Most accessibility advice treats "screen readers" as one thing. They aren't. JAWS, NVDA, and VoiceOver each handle images, links, and decorative graphics differently. If you only test against one, you'll ship alt text that works for some users and fails for others.

This guide explains how each major screen reader announces alt text, gives you a testing checklist, and shows the patterns that work across all three.

Key Takeaways

• JAWS, NVDA, and VoiceOver all read the alt attribute, but with different prefixes ("graphic", "image", "link graphic").
• All three skip images with alt="" entirely — that's the correct behavior for decorative graphics.
• None of them reliably announce the title attribute. Don't depend on it.
• Linked images are announced as links, with the alt text serving as the link's accessible name.
• Test in at least two screen readers before shipping — JAWS and NVDA cover Windows; VoiceOver covers macOS and iOS.

The Three Major Screen Readers

JAWS (Windows)

JAWS is the most-deployed screen reader in enterprise environments and U.S. government. It costs around $1,200 per license but dominates corporate accessibility testing. When JAWS hits an image with alt text, it announces "graphic, [alt text]." Empty alt is skipped silently. Missing alt makes JAWS announce the filename — usually disastrous.

NVDA (Windows)

NVDA is free and open-source, distributed by NV Access. It's the most popular screen reader for personal use and accessibility testing because it's free to install. NVDA announces "graphic [alt text]" or "[alt text] graphic" depending on configuration. Behavior on missing alt depends on user settings — most users have it configured to announce filenames as a last resort.

VoiceOver (macOS, iOS)

VoiceOver is built into Apple's operating systems. It's the dominant screen reader on iPhone and iPad — and a meaningful share of macOS users. VoiceOver announces "image, [alt text]." Empty alt is skipped. Behavior on missing alt is to announce the filename if the file is referenced via src.

Where They Diverge

Linked images

An image inside an <a> tag becomes the link's accessible name. JAWS announces "link graphic [alt text]"; NVDA announces "link, graphic [alt text]"; VoiceOver announces "[alt text], link, image." All three use the alt text as the link label — which is why a logo that links to your homepage should have alt="AltText.ai home" not alt="Company logo".

The title attribute

JAWS may announce the title attribute if configured to (most users don't). NVDA reads it after a delay if "report tooltips" is enabled. VoiceOver handles it inconsistently. The pattern: don't rely on title for information. Put it in alt. See Alt Text vs Title Attribute.

aria-label and aria-labelledby

If you set aria-label on an image, all three screen readers prefer it over the alt attribute. aria-labelledby (which references another element's text) takes priority over both. Use these for icons inside complex widgets where the alt attribute alone can't carry the right context.

Long descriptions

For complex images (charts, infographics), the historical mechanism was the longdesc attribute — but it was deprecated in HTML5 and modern screen readers ignore it. The current pattern is to provide a short alt text and a longer description in the surrounding text or a linked transcript page.

Testing Your Alt Text

Three-tier testing approach:

1. Automated scan first. axe DevTools, WAVE, or AltText.ai's crawl analyzer catch missing alt attributes site-wide.
2. Manual review on top-traffic pages. Read each image's alt text and ask: would a person who can't see the image understand what it shows in context?
3. Screen reader test on critical paths. Run NVDA (free) on Windows or VoiceOver (built in) on macOS through your homepage, product pages, checkout, and contact forms. Listen for filename announcements, redundant alt text, and missing image announcements.

You don't need to test every page in every screen reader. You do need to test the paths most users actually take.

Patterns That Work Across All Three

• Descriptive alt text 80–140 characters. Long enough to be useful, short enough to keep the audio flowing.
• Empty alt="" on decorative images. Don't omit the attribute.
• Functional alt text on linked images, describing the destination ("AltText.ai home") not the visual.
• Skip "image of" / "picture of" — screen readers already announce that the element is an image.
• Don't repeat the alt text in the caption — screen readers will say it twice.

Next Steps

Run a free crawl to catch missing alt across your site, then read What Is Alt Text? for the foundational rules. For sites with thousands of images, AltText.ai automates description generation.