About AXRI

How we evaluate Agentic AI readiness.

Traditional websites optimize for human browsing. Agentic AI systems need explicit, machine-readable signals to discover, understand, and safely use your site.

The ecosystem is moving quickly across standards bodies, AI clients, and commerce protocols. This page explains what we check today, why each signal or a proxy for it matters, and how the score is weighted. These scores are very much a starting point, and we expect both the checks and their weight to evolve as best practices and standards emerge.

Note: Even without these specific configurations, agents may still be able to process a website. These specific configurations make websites more accessible to agents and help improve reliability and consistency for automation and better agentic experience on web.

Back to evaluator

Core Checks

  • llms.txt

    Helps agents find a clear entry point for important pages, usage context, and crawl intent without reverse-engineering your full site structure.

  • JSON-LD structured data

    Reduces ambiguity by exposing entities and relationships in a machine-readable format that many search and assistant systems already parse reliably.

  • API hints

    Makes integration paths discoverable when agents need actions, not just reading. Without this, agents often fall back to brittle HTML interpretation.

  • Machine-friendly content

    Signals whether key pages can return markdown or JSON for reliable parsing and lower hallucination risk during extraction.

  • Sitemap discoverability

    Improves deterministic crawling and reduces missed important URLs, especially when site navigation is JS-heavy.

Optional Governance Checks

  • ai-plugin manifest

    Helps compatible AI clients discover tool metadata and endpoint details.

  • ai.txt policy

    Communicates AI/agent access expectations as this convention continues to mature.

  • security.txt and robots.txt

    Improves trust and operational clarity around responsible access and disclosure.

Shopping Signals

Shopping and checkout workflows require stronger interoperability contracts than content discovery. In the broader ecosystem, this is where Agent Card, capability advertising, and protocol-level actions are evolving fastest.

  • Agent Card discovery

    Indicates whether your commerce agent can advertise capabilities to other agents.

  • Commerce protocol endpoints

    Shows whether machine-usable endpoints exist for product, cart, or checkout style interactions instead of scraping human-only UI.

  • Payment capability signals

    Helps determine if autonomous buying flows can be completed safely, not just discovered.

Score Weightage

The readiness score is currently based on core non-shopping signals. Optional and shopping checks are shown for visibility but do not yet add points.

  • llms.txt: 100 points

    If present, it is treated as a strong readiness indicator and currently carries full weight.

  • Fallback core split (when llms.txt is missing)

    JSON-LD: 25, API hints: 35, machine-friendly content: 30, sitemap discoverability: 10.

  • Partial credit behavior

    Machine-friendly content can receive partial points when only plain text is detected.

  • Optional and shopping signals: 0 points (for now)

    They are surfaced because they matter operationally, and they are likely candidates for expanded scoring as standards mature.

Important Note

This space is fast evolving. These checks while may be proxies, are still important indicators, but they are not necessarily the only signals of Agentic eXperience readiness. Check back soon for more details as protocols, tools and practices continue to evolve.