Perceivable · Operable · Understandable · Robust
Accessibility rests on four principles — Perceivable, Operable, Understandable, Robust. Here's what each means and the success criteria every team should know.
Information and UI components must be presented in ways users can perceive — through sight, sound, or touch. If users can't sense it, they can't use it.
UI components and navigation must be operable. If something requires a mouse with pixel-perfect aim or a steady hand, large portions of users are locked out.
Information and the operation of the UI must be understandable. Predictable behavior and plain language matter as much as visual design.
Content must be robust enough to be reliably interpreted by user agents — including assistive technologies. Today's accessible page must still be accessible to tomorrow's screen readers and devices.
Text against its background must reach 4.5:1 for normal text or 3:1 for large text (18pt+, or 14pt+ bold). Anything below makes copy unreadable for low-vision users.
Every non-text element needs a text alternative that conveys the same meaning. Decorative images use empty alt; informative images describe the content.
Every interactive element must be reachable and operable using a keyboard. No mouse-only menus, no drag-and-drop without a fallback, no hover-only reveal of critical info.
Keyboard focus must always be visible. Removing the default outline without providing a clearer replacement strands keyboard users — they literally cannot see where they are.
Pointer targets must be at least 24×24 CSS pixels (AA, WCAG 2.2) — 44×44 for AAA. Tiny tap targets make mobile sites useless for anyone with low precision or large fingers.
Every UI component must expose its name, role, and current state to assistive technology. Use the right element for the right job — <button> for actions, <a href> for navigation, <label> for inputs. ARIA only when native HTML can't express it.
AI is becoming both a way to generate accessibility metadata at scale and the assistive layer itself — neither replaces human review.
Vision and speech models can describe images and transcribe audio at scale. But the output needs review — a confident, wrong caption is worse than none at all.
Screen readers, live captions, and "explain this page" are becoming AI features. Ship clean semantics so the model has accurate structure to work from.