Windows Narrator and Edge With Data Grids

Narrator is the screen reader built into Windows. It is rarely the primary tool of experienced screen reader users, who mostly use NVDA or JAWS, but it is the one available on every Windows machine without installation — on locked-down corporate laptops, on shared devices, and for people who have recently lost sight. Its table and grid behaviour in Edge is close to NVDA’s and JAWS’s in principle and different in the details: its own keys, its own “scan mode”, and its own verbosity choices.

This page describes how Narrator reads data tables and grids in Edge, how to test with it, and which differences are worth fixing. It belongs to assistive technology behaviour differences.

Spec reference

Narrator uses the UI Automation (UIA) accessibility API, which Edge and Chrome implement. It reads the same roles, names, states and relationships as other screen readers, mapped through UIA patterns:

Tables and grids expose the UIA Table and Grid patterns; header relationships come from th/scope, headers, or columnheader/rowheader roles.
aria-sort maps to UIA properties Narrator reads on header cells.
Live regions map to UIA LiveRegionChanged events; Narrator honours polite and assertive.

Narrator’s scan mode (toggled with Caps Lock+Space or Insert+Space) is its browse mode: arrow keys read the page. It turns off automatically on many interactive controls and composite widgets; when it is on, arrow keys do not reach the page’s scripts.

Criteria are the same as for any screen reader: SC 1.3.1, 4.1.2 and 4.1.3 are where Narrator differences show up.

Same semantics, different keys and verbosity — the markup is shared; the test script is not.

When Narrator testing matters — and how much

Include Narrator in the audit matrix for products used on managed Windows fleets — enterprise tools, government services, education platforms — where installing NVDA may be blocked, and in any product whose users include people new to screen readers.

It does not need to be in every per-pull-request smoke test. A per-release manual pass over the critical data journeys is usually enough, because most Narrator failures share causes with NVDA and JAWS failures.

The misapplication to name is treating Narrator differences as bugs to work around with Narrator-specific markup. The fix for a Narrator problem is almost always correct, standard semantics; Narrator-specific hacks tend to break NVDA or JAWS.

Annotated code example

Narrator + Edge test script for an interactive data grid
1. Ctrl+Win+Enter              → Narrator on
2. Load the page in Edge; press H until "Open invoices" heading is read
3. Tab into the grid           → expect: "Open invoices, grid, …, INV-1042, row 1, column 1"
                                 (scan mode should switch off; if not, Caps Lock+Space)
4. Right Arrow ×2              → expect: "Due, 2026-03-04"  (column header then value)
5. Down Arrow                  → expect: "row 2, 2026-03-09"
6. Tab to the Amount sort button, Enter
                               → expect: "Amount, sorted ascending" then the status message
7. Type in the filter field    → expect: one count message after typing stops
8. Caps Lock+Space (scan mode on); Ctrl+Alt+Down
                               → expect: table navigation works in scan mode for a static table
Record: exact strings, any missing header, any extra "clickable" or "group" noise

<!-- What makes step 3 work: a named grid with proper header roles -->
<div role="grid" aria-labelledby="inv-h" aria-rowcount="241">
  <div role="row" aria-rowindex="1">
    <div role="columnheader">Invoice</div>
    <div role="columnheader" aria-sort="none"><button type="button">Amount</button></div>
  </div>
  <div role="row" aria-rowindex="2">
    <div role="rowheader" tabindex="0">INV-1042</div>
    <div role="gridcell" tabindex="-1">1,280.00</div>
  </div>
</div>

Narrator tends to announce “clickable” for elements with click handlers but no role — a useful signal that a <div onclick> has crept into a grid. Treat “clickable” in a Narrator transcript as a finding.

Keyboard & AT behaviour

Key	Narrator result	Difference from NVDA/JAWS
`Caps Lock+Space`	Toggle scan mode	NVDA: `Insert+Space`; JAWS: `Insert+Z` or automatic
`Ctrl+Alt+Arrows` (scan mode, table)	Move by cell with headers	Same keys as NVDA and JAWS
`T` (scan mode)	Next table	Same as NVDA; JAWS also `T`
Tab into `role="grid"`	Scan mode usually turns off	NVDA/JAWS switch reliably; confirm with Narrator
Polite status update	Read after current speech	Narrator sometimes reads it before finishing the focused item
`Caps Lock+F1` on an element	Detailed element info	Useful for checking names and roles

A ten-minute pass per release catches the Narrator-specific issues that matter.

Integration context

Narrator’s scan mode is the same concept as NVDA’s browse mode and JAWS’s virtual cursor; the implications for grids — why arrow keys sometimes never reach your handler — are explained in JAWS virtual cursor versus forms mode in data grids. Sort announcement differences across readers are compared in NVDA versus JAWS aria-sort announcement differences.

Whether a component should be a grid at all — and therefore whether mode switching happens — is covered in choosing between grid and table roles.

If NVDA or JAWS would also be affected by the same cause, it is a fix; otherwise it is a note.

Gotchas

Chrome versus Edge. Narrator works in Chrome too, but Edge is the default on Windows and the combination most Narrator users have. Test Edge first.

Narrator versions. Narrator improves with Windows feature updates; record the Windows build with every transcript.

Keyboard layout. The Narrator key can be Caps Lock, Insert, or both depending on settings. Write test scripts that mention both.

Design system notes

Add a Narrator + Edge column to the component library’s AT support matrix, with the date and Windows build of the last check. Flag components where Narrator needs a manual scan-mode toggle so product teams can mention it in help content if needed.

Testing checklist

Critical grids have been tested with Narrator in Edge this release.
Entering a grid turns scan mode off, or the need to toggle it is documented.
Column and row headers are announced when moving between cells.
No grid cell is announced as "clickable" without a role.
Sort and filter status messages are read once.
Transcripts record the Windows build and Edge version.

FAQ

Should data grids be tested with Windows Narrator?

Yes, at least per release for products used on managed Windows devices or by people new to screen readers. Narrator is available on every Windows machine without installation.

What is Narrator's scan mode?

Narrator’s equivalent of NVDA’s browse mode or JAWS’s virtual cursor. Arrow keys read the page instead of reaching its scripts. It is toggled with Caps Lock plus Space and usually turns off automatically in interactive widgets.

Why does Narrator say "clickable" on my grid cells?

Because the elements have click handlers but no interactive role. Give cells and controls proper roles — gridcell, button — so Narrator and other readers announce what they are.

Do I need Narrator-specific markup?

No. Fix differences by making the markup correct and standard; Narrator-specific workarounds tend to break other screen readers.

JAWS virtual cursor versus forms mode — the JAWS equivalent of scan mode
NVDA versus JAWS aria-sort — sort announcements compared
Choosing between grid and table — why mode switching happens

← Back to Assistive Technology Behaviour Differences