Hover-triggered tooltips, focus-triggered popovers, and sub-menus are the kind of component an AI design agent is routinely asked to emit — and exactly the kind that WCAG 2.2 Success Criterion 1.4.13 Content on Hover or Focus (Level AA) governs. This guide treats the criterion as a computable contract: a set of conditions an agent can partially check in markup and DOM, an honest list of what it cannot verify from static sources, and a clear record of what the underlying W3C documents do not establish.
Everything factual here comes from W3C documentation: the W3C Understanding page for SC 1.4.13, the W3C ARIA Authoring Practices Guide (APG) tooltip pattern, and the WCAG 2.2 quickref entry for this criterion’s techniques list. Where the sources are silent, this guide says so rather than filling the gap.
What the criterion actually says
The Understanding document states the criterion as follows:
Where receiving and then removing pointer hover or keyboard focus triggers additional content to become visible and then hidden, the following are true:
Dismissible: A mechanism is available to dismiss the additional content without moving pointer hover or keyboard focus, unless the additional content communicates an input error or does not obscure or replace other content;
Hoverable: If pointer hover can trigger the additional content, then the pointer can be moved over the additional content without the additional content disappearing;
Persistent: The additional content remains visible until the hover or focus trigger is removed, the user dismisses it, or its information is no longer valid.
Exception: The visual presentation of the additional content is controlled by the user agent and is not modified by the author.
The Dismissible condition contains two inline exemptions: content that communicates an input error, and content that does not obscure or replace other content. The Understanding page’s Intent section adds commentary on these — that input error messages “can persist without being dismissible, as they generally require the user’s attention, explicit confirmation or remedial action,” and that content appearing only over white space or purely decorative content does not interfere. This commentary is supporting context, not normative wording, and it matters for an agent because neither exemption has a testable definition in these sources (more on that below).
Scope: in, out, and adjacent
Three scope Notes in the Understanding document draw the boundary:
Note 1: Examples of additional content controlled by the user agent include browser tooltips created through use of the HTML title attribute.
Note 2: Custom tooltips, sub-menus, and other nonmodal popups that display on hover and focus are examples of additional content covered by this criterion.
Note 3: This criterion applies to content that appears in addition to the triggering component itself. Since hidden components that are made visible on keyboard focus (such as links used to skip to another part of a page) do not present additional content they are not covered by this criterion.
The Additional Notes add three boundary rules an agent should encode:
- User-agent-rendered content is out. The classic case is the browser’s rendering of the HTML
titleattribute. - Modal dialogs are out of scope because they must take keyboard focus and thus should not appear on hover or focus; the Understanding page refers to SC 3.2.1 On Focus.
- Hover-triggerable content should also be keyboard-triggerable, referring to SC 2.1.1 Keyboard. That is an adjacent criterion, not part of 1.4.13 itself.
One housekeeping note: the criterion’s errata block on the Understanding page lists two items (“Conforming the phrasing in the exception for consistency” and “Rewording the normative preamble for clarity”), but the corrected text is not given in these sources. The errata page is where corrections live; an agent should not assume the quotations above differ from the current normative text on that basis.
Four patterns, side by side
| Pattern | Dismissible | Hoverable | Persistent | Exception that applies |
|---|---|---|---|---|
Native HTML title tooltip |
not established by these sources | not established by these sources | not established by these sources | User-agent exception (Note 1: browser tooltips via the HTML title attribute) |
| Custom hover tooltip | Required unless the input-error or non-obscuring exemption applies (SC wording) | Required — F95 names this failure | Required until trigger removed, dismissed, or information no longer valid | No user-agent exception; the two Dismissible exemptions may apply |
| Custom focus popover | Same SC wording as above | Applies only “if pointer hover can trigger the additional content”; otherwise not established by these sources | Required per SC wording | Same as custom hover tooltip |
| Non-modal dialog | Covered per Note 2 (“other nonmodal popups”); dialog-specific behaviour not established by these sources | not established by these sources beyond the SC wording | not established by these sources beyond the SC wording | Note 2 covers nonmodal popups on hover/focus; modal dialogs are out of scope (SC 3.2.1) |
The pattern of the table is the pattern of the criterion: the three conditions attach to additional content that appears and disappears with hover or focus, and the user-agent exception removes author-rendered native tooltips from the picture entirely.
What an agent can check in markup and DOM
These are the checks an AI design agent can run against component code. Each is labelled DIRECT (a direct reading of the SC wording or its Notes) or PROXY (evidence that correlates with a condition, but not a rule the sources state as such).
