Define the conversion journeys to test

List the shortest keyboard path from landing page to meaningful outcome: open navigation, choose service, open quote dialog, complete form, review errors and submit. Include sticky headers, cookie preferences, appointment widgets and mobile menus. Capture the expected next focus target for every transition, especially when content is inserted or removed.

Prioritize flows by commercial importance and interaction complexity. A brochure page with ordinary links may be low risk; a multi-step quote modal with nested dropdowns is high risk. Use the same browser viewport and input method each time so regressions can be reproduced.

Run the keyboard-only script

Start with the mouse untouched. Press Tab and Shift+Tab through visible controls and verify the focus indicator remains visible, the order follows the task, and no essential action requires hovering. Activate controls with Enter or Space as appropriate. Confirm that disabled-looking controls are actually disabled and that visually hidden links are not unexpectedly focusable.

Test on a desktop browser and a narrow viewport with the responsive menu active. A menu that is closed visually but still tabbable is a common failure. Similarly, a fixed banner can cover the focused control even though focus order is technically correct. Record both behavior and screenshots of the failing state.

Verify modal focus as a state machine

When a modal opens, focus should move inside it to an appropriate element; Tab and Shift+Tab should remain within an active modal; Escape should close it when appropriate; and focus should generally return to the triggering control. The W3C dialog pattern also calls for a clear accessible name and for background content to be inert while the modal is active.

Test the less obvious states: validation error inside the modal, submission success, asynchronous loading, nested picker, and trigger removed after success. If the trigger no longer exists, move focus to a logical next point such as the success heading. Never mark a dialog modal in ARIA while allowing keyboard users to interact with the obscured page.

Treat form errors as navigation problems

Submit the form with required fields empty and with a malformed email address. Confirm that error text identifies the field and explains the correction, that focus reaches the first actionable error or error summary, and that typed values survive correction. Re-test with browser autofill and a screen reader when possible.

Example: a three-step form opens step two but leaves focus on a button removed with step one. The visitor may think the form froze. Move focus to the new step heading or first relevant field, announce the state appropriately and preserve a visible progress indicator. A small focus fix can remove a large barrier without changing the offer.

Write regression tests developers can execute

Create a test matrix with URL, viewport, control, starting focus, action, expected focus, observed focus, pass or fail and severity. Include a case for Escape, a case for return focus and a case for each validation state. Automated checks can flag missing labels or some focus problems, but they do not replace manual keyboard traversal.

Add these tests to the release checklist for changes to navigation, modal libraries, forms, cookie banners and third-party booking widgets. Where possible, use stable test selectors and an end-to-end script for focus transitions, while retaining a manual check for visible focus and usability.

Measure business impact without reducing access to a metric

Monitor form starts, completion, validation failures and support complaints before and after fixes, segmented only where data is appropriate. A reduction in abandonment can support a CRO case, but accessibility is not justified solely by whether a conversion lift is statistically detectable. Fix barriers even in low-traffic paths.

Connect this playbook to the broader website accessibility and multi-step form audits. Use the website audit checklist to keep findings visible alongside performance work. A conversion flow is not finished until visitors can operate it reliably without a mouse.