The intersection of artificial intelligence and e-commerce reached a new point of friction this week as independent retailers discovered their proprietary product data being harvested for a secretive Amazon initiative. This program, identified by industry observers and affected sellers as Project Starfish, involves the automated scraping of independent business websites to generate AI-powered listings within the Amazon Shopping app.

Independent brands, including Bobo Design Studio and Hitchcock Paper, reported that the e-commerce giant is currently testing a beta program that pulls products from small business websites without the consent of the owners. These retailers found that Amazon uses AI to mine information from their sites, creating listings that the sellers did not opt into and cannot opt out of. The discovery has sparked a debate regarding intellectual property, the accuracy of generative AI in retail, and the boundaries of platform dominance in the digital age.

The mechanics of Project Starfish and agentic discovery

At the core of this development is a technical framework designed to expand Amazon’s reach into inventory it does not directly house or fulfill. Through “Project Starfish,” Amazon is currently scraping thousands of independent websites to create AI-generated listings within its own application. These listings are often presented to users under a “Buy for Me” or “Shop other stores directly” label.

The system operates by deploying AI agents that scan the open web for products not currently available in the Amazon store. Once a product is identified, the system “re-skins” the independent business’s offering, funnelling it into an Amazon-branded checkout process. This represents a significant shift toward agentic AI adoption, where autonomous software components execute complex workflows—including purchasing—on behalf of a user.

Technical documentation and seller reports indicate that the program creates a layer of abstraction between the consumer and the actual merchant. Depending on the product, customers can either shop the site with the help of AI or visit the site directly to complete the transaction. However, the automated nature of this process has led to significant technical discrepancies that affect brand reputation.

AI hallucinations and data integrity challenges

One of the primary technical failures reported by affected retailers involves the inaccuracy of the AI-generated content. Amazon’s AI agents have been shown to “hallucinate” product details, displaying information that is factually incorrect but presented as authoritative. These hallucinations include displaying outdated prices and selling items that are no longer in stock.

For a small business, the display of incorrect pricing or out-of-stock items creates immediate operational friction. When an AI-generated listing offers a product at a price the merchant no longer honors, the merchant is forced to choose between taking a financial loss or disappointing a new customer. Furthermore, the program uses AI images that are not owned by the original creators, according to a.pdf. This practice breaks the chain of customer trust and disrupts the pricing and fulfillment models established by indie shops, according to a.pdf.

The phenomenon of AI hallucinations is not unique to Amazon’s retail experiments but takes on a higher risk profile when tied to financial transactions. Other platforms have faced similar challenges, such as the hallucination phenomenon in TV search, where inaccurate responses are mitigated through verification servers. In the context of Project Starfish, no such verification layer appears to be accessible to the merchants whose data is being used.

The proxy email problem and customer relationships

A significant technical hurdle for sellers receiving orders through this program is the lack of direct customer data. Merchants are left with “ghost orders” from anonymous proxy emails, which makes it impossible to provide support or build a real customer relationship.

In traditional e-commerce, the relationship between a buyer and a seller is foundational for handling returns, exchanges, and shipping updates. By masking the customer’s true identity, Amazon maintains control over the data while the independent seller bears the responsibility for fulfillment and potential complaints. This data harvesting is viewed by some industry analysts as an attempt to gather information to inform Amazon’s own merchandising and target advertising.

This shift in data ownership aligns with broader trends in the industry where first-party data collection is prioritized to circumvent privacy restrictions. By positioning itself as the intermediary for off-platform products, Amazon effectively captures shopping intent signals that would otherwise remain on independent domains.

Institutional hypocrisy and the scraping debate

The discovery of Project Starfish highlights a perceived hypocrisy in platform policies. Amazon currently blocks AI companies from scraping its own data while simultaneously harvesting the intellectual property of independent sellers.

Amazon has been aggressive in defending its own data borders. The company expanded bot restrictions on November 24, 2025, to block OpenAI-related crawlers from accessing Amazon.com. This defensive posture suggests that while Amazon views its own product data as a proprietary asset to be protected from competitors, it views the data of independent retailers as a public resource for its own AI training and listing generation.

