How to test your site with a screen reader — five keys and ten minutes

Automated checks like Lighthouse only catch what a formula can catch: is the attribute there, is the contrast high enough. They won't tell you that your button is called "button," or that your form announces as "edit text, edit text, edit text."
You can only hear that. And you don't need to "learn to use a screen reader" — you need five keys and ten minutes.
One warning up front that will save you an hour of panic: first learn how to turn it off. The first thirty seconds feel like losing control of your computer. That's normal, and it passes.
Step 1. Pick the one that fits your system
- Windows — NVDA. Free, installs in a minute, and it's the primary screen reader for 37.7% of users. Download it from nvaccess.org.
- macOS — VoiceOver. Already installed, nothing to download.
- Phone — also an option, and an honest one: 91.3% of screen reader users run one on mobile.
When installing NVDA, uncheck "start automatically on log-on" — otherwise it greets you at every boot and that will surprise you.
Step 2. Turn it on, then immediately practise turning it off
- NVDA: start with
Ctrl + Alt + N. Quit withNVDA + Q(the NVDA key is Insert by default; you can remap it to Caps Lock in settings, which is handier). - VoiceOver:
Cmd + F5. On newer Macs, triple-press Touch ID. Same shortcut turns it off.
Toggle it on and off three times until it feels routine. Now you can work.
Step 3. Learn five commands and skip the rest
For NVDA on Windows:
H— jump to the next heading. This is the main one: 71.6% of users find information this way.Tab— next interactive element: link, button, field.Ctrl— shut up. You'll use this constantly.NVDA + F7— the Elements List: every link, heading and landmark on the page in one window. Your X-ray.↓— read the next line straight through.
For VoiceOver on macOS instead:
Ctrl + Option— the "VO keys," the modifier for every command.VO + →— next element.VO + U— the rotor: a list of headings, links, images and form controls. ← → switch the type, ↑ ↓ pick an item.Ctrl— shut up.- Inside the rotor you can just type to filter — press "2" in the headings list and only level-two headings remain.
Step 4. Walk the site in three passes
Don't listen to everything in order — nobody does. Do three short passes, each with its own question.
Pass 1 — headings. Press H (or open the rotor) from top to bottom. You're hearing your page's table of contents. The question: does it make clear what's on the page and in what order? The classic finding is that there are no headings at all, because big bold text was used instead. Which tags actually create structure is covered in the document tree.
Pass 2 — interactive elements. Press Tab to the end of the page. Every element needs a sensible name. If you hear "button" with no name, "link, read more" five times in a row, or "image, logo-final-2," write it down.
Pass 3 — the form. Reach a form and fill it in. Every field should announce with its label: "Email, edit text." If all you hear is "edit text," the field has no <label> and the person is filling it in blind. Check your contact form the same way.
Step 5. Know about modes, or you'll think it's broken
This is the wall every NVDA beginner hits.
You press H to jump to a heading — and the letter "H" gets typed into the input box. Nothing is broken: it switched into focus mode, where keystrokes pass straight to the browser. Otherwise you could never type anything.
- Browse mode — the default.
H,Taband arrows work as navigation commands. - Focus mode — turns on by itself when focus lands in a text field or a complex widget. A high beep marks the switch in, a lower one the way back.
- Toggle manually:
NVDA + Space.
There's a useful side effect: if you hear that beep somewhere you didn't expect it — on a plain link, say — some element carries a role it shouldn't. That's a markup bug, and you just found it by ear.
What you'll end up with
Ten minutes, and you have three to five concrete items: "menu button has no name," "email field has no label," "three level-one headings in a row." Not an abstract accessibility score — tasks, each fixable with a line of markup.
One more thing: after a test like this you'll stop trusting green checkmarks. Automated tools honestly verify that attributes exist — not that what's inside them means anything. A good companion to your pre-launch checklist, right next to the speed check.
Do I have to be blind to test this way?
No, but remember: you can see the screen and you'll unconsciously help yourself with your eyes. It's more honest to turn away from the monitor or switch it off. Better still is getting real screen reader users onto your project.
Does this replace Lighthouse and axe?
No, they complement each other. Automation quickly finds bulk formal errors; your ears find meaningless names, broken order and focus traps.
How do I test on a phone?
iPhone: Settings → Accessibility → VoiceOver, best mapped to a triple-press of the side button. Android: TalkBack, enabled by holding both volume keys. The gestures differ — swipe right is "next element," double tap is "activate."
The AI Architect 2027 course: step by step from computer basics to your own app and first customers. 543 short lessons on your phone.





