ASO: prepare for autonomous agents
ASO means Agentic Search Optimization in Seryvon. This pillar reviews signals that may help an automated agent discover a website's interfaces and understand its declared actions. The current rules inspect markup, forms, links, robots directives and supplied endpoint-probe results. They do not demonstrate that an agent can complete a transaction.
What the rules inspect
The ASO scoring source detects WebMCP-related tool declarations, JSON-LD potentialAction types, action-related schema types, usable-form signals and links resembling OpenAPI or API documentation. Several of these rules are explicitly marked experimental. Treat them as implementation-specific readiness checks rather than universal requirements for agent discovery.
The potential-action rule distinguishes types such as BuyAction, OrderAction, ReserveAction, SubscribeAction and RegisterAction from other detected actions. That distinction describes the declared action type. It does not establish that an endpoint exists, accepts valid requests or enforces the intended permissions.
Inspect form evidence before changing a page
The accessible-forms rule uses the proportion of detected forms classified as agent-usable when detailed counts are available. Its evidence includes missing actions, missing fields and missing labels. When no forms are found and none are usable, the result is not measured. Older signals without detailed counts use a presence-based fallback.
For example, if one form is usable but several others lack labels, the detailed result can expose that broader gap. Inspect each affected form and its purpose. A search form, a contact form and a reservation form may require different interaction tests. Static form evidence cannot establish successful submission, validation feedback or a completed business process.
External inputs and experimental discovery
Other rules consume AI discovery endpoint results, NLWeb probe status and brand-coherence measurements when provided. Without those inputs, they return not measured. The discovery rule uses a four-endpoint denominator; it is a Seryvon convention, not evidence that all agents require those endpoints.
The agent-access check compares evaluated bots with those blocked by robots.txt. An allowed result does not prove that a bot visited, that another access layer permits it or that a service should remove its intentional restrictions. Keep access policy decisions tied to the site's actual purpose.
Turn readiness findings into a test plan
- Choose a supported task, such as searching documentation, and identify its public entry point.
- Review labels, declared actions and documentation against what the interface actually supports.
- Resolve misleading or incomplete descriptions, then repeat the static audit.
- Separately test the permitted interaction in a controlled environment and record where it succeeds or stops.
No ranking promise
An ASO score does not guarantee agent selection, recommendations, ranking or purchases. A declaration can pass while authentication, validation or an unavailable service prevents completion. Use the Agentic search workflow for a practical review, and the Security page for related product information.
Seryvon is a public open-core release published by Powehi. The managed product is coming soon. These checks describe source capabilities and their limits, not a certification of autonomous operation.