All writing
Nov 04, 20268 min read

Testing a website without touching the mouse

A practical keyboard-only review that exposes missing focus states, awkward navigation, and interactions that only work for a pointer.

AccessibilityFrontendUX

A website can look finished and still become difficult to use the moment the mouse leaves the desk. The menu opens on hover. The focused link disappears against the background. A dialog closes visually but leaves the keyboard somewhere behind it.

None of those problems is obvious in a screenshot.

One useful review is deliberately simple: reload the page, put the mouse aside, and try to complete a real task with the keyboard. Find a project. Open the navigation. Send a message. Notice where the interface stops explaining itself.

This is not a complete accessibility audit. It is a focused check that can reveal problems before they become part of the handoff.

Start with a task, not a tab count

Pressing Tab until something looks wrong is better than nothing, but a task gives the review a purpose. On a portfolio, I might try to reach a case study, return to the writing section, and submit the contact form with an intentional mistake.

For ordinary page navigation, Tab moves between focusable elements and Shift+Tab moves backward. Enter activates links and buttons; Space activates buttons and operates controls such as checkboxes. Arrow keys have roles inside native controls and established widgets, not as a replacement for Tab everywhere.

Browser and operating-system settings can affect keyboard navigation. For example, Safari may need its setting for tabbing through webpage items enabled. Confirm the environment before deciding the page has no reachable links.

Write down the task, the browser, and the first point where progress becomes unclear. That is more useful than a general note saying that the page feels inaccessible.

Ask where focus went

At every step, I should be able to answer one question: what will respond if I press a key now?

A visible focus indicator provides that answer. It should remain distinguishable against the actual surface beneath it, including images, dark panels, and selected controls. A subtle colour shift that looks attractive in isolation may disappear in the page.

Removing the browser outline without a dependable replacement breaks this feedback. A shared focus treatment using :focus-visible can make keyboard interaction clear without requiring the same visual treatment for every pointer click.

Focus also needs to remain visible in context. A sticky header can cover a newly focused link. An overflow container can clip an outline. A cookie banner can sit over the control the browser has scrolled into view. The CSS rule may exist while the visitor still cannot see the result.

I check forward and backward navigation because some mistakes are easier to notice on the return journey.

Let the document order do the work

A sensible focus order usually follows the meaningful reading order of the page. Header controls come before the main content, related controls stay together, and a form proceeds through its fields in a sequence that makes sense.

Visual rearrangement can hide a mismatch. A grid may place a call to action beside a heading while its markup puts it much later. Keyboard focus then appears to jump across the page for no clear reason.

I fix the underlying structure before reaching for tabindex. Positive values create a separate ordering system that becomes fragile as controls are added. tabindex="0" makes an element reachable, but it does not turn it into a button. tabindex="-1" can support deliberate programmatic focus, such as moving to a section heading, without adding another stop to the normal Tab sequence.

Not everything should be a tab stop. Paragraphs and ordinary headings are readable content, not interactive controls. Making each one focusable creates extra work rather than extra access.

Use the element that already knows its job

If something navigates, I expect a link. If it changes the current interface or submits a form, I expect a button.

A clickable div may respond to the mouse while missing keyboard activation, an accessible role, and built-in disabled behaviour. Adding one key handler rarely reproduces everything the native element provides.

This is also why I test with both Enter and Space where appropriate. A custom control may work with one key only because someone patched the most obvious case.

Native HTML is not a guarantee that the whole interaction is accessible, but it removes unnecessary implementation work. A real button still needs a useful name. A real input still needs a label. The starting point is simply stronger.

Open the things that usually hide

Menus, disclosures, and dialogs deserve their own pass because the important behaviour happens between states.

For a simple navigation disclosure, I check that a button opens it, its expanded state is communicated, and its links become reachable. Closed content should not leave invisible controls in the Tab sequence. Calling an ordinary navigation list a menu can introduce expectations for more specialised keyboard behaviour, so I avoid that role unless the interaction genuinely needs it.

A modal dialog has different responsibilities. Focus should move into it at a sensible location, remain within it while it is modal, and usually return to the opening control when it closes. The background should not remain independently interactive. Escape should normally dismiss the dialog, and a visible close control should still be available.

A focus trap inside an open modal is intentional. A focus trap inside an ordinary page section is a defect. The distinction is whether the interaction requires a temporary boundary and provides a clear way out.

Make a mistake in the form

A successful submission tests only one path. I also leave a required field empty, enter an invalid value, and try again.

Can I reach every field and understand its label? Does the error explain the correction? Is the message associated with the relevant input? If focus moves after submission, does it land somewhere useful rather than returning to the top of the page?

Errors should not depend on a red border alone. Submission progress and success also need to be communicated without requiring someone to notice a visual change elsewhere. A status message may need an appropriate live region; not every update needs to steal focus.

When the server fails, the form should preserve the message I wrote. Keyboard access is not helpful if recovery means starting the entire task again. I cover that wider responsibility in Forms are product work, not afterthoughts.

Check the shortcuts through repeated content

A page can be technically navigable and still exhausting. Repeating a long header before every main-content task is one example.

A skip link gives keyboard users a direct route past repeated navigation. I check that it becomes visible on focus and that activation actually moves the navigation position to the main content, rather than merely changing the URL fragment. Its target must not disappear behind a fixed header.

Repeated links inside cards deserve attention too. An image, title, and “Read more” link may all lead to the same article. That is not automatically wrong, but it can create unnecessary stops. One clear primary link is often enough. Any linked icon or image also needs an understandable accessible name.

The goal is not the fewest possible tab stops. It is a sequence where each stop earns its place.

Turn the findings into repeatable checks

I record failures as small reproduction steps:

  • Open the mobile navigation with the keyboard.
  • Close it using its toggle.
  • Press Tab.
  • Observe that focus enters a now-invisible navigation link.

That gives the fix a clear acceptance condition: closed navigation items must no longer be reachable, while the toggle remains usable.

If the defect lives in a shared component, I check its other callers after fixing it. A focused interaction test can protect the repaired behaviour. Automated accessibility tools can catch additional issues, but they do not replace trying the task and judging whether the sequence makes sense.

Keyboard testing also does not replace screen-reader testing. Accessible names, reading order, announcements, zoom, contrast, touch input, and other needs require their own review. A smooth Tab sequence is evidence about one important part of the experience, not proof of overall accessibility.

A finished interface works beyond the screenshot

I would include this pass in a practical launch checklist, but I would not save it for launch day. Checking a menu or form while building it makes the behaviour easier to correct before it spreads through the site.

Put the mouse aside. Complete a meaningful task. Keep track of focus, open and close every layer, and deliberately take the error path.

The result should not feel like a special version of the website. It should feel like the same website, with another dependable way to use it.

Continue the conversation

Get the next field note.

Occasional writing about development, systems, and better web decisions.