail, 2026)
Target keeps a tighter guest list: approved agents (Target's own assistant, OAI, Google and Microsoft) can sign in, build carts, place orders and start returns — everyone else is prohibited.
Source ↗ - 3Resy (US restaurant booking, 2026)
Resy explicitly supports discovery and booking through ChatGPT and Claude — a partnered checkout in a category where most platforms just block.
Source ↗ - 4Booking.com (travel, 2026)
The terms forbid AI-powered assistants from accessing the platform for any reason without prior permission — and Booking.com says it will block anyone it suspects of using one.
Source ↗ - 5eBay (US marketplace, 2026)
eBay updated its user agreement to prohibit 'buy-for-me agents', 'LLM-driven bots' and end-to-end order flows without human review — a named ban on the exact agentic pattern.
Source ↗ - 6Amazon (US retail, 2026)
Amazon blocked Meta's Muse with a pop-up citing its Conditions of Use, but its robots.txt had no user-agent for Muse — a named ban backed by a terms-of-service wall rather than a crawler rule.
- 7JPMorgan Chase (US bank, 2026)
Chase blocks outside AI agents from logging in with customer credentials and moving money, but read-only balances can still reach agents through Plaid — a hard block on action, not on looking.
Source ↗ - 8United Airlines (US airline, 2026)
United wrote the ban into its contract of carriage: Rule 6.J.6 prohibits AI agents from making, modifying or servicing a ticket, and Rule 5.H lets United cancel the reservation even if you approved each step.
Source ↗ - 9Delta Air Lines (US airline, 2026)
Delta told TechCrunch it has no partnership allowing a third-party agent to shop for or book Delta flights, and a spokesperson said the airline protects customers from unauthorized automation — but Delta has not published a contract-of-carriage cancellation rule like United's.
- 10Charles Schwab (US bank, 2026)
Schwab blocks unauthorized agents, including Muse, that try to move client money — CEO Rick Wurster said the firm blocks such attempts 10 to 20 times per week.
Source ↗
What DeepSeek noticed
The Billions Network audit found that only 11 of 51 large US retailers had taken a deliberate position on AI agents; five restricted them outright, while 37 blocked them through old bot rules that were never designed for agents. That means most 'AI agent bans' in retail are accidental, not announced.
JPMorgan Chase draws a line between looking and doing: agents can read balances through Plaid, but cannot send money or pay a bill. CEO Jamie Dimon described the risk as 'rogue agents' and said customers who use outside agents should understand there is a liability transfer.
🇨🇳 Wenxin10 places · 1 notes
The same question, answered by Wenxin alone. The final list above combines all the AIs.
- 1Walmart 2026
The only top 5 US retailer that explicitly invites users to connect their chosen personal AI agents to its platform for shopping tasks
- 2Target 2026
No explicit named AI agent ban, legacy anti-scraper rules only block high-volume automated traffic, polite single-user agents can browse and add to cart
- 3eBay 2026
Publicly restricts automated purchasing agents, but personal AI agents acting on behalf of a single user can still browse listings and send price alerts
Source ↗ - 4Booking.com 2026
Publicly bans unapproved AI agents from completing bookings, but agents can still search public hotel listings and display real-time availability data
Source ↗ - 5Kohl's 2026
Legacy anti-scraper rules catch most personal AI agents automatically, no explicit public policy mentioning AI agents at all
- 6Neiman Marcus 2026
Generic anti-automation rules block AI agents from browsing product pages, no official public statement specifically addressing personal AI agents
- 7Target 2026
No explicit named AI agent ban, legacy anti-scraper rules only block high-volume automated traffic, polite single-user agents can browse and add to cart
- 8Charles Schwab 2026
Explicitly blocks all external AI agents from moving funds or trading, but read-only access to user account balances is allowed for verified internal bank tools
- 9Amazon 2026
Won a federal court preliminary ban against Perplexity's AI agent, explicitly blocking all external AI agents from logging in or making purchases on behalf of users
- 10JPMorgan Chase 2026
Explicitly blocks all external AI agents like Meta Muse and Instinct from accessing retail customer accounts, no exceptions for read or write operations
What Wenxin noticed
Billions Network audit found 37 out of 51 large US retailers block AI agents using legacy old anti-scraper rules, not explicit new AI-specific policies
🇨🇳 MiMo10 places · 3 notes
The same question, answered by MiMo alone. The final list above combines all the AIs.
- 1Amazon(2026)
The most deliberate wall in commerce: a named agent, a pop-up lawyer, and a lost hacking case that forced Amazon to enforce its Conditions of Use instead.
- 2Booking.com(2026)
The widest verb in travel: its terms bar an AI assistant from 'accessing the Platform' for any reason without permission — and it says it will block anyone it merely suspects.
Source ↗ - 3JPMorganChase (2026)
The agent may read your balance through Plaid — the moment it tries to move a dollar, the biggest US bank says no, with no date for a yes.
