--- title: "What Counts as Good When Agents Decide" canonical: "https://current.tinyfish.ai/issue/19/vision/article/10861/what-counts-as-good-when-agents-decide" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" format: "Default Settings" author: "Rina Takahashi" author_url: "https://current.tinyfish.ai/team/25" issue: 19 section: "vision" section_name: "Vision" published: 2026-03-04 updated: 2026-03-05 description: "AI shopping agents don't just filter products—their opaque criteria quietly harden into the operative definition of quality itself." word_count: 414 source_links: 3 --- # What Counts as Good When Agents Decide *By [Rina Takahashi]()* *March 4, 2026* [Read on the website]() ## The Structured Data Gate A [2025 Kantar study](https://www.kantar.com/north-america/inspiration/retail/new-study-shows-why-ai-agents-shop-like-super-consumers) found that removing a single key product attribute reduced an AI agent's probability of selecting that product by *20 to 40 percent*. Agents cannot infer what isn't there. A person might pick up a jar, turn it over, read the cramped label, and decide the small-batch hot sauce is worth trying despite the bare-bones packaging. An agent scanning structured data fields sees an incomplete record and moves on. One major search platform has been [explicit](https://www.foodnavigator-usa.com/Article/2026/02/18/why-cpg-brands-must-prepare-for-ai-shopping-agents/): if your product uses sustainable packaging but that fact isn't structured and tagged, an agent searching for "verified sustainable packaging" won't find it. The product's actual quality never enters the equation. The data describing it has to meet the agent's parsing requirements first. The gatekeeping structure here looks nothing like traditional advertising or search optimization. Retailers seeking visibility in AI shopping results navigate [allowlisting processes](https://almcorp.com/blog/chatgpt-shopping-research-the-complete-guide-to-ai-powered-product-discovery-and-llm-optimization-for-e-commerce-2025/) emphasizing machine-readable product data, structured markup, and comprehensive attribute coverage. The criteria for inclusion are formal. How products rank among those included remains unpublished. ## How Opacity Becomes the Standard Search rankings were always mysterious, but users could see the results page and apply their own judgment. Agent-mediated shopping collapses the browsing step entirely. The agent returns a curated recommendation. The user sees the output, not the filtering. What got excluded, and why, is invisible by design. Several platforms describe their shopping results as organic and unsponsored. A meaningful distinction from ad-driven marketplaces. And yet "unsponsored" and "transparent" are different words for a reason. The selection mechanism is neither paid nor explained. The product that wins the agent's recommendation wins it for reasons the user cannot inspect and the merchant cannot fully reverse-engineer. The opacity compounds through a specific loop. Merchants, unable to see the full ranking logic, optimize to the criteria they *can* observe: structured data completeness, attribute coverage, review volume. Those observable requirements harden. They become the checklist. And because every merchant optimizes to the same visible signals, the invisible criteria (whatever else the agent weighs) go unchallenged. Each cycle of optimization entrenches the legible requirements further while leaving the illegible ones untouched. The gate gets more formal. The logic behind it stays opaque. When millions of purchase decisions pass through these filters daily, the criteria stop being technical requirements and become what "good" means in practice. Nobody decided they should. Nothing in the system pushes back. The products that don't fit the pattern quietly stop appearing at all. ## More - [Vision — Issue 19]() - [Issue 19]() - [Publication guide]()