- Hover/focus activation on the trigger — event listeners or CSS
:hover/:focusrules that reveal content. This establishes the component is in scope. PROXY (the CSS pseudo-class approach is itself listed only as a “Potential future technique” in the Understanding document). - A keydown Escape handler that hides the content — evidence of a dismissal mechanism that works without moving pointer or focus. PROXY (“such as pressing the Esc key” in the Intent is illustrative, not exhaustive; the APG Escape rule is work in progress).
- Container has
role="tooltip"; the trigger references it viaaria-describedby— the APG wiring. PROXY (APG pattern only; no claim is made about what assistive technology users hear). - The tooltip container is not focusable — per the APG statement that “Tooltip widgets do not receive focus.” PROXY.
- Showing the tooltip does not move focus off the trigger — per APG note 1 (“Focus stays on the triggering element while the tooltip is displayed”). PROXY.
- Focus-invoked tooltips dismiss on blur of the trigger — per APG note 2. PROXY.
- Hide logic is bound to exit from both trigger and content, so the content is itself hoverable — per APG note 3 and the Hoverable condition’s wiring. PROXY.
- Hover and focus activation both exist — a cross-reference to SC 2.1.1 Keyboard in the Understanding page’s Additional Notes. PROXY for that adjacent criterion, not for 1.4.13.
- The only “tooltip” present is the
titleattribute — out of scope. DIRECT (Note 1). - Hidden-until-focus skip links — not covered. DIRECT (Note 3).
- The component is modal or takes focus — outside this criterion; refer to SC 3.2.1 On Focus. DIRECT (Additional Notes).
Explicitly out of scope as checks: any delay or timeout value, any pixel, gap, overlap, z-index, or contrast threshold, any screen-reader output, and any conformance percentage. Do not infer pass or fail from the mere presence or absence of timers — no timeout rule exists in these sources.
What an agent cannot verify
This is the half of the contract that keeps an agent honest. The sources leave the following unset, and none of it should be treated as a requirement:
Timing for this component. Neither the Understanding page nor the APG tooltip pattern page defines a minimum or maximum delay for showing content (the APG says only that a tooltip “typically appears after a small delay” — no number), nor any dismissal timeout, auto-dismiss duration, or Escape-handling deadline.
Geometry and obscuring. The Understanding document says the Hoverable condition “generally implies that the additional content overlaps or is positioned adjacent to the target,” but specifies no gap, offset, edge-proximity threshold, or minimum hit-target size. There is no testable definition of when content “obscures or replaces” other content — no z-index, contrast, or overlap rule. And no rule covers a tooltip the pointer moves toward but never enters: Hoverable only requires the pointer can be moved over the content without it disappearing.
Assistive-technology output. The sources say nothing about how screen readers announce tooltips — no claim about announcement text, verbosity, or read order. The role="tooltip" / aria-describedby wiring is specified; what users hear is not.
Conformance thresholds. No percentage of instances, components, or pages must comply. SC 1.4.13 is stated in absolute, per-instance terms.
Dismissal-mechanism exclusivity. Escape is an example (“such as pressing the Esc key”), the APG specifies Escape for its tooltip pattern, but nothing prohibits other mechanisms — for instance, a close control reachable without moving pointer or focus.
Touch. The Understanding page for SC 1.4.13 and the APG tooltip pattern say nothing about touch, tap, long-press, or mobile interaction.
“No longer valid.” The Persistent condition allows content to hide when “its information is no longer valid” — a phrase the sources use without defining any trigger, timeout, or validation rule.
Errata text and APG status. The corrected errata wording is not in these sources, and the APG tooltip pattern is explicitly work in progress without task-force consensus, with no published example.
Where the APG tooltip pattern fits
The APG tooltip pattern is the closest thing to a reference implementation recipe — with a mandatory caveat, quoted verbatim from the page itself:
NOTE: This design pattern is work in progress; it does not yet have task force consensus.
With that caveat stated, the pattern’s rules are:
- Keyboard: “Escape: Dismisses the Tooltip.”
- Focus stays on the trigger: “Focus stays on the triggering element while the tooltip is displayed.”
- Blur dismissal: “If the tooltip is invoked when the trigger element receives focus, then it is dismissed when it no longer has focus (onBlur).”
- Hover persistence: “If the tooltip is invoked when a pointing cursor moves over the trigger element, then it remains open as long as the cursor is over the trigger or the tooltip.”
- Roles (two separate source bullets): “The element that serves as the tooltip container has role tooltip.” and “The element that triggers the tooltip references the tooltip element with aria-describedby.”
The pattern page also notes that “Tooltip widgets do not receive focus” and that “a hover that contains focusable elements can be made using a non-modal dialog.” It describes a tooltip as typically appearing “after a small delay” — no value given — and states that no example is published yet; the example work is tracked in issue 127 and the pattern discussions in issue 128.