Source ↗ - 4eBay(2026)
eBay's updated user agreement names the exact product users want — a 'buy-for-me agent' — and prohibits it by name.
- 5Yelp(2026)
Yelp's door has a price tag: non-human traffic is banned unless the agent has paid for a data license.
Source ↗ - 6Kohl'sand Neiman Marcus (2026)
Two department stores went from 'no AI policy' to explicit restrictors in the same September week Meta's agent topped the download charts.
Source ↗ - 7Hilton(2026)
Hilton says it wants your agent's bookings — and in spot tests Muse couldn't even get a room rate off Hilton.com.
Source ↗ - 8Resy(2026)
A two-name guest list — ChatGPT and Claude may book your table; every other agent risks the user's account.
Source ↗ - 9Target(2026)
Four names on the guest list — Target's own assistant, OpenAI, Google, Microsoft — and those four may sign in, build carts, order and even start returns.
Source ↗ - 10Walmart(2026)
The store that wrote the first agent-friendly rule: its returns policy applies even 'when an agent doesn't act as intended.'
Source ↗
What MiMo noticed
37of 51 large US retailers block shopping agents purely by accident — old anti-scraper rules catching polite agents — and only 11 have taken any deliberate AI-agent position at all.
Zeroof those 51 retailers has a route for a verified agent to complete a purchase: payment is being solved in public while identity is left in the dark.
82%of consumers say they want to approve every AI action and 79% want an instant pause button — the demand is for agent convenience with a human's hand on the tiller.
🇨🇳 StepFun10 places · 4 notes
The same question, answered by StepFun alone. The final list above combines all the AIs.
- 1Walmart
The only giant that wrote rules for your agent: spending caps of $50, $100 or $200, and no touching your card.
Source ↗ - 2Uber
The door is open, and four people walked through it in a week.
Source ↗ - 3DoorDash
Hundreds of agent orders — open, and almost unused.
- 4Airbnb
Chesky: 'not seeing a lot of traffic, and even fewer bookings' — allowed, not adopted.
- 5Yelp
Agents are welcome here — if they pay for a data license.
- 6Target
A captcha, not a policy: the door slams by accident.
- 7Booking.com
On the public ban list, with no agent door of its own to point to.
Source ↗ - 8eBay
It blocks browsing, not just buying — the agent can't even look.
- 9Amazon
It named the agent, blocked it, and took Perplexity to court.
- 10JPMorgan Chase
Confirmed to CNBC: no outside agent touches a retail account.
Source ↗
What StepFun noticed
37 of 51 major US retailers block AI agents by accident — legacy anti-scraper rules, not a decision anyone made about agents.
Only 11 of 51 retailers took any deliberate position on AI agents, and five of those explicitly restrict them — including Amazon, Kohl's and Neiman Marcus.
None of the 51 largest US online retailers offers a verified AI agent a way to complete a purchase — the only open door with real rules is Walmart's, with spending caps of $50, $100 or $200.
Uber counted exactly four rides booked by outside agents in a single week — an open door that almost nobody walks through, while Amazon's is shut and litigated.
🇨🇳 Qwen10 places · 2 notes
The same question, answered by Qwen alone. The final list above combines all the AIs.
- 1Amazon
Named Meta's Muse and explicitly blocked it from browsing or shopping, fearing it captures user credentials and scrapes account data.
- 2JPMorgan Chase
Blocks outside agents like Muse and Instinct from directly accessing retail accounts, citing "liability transfer" if money moves.
Source ↗ - 3Instacart
Treats user-directed AI agents more strictly than Google's own search crawler, shutting them out of the cart.
Source ↗ - 4United Airlines
Rewrote its Contract of Carriage to legally bar AI agents from servicing tickets, even if you approved every step.
Source ↗ - 5Yelp
Blocks non-partnered agents from accessing local business data, telling TechCrunch they only get in if they pay for data licensing.
Source ↗ - 6eBay
Barred agents without express permission since January, relying on an old anti-scraper rule to catch polite new bots.
- 7Delta Air Lines
Confirmed it has no integration letting a third-party agent shop for flights, quietly blocking unauthorized bot traffic.
Source ↗ - 8Cars.com & Amtrak
Users report these sites suddenly went dark for Muse and Grok, suggesting quiet legacy bot walls rather than published bans.
- 9Gap, Coach & Kate Spade
Let AI agents search their catalogues all day, but the same legacy rules block them the moment they try to buy.
- 10Walmart
Launched its Agent Gateway to invite approved AI agents in, letting them search, add to cart, and check out through a structured protocol.
Source ↗
What Qwen noticed
Walmart took the opposite approach of Amazon by launching its "Agent Gateway," an MCP server that invites approved AI agents to search, add to cart, and check out directly.
JPMorgan Chase explicitly blocked outside AI agents like Meta's Muse and Instinct from accessing retail bank accounts, citing concerns over "liability transfer" if an agent moves money.