Opens in a new tabSkip to content
Agent LighthouseAgent Lighthouse

    Searches the text of every published page. The evidence sources themselves are not in this index — search all of them on the trusted sources page.

    GitHub ↗
    Browse checks and page contents
    agent-interfaces/search-endpoint

    Site search reachable by agents

    What it checks

    One site-search audit over both halves of the declaration: the Schema.org SearchAction URL template, and a GET search operation in the OpenAPI spec.

    State Result
    a SearchAction URL template whose probe returns HTTP 200 with results, or a GET operation on a path with a search segment in /openapi.json pass
    a SearchAction template that 200s with an empty payload, is gated (401/403/407/451), returns another non-200, or is unreachable warn, priority low
    a SearchAction whose target has no urlTemplate or no {placeholder}, or a WebSite node that declares no SearchAction at all warn, priority low
    neither half present fail, priority low

    Why it matters

    An AI agent reads a site’s WebSite → potentialAction → SearchAction urlTemplate (or a GET /search operation in its OpenAPI spec), substitutes the query term, and fetches results instead of crawling the site.

    Evidence

    • SearchAction is a stable schema.org type (“The act of searching for an object”), used as a potentialAction on WebSite with an EntryPoint urlTemplate carrying the query placeholder; schema.org’s Google-index aggregation reports adoption on 10M+ domains — schema.org/SearchAction (verified 2026-08-21)
    • Google retired the only documented consumer of that markup, the sitelinks search box, in October 2024 (“Farewell, Sitelinks Search Box”); the feature no longer appears in Search results and its documentation was archived — developers.google.com/…/sitelinks-search-box (verified 2026-08-21)
    • The OpenAPI half of the signal depends on an agent obtaining the spec at all, which is documented only for developer-registered documents (GPT Actions, Copilot API plugins) — developers.openai.com/…/getting-started (verified 2026-08-21)

    Limits

    No crawler or agent documentation from OpenAI, Anthropic, Google, Microsoft or Perplexity states that a named agent reads SearchAction to query a site, rather than crawling it or using a general web-search tool. The documented server-side tools those vendors ship — Anthropic’s web_search and web_fetch, for example — query the open web, not a site’s declared search template. High markup adoption therefore reflects legacy SEO practice, not proven agent consumption — a community convention with a plausible but unverified mechanism.

    How it scores

    SearchAction is a ratified schema.org term with very large adoption. But its one documented consumer was retired by Google in 2024, and no vendor documents a named AI agent that reads it. The OpenAPI fallback inherits the unproven discovery leg graded in 5.1.

    Sources