Amazon AI Agent Rules and Scraping in 2026: What Changed, What Still Works

By Elena Park · 2026-07-25 · 7 min read read · Engineering

Amazon is blocking AI shopping agents and restricting AI crawlers while courts decide where the line is. What changed in 2026, whether scraping Amazon is legal, and safer ways to get product data.

Short answer: can you still scrape Amazon in 2026?

Technically yes for public product pages, but not without risk. Amazon actively restricts automated access, it says unauthorized AI agents breach its Conditions of Use, and it has gone to court to stop AI shopping agents. The safest options are official data programs for sellers and affiliates, or a data provider that handles collection for you.

This article covers what changed, what courts have said so far, whether scraping Amazon is legal, and what works in practice. It is general information, not legal advice.

What changed with Amazon and AI bots

August 2025: Modern Retail reported that Amazon had added several AI-related crawlers, including ones from Meta, Google, Huawei and Mistral, to the bots its robots.txt tells not to crawl, after earlier restrictions reported on crawlers from Anthropic, Perplexity and Google's Project Mariner agent. A robots.txt file is a request, not a technical block, so only bots that follow it are affected.

November 2025: Amazon sued Perplexity over its Comet browser agent. Amazon accused Perplexity of computer fraud for not disclosing when Comet shops on a real person's behalf, and for continuing after Amazon asked it to stop.

March 2026: a federal district court granted Amazon a preliminary order stopping Comet from making purchases on Amazon while the case continued. Reports say the judge found Amazon likely to prove a violation of the Computer Fraud and Abuse Act. Reports also say Amazon added an agent policy to its Business Solutions Agreement, effective 4 March 2026.

4 August 2026: the Ninth Circuit lifted that order. According to MediaPost's report of the opinion, the court said the user, not the agent, is the one who accesses Amazon's site, and that holding Perplexity liable would need a novel reading of the law. Amazon said it was evaluating its next steps.

September 2026: Meta released its Muse agent on 8 September, and on 20 September Amazon said Muse could no longer place orders on Amazon.com. Users saw a message that continued access by an unauthorized AI agent violates Amazon's Conditions of Use. Amazon said Muse does not identify itself and appears to store customer credentials. Meta said Muse has no visibility into passwords or payment methods, according to PPC Land's report.

What this means: AI agents versus scrapers

These disputes are about shopping agents acting for logged-in customers, which is different from collecting public product data. They still show where Amazon stands: agents should identify themselves and operate with permission, and Amazon will block those that do not. Amazon's own Buy for Me feature identifies itself and lets brands opt out, per PPC Land.

If your tool places orders, fills carts or logs in to accounts, treat it as an agent and expect Amazon to block it. If it only reads public product pages, you are in a different position, but not a risk-free one.

Is scraping Amazon legal?

There is no single answer. It depends on what you collect, how you collect it, and where you and your users are. In the United States, courts in the hiQ v. LinkedIn litigation treated scraping publicly accessible pages as unlikely to violate the Computer Fraud and Abuse Act, and the Supreme Court's Van Buren decision read that law's 'exceeds authorized access' wording narrowly.

Those cases do not remove contract risk. Amazon says unauthorized automated access breaches its Conditions of Use, and the Muse notice shows it will say so to the public. The Ninth Circuit's Perplexity ruling is about whether an agent acting for a user 'accesses' Amazon's servers under the CFAA. It is not a ruling that scraping Amazon is permitted.

Personal data is a separate issue. Reviewer names, seller details and anything that identifies a person can bring in privacy law such as the GDPR. Collect product-level facts such as price, availability and ratings, not personal data.

Official ways to get Amazon data

Sellers: the Selling Partner API gives sellers programmatic access to their own catalog, orders and pricing data, and Seller Central has built-in reports. Use these before you scrape your own account.

Affiliates and publishers: Amazon has deprecated the Product Advertising API (PA-API 5). Amazon's notice says calls to it now return HTTP 403 and tells developers to migrate to the Creators API, which it describes as a REST API for programmatic access to Amazon's product catalog data for publishers, influencers and affiliate partners. If an old integration stopped working, this is the likely reason.

Everyone else: data providers and scraping APIs sell structured Amazon product data. Check what their terms promise about compliance, how often the data refreshes, and what happens when Amazon changes its pages.

If you still collect public Amazon pages: what works in practice

Treat it as a risk you manage, not a loophole. Use only publicly visible product pages, never log in, never collect personal data, and do not scale faster than you can justify.

Large retailers widely flag datacenter IP addresses, so teams that collect public retail pages generally use residential proxies. Use country-matched IPs for local marketplaces, for example German IPs for amazon.de, because prices and availability depend on the delivery location.

Keep request rates low and spread out, reuse a session when a page flow needs it, expect CAPTCHAs, and back off when you see errors instead of retrying immediately. Use a real browser only for pages that need JavaScript, because plain HTTP requests use far less bandwidth.

Measure cost per successful page, not cost per gigabyte. Retries, blocked pages and heavy page assets all use bandwidth, so a cheaper network that fails more often can cost more.

If Amazon starts blocking you consistently, treat it as a signal. Check whether an official or licensed source covers your need before you escalate your methods.

Common Amazon scraping problems and fixes

Problem: CAPTCHA pages. Fix: slow down, reduce parallel requests, keep one IP and session per browsing flow, and switch from datacenter to residential IPs if you have not already.

Problem: 503 errors or empty pages. Fix: back off for longer after each failure, spread requests across time, and check that your headers and TLS behaviour match a normal browser if you use plain HTTP.

Problem: prices that differ from what you see in your browser. Fix: set the delivery location, and use a proxy in the same country, because Amazon shows price and availability by delivery location.

Problem: an old Amazon API integration that stopped working. Fix: check whether it used the Product Advertising API, which Amazon has deprecated, and migrate to the Creators API.

What we could not confirm

We could not confirm the number of detection signals Amazon uses, the text of Amazon's agent policy, or any rule that bans automation of Seller Central. An earlier version of this article stated those as facts. We removed them because we could not find a source.

The Amazon v. Perplexity case is still moving, and Amazon has said it is evaluating its next steps after the Ninth Circuit ruling. We will update this article when the position changes.

Frequently Asked Questions

Sources and further reading

• PPC Land: Amazon blocks Meta's Muse AI agent from shopping on its site
• Modern Retail: Amazon quietly blocks AI bots from Meta, Google, Huawei and more
• MediaPost: Appeals court sides against Amazon, lifts Perplexity ban
• Bloomberg Law: Amazon wins court order blocking Perplexity's AI shopping bots
• Amazon Creators API: PA-API 5 deprecation notice