Interactive elements carry accessible names for AI agents

This check is about agent operability, not search citation. AI agents that operate a browser on a user's behalf navigate your UI the way screen readers do. OpenAI's developer guidance is explicit: "ChatGPT Atlas uses ARIA tags—the same labels and roles that support screen readers—to interpret page structure and interactive elements. To improve compatibility, follow WAI-ARIA best practices." Google's generative-AI guide likewise describes browser agents "inspecting the DOM structure, and interpreting the accessibility tree." An icon-only button with no accessible name is invisible to those agents: a user who asks their assistant to sign up, check pricing, or run your product hits a wall your human visitors never see. The same fix serves disabled users — accessibility work now does double duty.

Give every interactive element an accessible name. Native semantics count — <button>Submit</button> is already named; the failures are icon-only buttons, unlabeled inputs, and clickable divs.

// Icon-only button: name it
<button aria-label="Copy audit prompt">
  <CopyIcon aria-hidden="true" />
</button>

// Input: associate a label
<label htmlFor="email">Email</label>
<input id="email" type="email" />

For full coverage (roles, states, keyboard interaction), the Accessibility WCAG audit goes much deeper — this check covers only the naming floor agents need.

• ID: agent-accessible-interactives

• Severity: low

• What to look for: Scan component and page source under src/app and src/components (or the framework equivalent) for interactive element source sites: each distinct <button>, <a>, <input>, <select>, <textarea>, or element with an interactive role/onClick in the source code counts once, regardless of how many times it renders. For each source site, determine whether it has an accessible name: visible text content, aria-label, aria-labelledby, an associated <label>, alt on an image button, or a title as last resort. Native semantics count — a <button> with text children is named. Count-and-report: "X of Y interactive element source sites have accessible names."

• Pass criteria: Count all interactive element source sites and the subset with accessible names. At least 90% must have accessible names (a house threshold, not a vendor number — vendors say "follow WAI-ARIA best practices" without quantifying). Report even on pass: "Y interactive source sites scanned, X named (Z%) — unnamed: [list or none]."

• Fail criteria: More than 10% of interactive element source sites lack accessible names, or any primary conversion control (sign-up, checkout, main CTA) is unnamed regardless of the percentage. Report: "X of Y interactive source sites lack accessible names (Z%, threshold 90%): [element — file]".

• Do NOT pass when: the percentage clears 90% but the main CTA is an icon-only or image-only control with no name — the elements agents most need are weighted by importance, not just count.

• Skip (N/A) when: The project has no interactive UI (static content site, API-only project). Note: "No interactive elements found — agent operability not applicable."

• Detail on fail: Frame as agent operability, never search visibility: "3 of 14 interactive source sites are unnamed icon-only buttons (copy, share, menu) — per OpenAI's guidance, Atlas-style agents interpret controls through ARIA labels and roles; these are invisible to them (and to screen readers)."

• Cross-reference: This check is the naming floor only. The Accessibility WCAG audit covers roles, states, keyboard interaction, and contrast in depth — fixes there satisfy this check automatically.

• Remediation: Name the unnamed controls:

  <button aria-label="Run the free Stack Scan">
    <PlayIcon aria-hidden="true" />
  </button>