Manual Accessibility Checks You Still Need
On this page
Before writing a single automated check, spend twenty minutes testing a page by hand. It shows you what accessibility bugs feel like from the user's side, and it covers the criteria no rules engine can judge. This chapter gives you three manual passes that find most real problems: keyboard, screen reader and zoom. None needs special software beyond what your computer already has.
Use any page you own, or Playwright's TodoMVC demo at https://demo.playwright.dev/todomvc/, which the later chapters scan automatically.
Pass 1: keyboard only#
Put the mouse out of reach and use the page with these keys:
| Key | Expected behaviour |
|---|---|
Tab / Shift+Tab
|
Move focus to the next / previous interactive element |
Enter |
Follow a link, press a button, submit a form |
Space |
Press a button, toggle a checkbox |
| Arrow keys | Move inside a widget: radio group, menu, tabs, slider, select |
Escape |
Close a dialog, menu or popup |
While you tab through, check four things:
-
Can you reach and operate everything? (WCAG 2.1.1 Keyboard, Level A.) A
<div>with a click handler is the classic failure: it looks like a button, but it is not focusable and ignoresEnter. Custom dropdowns, date pickers and drag-and-drop lists are the usual suspects. -
Can you always see where focus is? (2.4.7 Focus Visible, AA.) A CSS reset with
outline: noneand no replacement makes the page unusable from the keyboard. New in WCAG 2.2, 2.4.11 Focus Not Obscured also requires that the focused element is not completely hidden behind a sticky header, cookie banner or chat widget. -
Is the order logical? (2.4.3 Focus Order, A.) Focus should follow the visual reading order. Positive
tabindexvalues (tabindex="3") are the common cause of focus jumping around the page. -
Can you get out again? (2.1.2 No Keyboard Trap, A.) Open every dialog, then close it with
Escape. Focus should return to the control that opened it, not to the top of the page. Embedded video players and rich text editors are frequent traps.
On TodoMVC, type a todo and press Enter, then press Tab. Focus lands on "Mark all as complete" first and then on the new item's checkbox, which Space toggles. That order is fine. Now hover over the item: a delete button appears. Try to reach it with Tab. Whether you can, and whether you can see it when you do, is exactly the kind of finding this pass produces.
Safari note: by default Safari moves Tab focus only between text fields. Hold Option while tabbing, or turn on "Press Tab to highlight each item on a webpage" in Safari's Advanced settings, before you report a Safari keyboard bug. Chapter 8 handles the same difference in an automated test.
A "Buy now" control is a <div class="btn" onclick="buy()">. It looks and works fine with a mouse. What does the keyboard pass find?
A div is not focusable and has no keyboard behaviour. The fix is a real <button>, which brings focus, Enter and Space handling and the button role for free. Visible text alone does not make something operable.
Pass 2: screen reader basics#
You do not need to become a fluent screen reader user to find the big problems. Learn to start it, read the page and jump by headings and form fields.
macOS: VoiceOver is built in. Toggle it with Cmd+F5. Control+Option is the "VO" modifier: VO+Right Arrow reads the next item, VO+U opens the rotor, a list of headings, links, form controls and landmarks you switch between with the left and right arrows. Use Safari, which is the browser VoiceOver supports best.
Windows: NVDA is free from nvaccess.org. Once it runs, Down Arrow reads the next line, H jumps to the next heading, F to the next form field, B to the next button, and Insert+F7 lists elements. Use Chrome or Firefox.
Phones: VoiceOver on iOS and TalkBack on Android are both built in. Swipe right to move to the next element, double-tap to activate.
What to listen for:
- Every control announces a name and a role. "Email address, edit text" is good. "Edit text" alone means an unlabeled input. "Button" alone means an icon button without an accessible name.
-
Images say something useful. "IMG_2041.jpg, image" is noise; decorative images should be silent (
alt=""). -
Headings outline the page. Open the rotor or heading list: it should read like a table of contents, with one
h1and no skipped levels used just for styling. - State changes are announced. After you submit a form with an error, does the screen reader say anything? If the error only appears visually, a blind user never learns why nothing happened.
-
Landmarks exist.
header,nav,mainandfooterlet people skip straight to the content.
Screen readers differ from each other, and each pairs best with particular browsers. Test the combinations your users actually have; for most public sites that means at least VoiceOver with Safari and NVDA with Chrome.
Pass 3: zoom, reflow and text spacing#
Low-vision users zoom. Three checks:
-
Zoom to 200% (
Cmd/Ctrland+). Text must still be readable and nothing may be cut off or overlap (1.4.4 Resize Text, AA). - Zoom to 400% in a 1280 pixel wide window, which makes the viewport 320 CSS pixels wide. Content must reflow into one column, without scrolling sideways to read lines of text (1.4.10 Reflow, AA). Data tables, maps and toolbars are allowed exceptions.
- Increase text spacing. WCAG 1.4.12 requires that nothing breaks when users override line height to 1.5, paragraph spacing to 2 times the font size, letter spacing to 0.12 and word spacing to 0.16. Text in fixed-height boxes is what usually breaks.
Also turn on your operating system's "reduce motion" setting and check that large animations stop or soften, and check that information is never conveyed by colour alone, such as a red border as the only sign of an invalid field.
Turning a manual finding into a regression test#
Some manual findings can be pinned down with a plain Playwright test once you know what to check. Reflow is a good example: at a 320 pixel viewport, the document should be no wider than the window.
import { test, expect } from '@playwright/test';
// WCAG 1.4.10 Reflow: content at 320 CSS pixels wide without scrolling in two directions.
// 320px is what a 1280px-wide window shows at 400% zoom.
test.use({ viewport: { width: 320, height: 640 } });
test('todo app reflows at 320 CSS pixels', async ({ page }) => {
await page.goto('https://demo.playwright.dev/todomvc/');
await page.getByPlaceholder('What needs to be done?').fill('A todo with a fairly long title that has to wrap');
await page.keyboard.press('Enter');
const overflow = await page.evaluate(
() => document.documentElement.scrollWidth - document.documentElement.clientWidth,
);
expect(overflow).toBeLessThanOrEqual(0);
});
This passes on TodoMVC. It is a narrow check: it catches a page that grew a horizontal scrollbar, not text that is clipped inside a fixed-width box. That trade-off, a precise automated check for one symptom plus a human pass for the rest, is the pattern for the whole course.
Which manual check does WCAG 1.4.10 Reflow describe?
Reflow is about the 320 CSS pixel viewport. 200% zoom is 1.4.4 Resize Text, and line height overrides are 1.4.12 Text Spacing; both are separate criteria worth checking too.
Writing findings down#
Record each finding with the WCAG criterion, the steps to reproduce (including browser and assistive technology), what happened and what should happen: "NVDA with Chrome on Windows 11: the email field announces 'edit text' with no name (1.3.1, 4.1.2). Expected: 'Email address, edit text'." A finding written like that is something a developer can fix without asking you, and something you can turn into a test.
The next chapter adds the automated half: one line of Playwright that checks the whole page against dozens of rules.