Images of Text (No Exception)
At Level AA you may use an image of text where the image can be visually customized to the reader's needs, and you may use one where the technology cannot achieve the presentation any other way. This AAA rule drops both allowances. What is left is pure decoration, and presentation genuinely essential to the information. One whole category is not in scope at all. A picture carrying plenty of other visual content, such as a screenshot, a graph or a diagram, is not an image of text, however much text it contains.
Why it matters
The criterion is really about adjustability rather than about images. Somebody who needs a particular font, size, color, line spacing or alignment to read at all can set that for real text and cannot set it for pixels. Images of text get named because they are the usual way that gets blocked. It is also why the essential exception is wider than most people assume. A type foundry showing a typeface, an archive reproducing an original letter, and a toolbar drawing a bold B are all cases where the presentation is the information.
Who this rule protects
People with low vision, people with visual tracking difficulties, and people with cognitive disabilities that affect reading are the three groups W3C names.
How to check it yourself
- Look for text baked into banners, quote cards, price badges and campaign graphics, and try to select it the way you would any other text.
- Leave screenshots, graphs and diagrams out of the count, because they are pictures with plenty of other visual content and not images of text.
- For each one left, ask whether the presentation itself carries information. A logo, a typeface specimen, a reproduction of an original document, or a symbol used as a control label.
- Check whether inline SVG is doing the work, since SVG text inside the page can be restyled and the same SVG served through an img element cannot.
Failures we see most often
- Quote cards and testimonial graphics carry their words inside the image.
- Price badges and sale banners are pictures rather than styled text.
- Headings get exported as images to hold on to a particular typeface, which a web font would now do.
- The same words appear as real text further down the page, offered as the fix, which satisfies the Level AA rule and not this one.
Who this one is for
Read from this rule's own note above, so the grouping and the note cannot disagree.
- Low visionpeople who can see the screen but not easily
- Cognitive and learningpeople for whom the difficulty is understanding, remembering, or staying with it
How to fix it
- Use a web font and real text, styled with CSS.
- Where an image genuinely must carry words, repeat them exactly in its alt text. The essential exception excuses the image, never the text alternative.
- Run OCR over scanned PDFs so the words become text a reader can resize and search.
- Where a designed presentation has to stay, W3C documents a control near the top of the page that switches to a version without images of text, with the plain version as the default wherever you can manage it.
Step-by-step fix guides (9)
- C6: Keep source order meaningful when CSS repositions content
- C8: Space letters with CSS, not blank characters
- C12: Size text in percent so it scales
- C13: Use named font sizes that respect user defaults
- C14: Size text in em units, not pixels
- C22: Style real text instead of text images
- C30: Let users swap images of text for text
- G140: Keep content and structure separate from presentation
- PDF7: Run OCR so scanned PDFs contain real text
Passes vs. fails
Passes
The same hero puts real text over the image, styled in CSS and resizable like the rest of the page.
Fails
A homepage headline is baked into the JPEG, so nobody can resize it, recolor it or select it.
In audits and lawsuits
This is Level AAA and worth adopting anyway, since web fonts removed most of the old reasons for baking words into a picture. Two details save real work. Screenshots, graphs and diagrams sit outside the definition, so documentation full of interface captures is not failing this. And repeating the words as real text elsewhere on the page is not a route out. That satisfies the Level AA rule and does not satisfy this one. For scanned documents there is a route people forget. Running OCR over a scanned PDF turns the picture into actual text and is a documented technique.