Action-Oriented API

The Action-Oriented API parameter checks whether your site exposes API endpoints that let an AI agent do something — search, submit, book, order — instead of only reading content. friendly4AI counts the distinct action endpoints it finds: three or more scores pass (100), one or two scores partial (50), and none scores fail (0).

What does this parameter measure?

It measures one thing: can an AI agent act on your site through callable endpoints, or can it only read? friendly4AI pulls candidate endpoints from three places:

• Homepage HTML links — <a href> paths that contain an action keyword: /search, /submit, /book, /reserve, /order, /checkout, /create, /update, /delete, /apply, /register, /login, /signup.
• Inline method examples — text in the page body such as POST /api/submit.
• OpenAPI spec (when prefetched) — paths whose path, summary, or description names one of those actions.

Before counting, the scanner de-duplicates what it finds. A documented OpenAPI path and a homepage link pointing at the same action count once, not twice.

Why does it matter for AI-readiness?

Read-only access lets an assistant describe your site. Action endpoints let it finish the job for the user. Give an agent a callable /search, /book, or /order and it can run the whole sequence — find a result, reserve it, confirm it — rather than dumping the user back on a manual form. Well-named action endpoints turn your site from a thing AI reads into a service it can operate for a customer.

How is it scored?

This is a gradient AI-Specific Signals parameter. The scanner counts distinct action endpoints in the de-duplicated set it builds across HTML and spec sources:

• Pass (100) — 3 or more distinct action endpoints detected.
• Partial (50) — 1 or 2 distinct action endpoints, or unclear/mixed signals.
• Fail (0) — no action endpoints detectable in the HTML or the spec.

Scoring here is API-conditional. The weighted contribution counts only when an API is detected on the site via api-discoverability. With no API present, the effective weight drops to zero and the parameter is excluded from the score, so a content-only site is never penalized for lacking action endpoints.

How do you fix a low score?

• Expose at least three clearly named action endpoints — say /api/search, /api/book, and /api/submit.
• Put action verbs in the path or the OpenAPI summary so the scanner can tell what each endpoint does.
• Document every action in your OpenAPI spec with a descriptive summary and description — see api-schema-quality-for-agents.
• Stuck at one or two actions? Add the obvious next ones. A search endpoint plus a contact/submit endpoint already covers most sites.
• Re-scan after you publish to confirm three or more distinct actions show up.

Related parameters

• api-discoverability
• capability-metadata
• api-schema-quality-for-agents