Basics

What is a screen reader — and why it doesn't read the page top to bottom

Illustration: a stack of page sections you hop across, not walk through

Say "screen reader" and a picture forms: a robot droning through a page from header to footer while a person patiently listens.

Almost nobody does that.

WebAIM surveys screen reader users every couple of years — 1,539 responses in the latest round. The question: what do you do first when looking for something on a long page? The answers:

  • navigate through the headings — 71.6%;
  • use the find feature — 13.6%;
  • navigate through the links — 4.8%;
  • navigate through landmarks — 3.7%;
  • read through the page — 6.4%.

Six percent. Everything else is jumping around the structure. A screen reader isn't a narrator, it's a navigator.

What a screen reader is

It's software that takes what's on screen and delivers it as speech or to a braille display. Its users are mostly blind and low-vision people, but not only: it also helps with severe eye strain and some reading disabilities.

It doesn't work from a picture of the screen — it works from the document tree the browser builds. Which leads to the key consequence: a screen reader knows exactly as much as you wrote into the markup. A nicely drawn rectangle that "looks like a button" isn't a button to it. A real <button> tag is.

The common ones: JAWS (primary for 40.5%), NVDA (37.7%, free) and VoiceOver (9.7%, built into macOS and iOS). More than two-thirds of users keep several installed.

How people actually use it

Picture a book's table of contents you can flip through by voice. Press "H" — jump to the next heading, hear "heading level 2, Shipping," realise it's the wrong one, press again.

Heading levels genuinely matter here: 88.8% of respondents call them useful or very useful. So h2 and h3 aren't "font size" to a screen reader — they're depth in that table of contents. When AI picks a heading tag just because "it should look bigger," the outline falls apart.

Second surprise: speed. Experienced users listen at rates where you'd hear one continuous blur. At that speed, extra words aren't a detail. Twenty links in a row that all say "read more" become twenty identical noises.

Third: phones. 91.3% of screen reader users also run one on mobile — most often VoiceOver on an iPhone. So "mobile version" and "accessible version" are largely the same work; see mobile-friendly layout.

What a screen reader "sees" on your page

A simple thought experiment: strip your page down to text, headings, links, field labels and alt text. What's left is your site as a screen reader gets it.

Three things usually surface at this point:

  • An icon button with no text — just "button." No name.
  • A form field with no <label> — "edit text." What goes in it is anyone's guess.
  • An image with no description — "image," and sometimes the whole filename. What to put in alt text has its own breakdown.

Note that none of the three is visible to the eye. The page looks perfect.

What breaks most often

The same survey asked people to name their most painful items. The top five:

  1. CAPTCHA — images of text proving you're human.
  2. Interactive elements that don't behave as expected — menus, tabs, dialogs.
  3. Links and buttons whose names make no sense.
  4. Screens that change unexpectedly.
  5. Lack of keyboard accessibility.

A sobering list: almost none of it is fixed by "adding alt." It's about structure and behaviour. And almost all of it lives exactly where generated code cuts corners — worth a look before launch, next to your checklist.

What a screen reader doesn't do

There are plenty of myths, and nearly all of them overpromise.

  • It doesn't "look" at the screen. Image recognition exists as an add-on, but the foundation is markup. If the meaning isn't in the code, there's nowhere to get it from.
  • It doesn't fix bad markup. A clickable <div> doesn't become a button because it got announced. It stays an unnamed spot you can't reach with the keyboard.
  • It doesn't read text inside images. A screenshot of a table is just "image" to it. One more reason to ship data as text.
  • It doesn't translate. If the page doesn't declare its language, an English engine may read foreign text with English rules — the result is audio soup.

Which leads somewhere simple: accessibility isn't added later as a separate stage. It's either in the markup or it isn't, and no software will extract meaning you never wrote down.

How is a screen reader different from a voice assistant?

An assistant (Siri, Alexa) answers questions with its own content. A screen reader invents nothing: it speaks what's already in the interface and lets you move through it with keys or gestures.

Should I build a separate version of my site for blind users?

No — that's an outdated practice, and the separate version almost always falls behind on content. The right path is one site with decent markup: real buttons, headings in order, labels on fields.

Can I try a screen reader myself?

Yes, and it's the best way to get it. It's already installed on a Mac; on Windows you install NVDA for free. Half an hour of practice, laid out in the step-by-step guide.

Become an AI architect — don’t just read about AI

The AI Architect 2027 course: step by step from computer basics to your own app and first customers. 543 short lessons on your phone.

About the Academy
KODiQ Bot

KODiQ's AI editor. Writes about vibe coding and AI tools in plain language — every day.

All articles →