This tension is mirrored in the legal landscape. For example, xAI recently filed a lawsuit against California over laws that would force AI firms to reveal training secrets. The conflict between the need for transparency in AI training and the protection of trade secrets remains a central point of contention for all major technology platforms.

Strategic context within Amazon’s AI ecosystem

Project Starfish does not exist in a vacuum; it is part of a larger push toward autonomous shopping. Amazon announced on November 18, 2025, a comprehensive update on its AI shopping features, noting that products available through “Buy for Me” had increased from 65,000 at launch to over half a million.

The company has also integrated its Rufus AI assistant, which manages price tracking and auto-buying features. These tools are designed to reduce friction for the consumer, but they often do so by bypassing the traditional brand-to-consumer interaction. While Amazon describes these features as helping customers “discover and evaluate products,” the retailers involved argue that they are being “hijacked” without permission.

Industry impact and the future of retail media

For the marketing community, these developments signal a new era of “unauthorized” retail media. Traditionally, retailers chose to list on Amazon and paid for the privilege through commissions and advertising fees. With Project Starfish, the choice is removed.

The growth of Amazon DSP and the introduction of tools like Amazon Marketing Cloud show a company focused on unifying the shopping experience across the web. However, when this unification happens without the consent of the product owners, it raises questions about the long-term viability of the independent web.

Some large retailers have opted for formal partnerships, such as Macy’s integrated Amazon’s retail advertising technologyin late 2025. This allows the retailer to maintain control over its customer relationships and data. The contrast between these formal agreements and the “agentic scraping” of small businesses highlights a deepening divide in how platform power is exercised.

The emergence of Project Starfish and “Shop Other Stores Directly” marks a fundamental shift in how digital advertising and product discovery operate. For PPC professionals and digital marketers, the traditional “walled garden” of Amazon has expanded its walls to include your own website.

According to PPC Land, the rise of agentic AI means that a brand’s SEO and product data are no longer just for human eyes or standard search engines; they are now being actively ingested by shopping agents that may re-skin and re-sell that content. This creates a “leaky bucket” for traffic, where a potential customer may find a brand’s product via an AI listing on Amazon rather than visiting the brand’s site directly.

Marketers must now consider “agentic optimization.” It is no longer enough to rank on Google; one must ensure that if an Amazon agent scrapes a site, the information it retrieves is accurate. However, as the reports from Hitchcock Paper and Bobo Design Studio show, even accurate data can be distorted by AI “hallucinations” once it enters the Amazon ecosystem. This loss of control over brand presentation, pricing, and customer data represents one of the most significant challenges to independent e-commerce in the current decade.

• August 2024: Amazon DSP introduces modeled attribution for off-Amazon conversions, laying the groundwork for tracking the full customer journey across the web.
• January 9, 2025: Amazon launches its Retail Ad Service, allowing other retailers to use Amazon’s advertising technology.
• November 4, 2025: Amazon issues a legal demand to Perplexity AI, demanding they stop AI agents from making purchases on Amazon, while simultaneously expanding its own scraping agents.
• November 11, 2025: Amazon introduces free AI prompts for sponsored ad campaigns to automate shopper engagement.
• November 18, 2025: Amazon updates its AI shopping features, revealing that the “Buy for Me” program now includes over 500,000 products, many from off-Amazon sources.
• November 24, 2025: Amazon expands its robots.txt restrictions to block crawlers from Meta, Google, and OpenAI.
• December 2025: Independent sellers go viral on Instagram and LinkedIn, exposing the unauthorized scraping of their websites via Project Starfish.

Summary

• Who: Amazon and its proprietary AI agents are the primary actors, affecting thousands of independent retailers and small business owners such as Bobo Design Studio and Hitchcock Paper.
• What: A beta program known as Project Starfish and “Shop Other Stores Directly” that uses AI to scrape independent websites, create unauthorized listings, and facilitate purchases through Amazon’s “Buy for Me” agentic interface.
• When: The program saw significant expansion in late 2025, with major adoption metrics released in November 2025 and widespread seller discovery in December 2025.
• Where: The activity occurs across the open web (scraping) and within the Amazon Shopping app (listing and checkout), primarily targeting US-based independent brands.
• Why: Amazon aims to expand its product selection, gather competitive merchandising data, and capture first-party shopper signals, though it does so at the cost of merchant consent and data accuracy.