Framed for an agent: the APG pattern is a useful, non-normative checklist that lines up with the three conditions — Escape maps to Dismissible, cursor-over-trigger-or-tooltip maps to Hoverable, and the blur and exit rules map to Persistent. It is not a WCAG mandate, and its consensus status is explicitly unsettled.
Techniques and failures that name the checks
The Understanding document’s Techniques section gives an agent three named anchors:
- SCR39: Making content on focus or hover hoverable, dismissible, and persistent — the sufficient technique, and effectively the whole contract in one name.
- F95: Failure of Success Criterion 1.4.13 due to content shown on hover not being hoverable — the named failure for a tooltip that vanishes when the pointer moves onto it.
- F114: Failure of Success Criterion 1.4.13 Content on hover or focus when content cannot be dismissed without moving the pointer or the keyboard focus — the named failure for a missing dismissal mechanism.
Two further items in the Understanding document are listed without IDs, each marked “(Potential future technique)”: a CSS technique (“Using hover and focus pseudo classes”) and an unnamed persistence failure (“content on hover or focus not remaining visible until dismissed or invalid”). The quickref entry for this SC is a different list, not a mirror: it adds ARIA: Using role="tooltip" and a “Failure to make content dismissible without moving pointer hover or keyboard focus”, both marked “(Potential future technique)”, and it omits F114. For how techniques relate to conformance generally — including that other techniques not listed may also suffice — see Understanding Techniques for WCAG Success Criteria.
FAQ
Does the HTML title attribute need to pass this criterion?
No. Note 1 places browser tooltips created through the title attribute under the user-agent exception: their presentation is controlled by the user agent and not modified by the author.
Is Escape mandatory for dismissal?
Not by the SC. The Intent says “such as pressing the Esc key” — illustrative, not exhaustive. The APG tooltip pattern does specify Escape, but that pattern is work in progress without task-force consensus, and nothing in the sources prohibits other dismissal mechanisms that work without moving pointer or focus.
Is there a maximum delay or dismiss timeout?
No. No source states any delay duration, dismissal timeout, or auto-dismiss value. The APG says only “typically appears after a small delay,” with no number.
Are skip links covered?
No. Note 3 excludes hidden components made visible on keyboard focus, such as links used to skip to another part of a page, because they do not present additional content.
What about modal dialogs?
Out of scope. The Understanding page states modal dialogs must take keyboard focus and thus should not appear on hover or focus, referring to SC 3.2.1 On Focus.
Must tooltips work on touch?
Not established by the Understanding page or the APG tooltip pattern. Neither addresses touch, tap, long-press, or mobile interaction.
Does aria-describedby guarantee screen-reader announcement?
No claim about announcement is made in these sources. The APG specifies the wiring; nothing in either source states what users hear, in what order, or at what verbosity.
Can static markup prove conformance?
No. The criterion is per-instance and absolute — no percentage threshold is defined. Markup checks are triage. The Dismissible exemptions (“obscures or replaces”), the Hoverable pointer-movement claim, and the Persistent “no longer valid” clause all require judgement or interaction beyond static code.
The Bottom Line
- Dismissible — partially checkable. An Escape handler is a PROXY for the mechanism; its absence is not failure, since “such as” makes Escape an example. The two exemptions need human judgement: “obscures or replaces” has no testable definition in these sources.
- Hoverable — least checkable from static markup. It is a claim about pointer movement. Binding hide logic to exit from both trigger and content is a PROXY; no gap tolerance between trigger and content is defined.
- Persistent — checkable only as binding. Hide must be tied to trigger or content exit, dismissal, or invalidity. “No longer valid” is undefined, and there are no timeout values to test against.
- Overall: markup checks are triage, not conformance. The criterion applies per instance, and the APG pattern that most closely resembles a recipe is work in progress without task-force consensus.
For an AI design agent, the practical contract is: emit hover and focus triggers, keep focus on the trigger, wire role="tooltip" and aria-describedby, bind Escape, keep the content hoverable by binding hide logic to exit from trigger and content alike, and flag every exemption judgement — input error, non-obscuring, invalidity — for human review rather than deciding it silently.
How This Guide Was Built
This guide is built from official WCAG documentation — the W3C Understanding page for SC 1.4.13, the W3C ARIA Authoring Practices Guide tooltip pattern page, and the WCAG 2.2 quickref entry for the techniques comparison. No hands-on testing was run — no tools, no audits, no measurements. All quotations are reproduced from the Understanding page or the APG tooltip page as labelled. The sources give no delay, timeout, or geometry values; where they are silent on a point — timing, screen-reader output, touch, obscuring definitions, “no longer valid” — this guide says so explicitly rather than supplying an assumption.
