Amazon's Agent Terms are a short addition to its Conditions of Use, made on May 30, 2025, that require AI agents to identify themselves when they use Amazon. The required format is a user-agent string of "Agent/[agent name]". The rule sounds technical, but it sets up Amazon's whole position on outside shopping agents: identify yourself and respect our decision, or expect to be blocked.
This explainer reflects November 2026 and is based on Amazon's announcements, press coverage and our own reading of Amazon's robots.txt. Amazon's Conditions of Use page did not load for our checker, so the wording below comes from reporting by TNW and others.
What do Amazon's Agent Terms say?
The core requirement, as reported, is simple: an AI agent that accesses Amazon must say it is an agent, using the user-agent format "Agent/[agent name]". A user-agent string is the label every browser and bot sends with each web request. A normal Chrome browser says it is Chrome. Under Amazon's terms, an agent must not pretend to be Chrome.
That matters because many browser agents run a real browser and send an ordinary browser label. To Amazon's servers, they look like a person. Identification lets Amazon decide, agent by agent, whether to allow, limit or block.
- Amazon adds Agent Terms to its Conditions of Use
- Amazon sues Perplexity over the Comet browser agent
- Federal court grants Amazon a preliminary injunction against Comet
- Amazon merges Rufus and Alexa+ shopping into Alexa for Shopping
- Ninth Circuit reportedly reverses the injunction
- Amazon blocks Meta Muse from Amazon.com
How does Amazon enforce the rules?
Amazon uses several layers together.
robots.txt
On November 10, 2026, amazon.com/robots.txt gave "Disallow: /" to 99 named user agents. The list includes ChatGPT-User, GPTBot, OAI-SearchBot, Claude-User, ClaudeBot, GoogleAgent-Mariner, GoogleAgent-Shopping, Google-Extended, Copilot, Perplexity-User, meta-externalagent, xAI-Grok, GrokBot, Manus-User and Devin. In other words, almost every agent that identifies itself honestly is told to stay out.
robots.txt is a request, not a lock. But it creates a paper trail: an agent that ignores it, or hides its identity to get around it, is easier to describe as unauthorised.
Detection and blocking
Amazon detects agent traffic and shows a block message. When it cut off Muse on September 20, 2026, users saw a popup saying continued access by an unauthorised AI agent violates Amazon's Conditions of Use, according to TNW, citing Amazon's statement to Bloomberg.
Litigation
Against Perplexity, Amazon went to court under computer-fraud law. That case is covered in our Amazon v. Perplexity explainer.
- Agent Terms added
- May 30, 2025
- Required user-agent format
- Agent/[name]
- AI agents disallowed in amazon.com robots.txt (Nov 10, 2026)
- 99
Who has Amazon blocked?
Meta Muse
Amazon began blocking Muse on Sunday, September 20, 2026, less than two weeks after Muse launched. Amazon gave four reasons, as reported by TNW and GeekWire:
- Meta never said Muse would access the store.
- Muse does not identify itself while browsing.
- Muse appears to capture and store customer credentials.
- Muse allegedly bypasses Amazon's personalisation, showing products that ignore purchase history.
Amazon said it had asked Meta to remove Amazon from Muse and Meta declined. Meta's launch post says Muse "has no visibility into people's passwords or payment methods." As of November 2026, no report says the block has been lifted. Search Engine Journal noted that Amazon's robots.txt had no Muse-specific entry, which shows the block relied on detection and terms, not robots.txt.
Perplexity Comet
Comet is the case that tested the rules in court. Amazon alleges Comet sent a Chrome user-agent string and that Perplexity changed it each time Amazon built detection. Perplexity says Amazon wants it to stop users from using AI assistants on Amazon at all.
Why does Amazon want agents to identify themselves?
Amazon's public argument is about transparency and security. A spokesperson said third-party shopping agents should "operate openly" and respect service providers' decisions about whether to take part. Hidden agents also make fraud detection harder, and agents that store a customer's Amazon login create account-security risk.
There is also a commercial side, which critics point out. Amazon earns heavily from sponsored listings and its own recommendations. An outside agent that compares prices and skips ads changes that. Amazon's own agents, Alexa for Shopping and Buy for Me, keep the shopper inside Amazon's ecosystem. Both readings can be true at once.
Pros
- Identification lets stores spot fraud and protect accounts
- Gives shoppers and merchants a clearer view of who is acting
- Fits emerging standards like Web Bot Auth and Visa's Trusted Agent Protocol
Cons
- Amazon can still refuse identified agents, so identifying yourself is no guarantee of access
- Keeps rival agents away from price comparison on the largest US store
- Leaves user-directed browsing in a legal grey area
Which agents can shop on Amazon in November 2026?
| Agent | Can it buy on Amazon? | Notes |
|---|---|---|
| Alexa for Shopping (Amazon) | Yes | Can buy at a target price with your default payment method |
| Alexa+ (Amazon) | Yes | Free with Prime, $19.99 a month otherwise |
| Meta Muse | No | Blocked since Sep 20, 2026 |
| ChatGPT agent mode, dots | No official route | ChatGPT-User disallowed in robots.txt |
| Claude Duet | No official route | Claude-User disallowed in robots.txt |
| Perplexity Comet | Disputed | Litigation ongoing; Perplexity disabled Comet's assistant on Amazon for iOS in May 2026, per court filings reported by TNW |
| Google agentic checkout | No | Named merchants are Wayfair, Chewy, Quince and Shopify stores |
If you mainly buy from Amazon, an outside agent is the wrong tool for now. Use Alexa for Shopping for price watching, and use outside agents for stores that welcome them, listed in our guide to which stores allow AI agents.
How identification could change things
The industry is moving towards agents that prove who they are cryptographically. Cloudflare's signed agents programme, the IETF Web Bot Auth draft and Visa's Trusted Agent Protocol all let an agent show which company runs it. Web Bot Auth verifies the operator, not the shopper. If those standards spread, Amazon's "Agent/[name]" rule could become one part of a wider system where stores allow signed agents and block anonymous ones. Merchants can read about the standard in our Web Bot Auth guide.
Bottom line
Amazon's Agent Terms ask one thing of AI agents: say what you are. In practice Amazon pairs that rule with a robots.txt that disallows almost every named agent, active blocking and a lawsuit. As of November 2026, the only agents that reliably shop on Amazon are Amazon's own. For a broader comparison of agents, see getpersonalaiagents.com's Alexa+ versus personal AI agents.


