To Sam Altman, Dario Amodei, Demis Hassabis, and all leaders shaping the future of artificial intelligence,

As you guide humanity into an era increasingly defined by AI-powered language models and intelligent interfaces, we respectfully offer a proposal that could serve the long-term interests of both the AI ecosystem and the open web: the creation and adoption of a public, standardized architecture that allows businesses to deterministically deliver structured, query-responsive product listings to AI systems.

We propose an Open Web AI Query Architecture (OWAQA)—a set of conventions and protocols that empower e-commerce businesses to serve highly relevant, structured product content to AI systems, enabling those systems to provide more contextually accurate, user-aligned recommendations, while giving businesses the control necessary to optimize product-market fit.

The Core Architecture Proposal

1. robots.txt + llm.txt + API Endpoint
   • Extend the existing robots.txt convention to include a reference to a llm.txt file.
   • The llm.txt acts as a publicly accessible manifest that includes a human-readable and machine-readable summary of the company’s site, product taxonomy, and preferred query routes.
   • This file also points to one or more RESTful API endpoints available to approved AI crawlers or systems.
2. REST API for AI Queries
   • The REST API accepts structured AI queries (e.g., product intent, budget, use case, features needed).
   • It routes the query to the business’s own Model Context Protocol (MCP)—a logic layer architecture introduced by Anthropic that facilitates structured context exchange between LLMs and external data sources.
3. Vector Database Matching
   • The MCP then queries a vector database maintained by the business. This database contains thousands of micro-targeted product variants, each optimized for specific personas, needs, or contexts.
   • The system identifies the best-fit variant and sends back a complete product listing: title, description, pricing, images, schema.org metadata, and a canonical product URL.
4. Response to AI System
   • The AI assistant presents the matched product listing, with attribution and linkage, to the end user in the chat or result interface.
   • Optional: systems can exchange performance signals or analytics (with consent) to help businesses further optimize listings.

Why This Matters: Determinism in a Probabilistic World

AI systems are inherently probabilistic. But businesses operate on determinism: they require accuracy, brand consistency, and control. They need to know:

• What messaging is shown to customers.
• Which products are surfaced.
• Why a certain product was recommended.

This proposed architecture bridges the gap. It offloads the responsibility of granular content optimization to businesses—the parties best positioned to align product offerings to nuanced user needs. It gives AI systems a structured, interpretable, and optimized content pipeline. It gives marketers a coliseum to compete in—a framework where better messaging, better product-market alignment, and better user fit are rewarded.

Unlocking Micro-Targeting at Scale

In traditional web SEO and SEM, a business might maintain one or two variants of a product listing. But AI-driven interfaces change the game: a single product may resonate differently with a commuter, a college student, and a retiree. This architecture allows businesses to:

• Create hundreds of micro-targeted variants of a single product.
• Match them to intent-derived segments generated by AI queries.
• Optimize messaging for each micro-audience—emphasizing different features, benefits, use cases, or cultural language.

This model delivers more value to end users. The AI presents more relevant content. The business captures more conversions. Everyone wins.

Guardrails, Attribution, and Reputation

This system also enables:

• Rate-limited, secure access to business APIs for AI agents.
• Business-defined gating mechanisms (e.g., API keys, usage quotas).
• Fair attribution so users know where the information originated.
• Reputation systems to incentivize high-quality data and penalize manipulation.

The Path Forward

This architecture is not science fiction. All its components—REST APIs, vector databases, schema.org, manifest files—exist today. What we need is consensus, adoption, and encouragement from the major AI players:

• OpenAI, Anthropic, Google DeepMind, Mistral, Meta, Amazon—your influence can guide the creation of llm.txt and REST discovery standards.
• W3C and Schema.org—your stewardship can formalize the protocols.
• Browser vendors and developer platforms—your support can ease implementation.

We urge you to collaborate with the open web community and e-commerce ecosystem to create a future where AI models interact with structured, high-quality, business-optimized content in a transparent, accountable, and mutually beneficial way.

The future is not closed. It is collaborative.

Signed,

Matt Brutsché - 500 Rockets Marketing Think Tank & Digital Agency