Everyone Wants a Marketing Taxonomy
Feature hubs, solution pages, glossaries, directories. If you use AI to generate your site structure, the temptation is a parallel fiction — a structure that describes a product you did not actually build. Search engines are just another user who will notice.
The SpielOS site exists to send qualified visitors to the waitlist. The shortest honest path is to describe what the product actually is. SpielOS is a harness: roles, skills, workflows, evals, context, providers, connections. That is the architecture, and it is also the sitemap.
The instruction I gave myself was blunt: do not invent a separate marketing taxonomy. No parallel directory, no solution pages, no teams or templates that do not exist in the product. The features the product ships are the features the site ranks for.
A marketing taxonomy that does not match the product is content debt you will pay forever.
The Product Architecture as the Sitemap
Four hubs, eleven pages, zero invented categories.
The result is a route tree that mirrors the harness directly. Each hub is a layer of the product, and each page under it is a real capability with a real search intent.
Chat Director Mode for long-running agent sessions, Direct Mode for workflow execution and scheduling.
Context Files as the knowledge base, Strategy for prompt and instruction management, Memory for persistent state.
Harness Agents as roles, Skills as reusable capabilities, Workflows as multi-agent pipelines, Evals as quality gates.
Infrastructure Providers for model selection, Connections for MCP, OAuth, and API integrations.
The sitemap is the product. Every page earns its place by describing something the product actually does.
Two Languages, One Structure
The same route tree, mirrored under /fa, with real Persian search intent.
Bilingual SEO was a direct extension of the architecture decision. Each English page has a Persian wrapper under /fa/, sharing the same components and the same taxonomy. The search intent is different — Iranian teams searching for agent orchestration in Persian — but the structure is identical.
That symmetry is what makes the SEO maintainable. One architecture change updates both languages. No parallel content strategy drifting in opposite directions.
hreflang, canonical, and locale metadata are automatic because the URL structure is. Architecture once, SEO twice.
Search Wants What the Product Is
When model capability converges, the differentiator is the harness around it. Search engines are just another user of that same answer.
The search result for "agent harness" should match a page about the harness because that is what the product is. The search result for "workflow execution" should match the workflows page because workflows are how execution happens. No translation layer, no marketing gloss.
The fastest SEO is honesty about what you built. The architecture is the answer; everything else is decoration.
This is the same decision that should shape how you use AI to present your own business: write about what the system actually does, including the parts that broke. Product truth compounds. Marketing fiction does not.
Architecture-First SEO
Three moves that cost almost nothing and change everything — especially if you generate pages with AI.
Map the real product. List every capability the product ships. That list is your sitemap. Do not add a page the product cannot answer.
Ban parallel taxonomies. If the marketing structure does not match the product structure, one of them is wrong. It is usually the marketing one.
Mirror it in every language. One architecture, mirrored wrappers. Bilingual SEO becomes a routing problem, not a content problem.
Stop optimizing the homepage. Optimize the truth. The architecture already did the keyword research.
The features hub is the product, not a description of it.
Chat, Context, Harness, Infrastructure. Four hubs, eleven pages, two languages, one architecture.
See the product structure