Frames are titled
What it checks
Agents need a title to understand what each iframe contains. Untitled or duplicate-titled frames are opaque embedded contexts.
Why it matters
The title attribute supplies an <iframe>’s accessible name, so an agent reading the accessibility tree sees iframe "Checkout payment form" instead of an anonymous iframe node and can choose which embedded context to enter.
Evidence
- MDN states: “People navigating with assistive technology such as a screen reader can use the
titleattribute on an<iframe>to label its content… Without this title, they have to navigate into the<iframe>to determine what its embedded content is” — developer.mozilla.org/…/iframe (verified 2026-08-21) - axe rule maps this to WCAG 4.1.2 Name, Role, Value (Level A) with impact “serious”; the stated consumer is screen reader users, who “can access a list of all frame titles on a page” — dequeuniversity.com/…/frame-title (verified 2026-08-21)
- HTML Accessibility API Mappings, which specifies the
title-to-accessible-name mapping for frame elements, is still a W3C Working Draft (05 August 2026) — w3.org/…/html-aam-1.0 (verified 2026-08-21) - Accessible Name and Description Computation 1.1 (W3C Recommendation) defines the
title-attribute fallback step in the name computation — w3.org/…/accname-1.1 (verified 2026-08-21)
Limits
Playwright’s ARIA snapshot documentation covers roles, names, values and text content, but never mentions frames — playwright.dev/…/aria-snapshots (verified 2026-08-21). The Chrome DevTools MCP tool reference contains no iframe handling in its snapshot description, only per-element uids — github.com/…/tool-reference.md (verified 2026-08-21). The listed consumer is a human using a screen-reader frame list, a navigation affordance an agent does not need: it receives the frame’s contents (or a per-frame reference) directly. Cross-origin third-party embeds — the dominant real-world failure — are unremediable by the site owner and undriveable by the agent regardless of their title, so the signal cannot change the agent outcome there.
How it scores
The name computation and the screen-reader consumer are ratified and documented, but the agent-side dependence is unproven. The documented agent snapshot tools address elements by per-element reference or uid, and neither documents frame titles as part of what the model reads. An agent can also read a frame’s contents directly rather than choosing frames by name.
Sources
- MDN — the iframe element — MDN, vendor-doc (verified 2026-08-21)
- axe rule — frame-title — Deque, vendor-doc (verified 2026-08-21)
- HTML Accessibility API Mappings 1.0 — W3C, draft-spec (verified 2026-08-21)
- Accessible Name and Description Computation 1.1 — W3C, spec (verified 2026-08-21)
- Live Chromium accessibility snapshot of a probe page (own experiment) — Own experiment via Playwright MCP + Chromium, study (verified 2026-08-21)
- Chrome DevTools MCP — tool reference — Google / Chrome DevTools, repo (verified 2026-08-21)