PRACTICAL ACCESSIBILITY TESTING
How to record useful manual accessibility test evidence
Updated October 1, 2026
A reviewer note should help another tester repeat the check. “Looks fine” does not say which page state, interaction or assistive technology was used. Keep the automated scan result and the human assessment distinct.
Start with a defined scope
Record the page URL, date, tester, browser, operating system and assistive technology with versions. Identify the state: for example, a contact form before submission, a validation error or an open dialog. Do not assume an initial-page scan exercised these interactions.
Check different ways of using the page
- Keyboard: navigate and operate the relevant controls without a mouse; check that users can leave menus and dialogs.
- Focus: follow its order and visibility, including after opening or closing a dialog.
- Screen reader: check useful control names, roles and states, headings, instructions, and relevant error or status information.
- Pointer targets: check the clickable area and space between controls, not just the label. Confirm that users can select the intended control without accidentally activating a neighbor. For WCAG 2.2 AA, review the 24 by 24 CSS pixel minimum and its spacing and other exceptions in Target Size (Minimum), 2.5.8.
- Dwell clicking and on-screen keyboard: with an actual dwell-click tool enabled, hold the pointer over a control until the tool activates it. Confirm the intended action occurs once, including in menus and dialogs. Enter form text with an on-screen keyboard and check that fields and actions remain reachable. Hovering with an ordinary mouse is not a dwell-click test. Record the tool, version and dwell settings; if unavailable, leave the check Not reviewed.
- Zoom and reflow: use actual browser zoom at 200% and 400%. Check that text, buttons, errors and dialogs remain visible and operable without clipping or overlap. At a 1280 CSS pixel starting width, 400% zoom corresponds to a 320 CSS pixel viewport. Check ordinary vertically scrolling content without two-directional scrolling; document essential two-dimensional content separately. See Reflow, 1.4.10. A narrow viewport alone does not verify browser zoom.
These are starting points, not an exhaustive audit. W3C's Easy Checks explains preliminary accessibility review and its limits.
A reusable evidence record
Page and state:
Tester and date:
Browser / OS / assistive technology and versions:
Input method and device (keyboard, pointer, touch, dwell tool, on-screen keyboard):
Input settings (for example, dwell delay):
Browser zoom and starting viewport dimensions:
Check and scope:
Steps to reproduce:
Expected behavior:
Actual behavior:
Outcome: Pass / Fail / Not applicable / Not reviewed
Evidence reference:
Retest date and result:
Write an outcome that another person can verify
Fictional example: “Contact form, error state. Tab to the empty email field, activate Submit, then return to the field. Expected: the error is available with the field. Actual: only the field name is announced; the visible error is not associated. Fail for this check in the recorded browser/screen-reader combination.” Record your actual observations and environment rather than copying this example as test evidence.
Record a separate result for each input method
A successful keyboard check does not establish that pointer or dwell activation works. Record each method, page state and outcome separately. Keep checks Not reviewed until performed; describe a blocked check and its reason rather than marking it Pass. These interaction checks require human observation even when an automated rule reports no finding.
Keep reviewer decisions separate from automation
In a saved GatekeeperQA report, use Saved tester review to record notes and outcomes. Download a reviewed copy to retain that evidence. A saved note does not change the original automated result, and a manual Pass applies only to the stated scope.
After a fix, repeat the same steps, rescan where relevant and retain the original and retest records. If a component was absent or could not be reached, record that limitation instead of declaring a pass. Keep credentials and personal form values out of shared evidence.
Read the report interpretation guide or practice with fictional reviewer notes.