CategoryProduct design

Accessibility you can verify: where the guideline stops

6 min read

Contrast ratios, focus rings, alt text length and font sizing are all measurable before a single person opens the page. You can run those checks today and fix what fails. What you can’t check alone is whether the result works for someone using a screen reader on a phone. That is where the guideline stops and your own testing has to start. NN/g keeps a topic index on accessibility that gathers the articles behind this post.

What you can verify without a user

Start with the mechanical checks. They are cheap, and when they fail, you can see it. Low contrast hurts legibility, content discovery, trust and reading under bright light, according to the article on low contrast. That page asks you to judge each color pairing together with the font size, since some pairings only clear the ratio when the text is large enough. It notes that larger text gets read first and lists fixes that don’t require darkening everything: reduce density, change size, reposition, or split a page into sections that open on demand.

In practice, the first thing to measure is the contrast ratio of body text against its background. A tool gives you a number in seconds, and it either meets your target or it doesn’t. What the number won’t tell you is whether the text landed on a busy photo, or whether a caption became too small to read at arm’s length. The semantic structure that screen readers depend on is covered in the post on accessible HTML.

Contrast: the minimum, the measurement and the limit

Contrast shows up in two different jobs, and mixing them up causes trouble. In one job it protects reading, and the piece on low contrast ties weak contrast to lost legibility and lost trust. In the other job it steers attention inside a chart, and the article on contrast in charts means something else by the word. The advice there is to start in gray and color only the series that matters. The title should state the finding, and colors of similar brightness are out because of color blindness.

Checking chart colors against a reading ratio is a waste of time. Treating body text as decoration gives you a page nobody can read.

The same target has to hold in dark mode, where a palette that passed in light mode often fails. The post on dark mode contrast explains why an inverted palette isn’t a free pass.

Alt text: what to describe and what to leave out

The guidance on writing alt text keeps it short, on the order of 150 characters. A decorative image gets an empty alt. A functional image describes the action it triggers, and an informative image describes the message it carries. The same article warns that current AI tools still can’t write alt text on their own, and that they tend to overdescribe a complex image.

A second piece pushes further. Alt text usability argues that an image repeating the text beside it often needs no alt at all. It offers three questions: is it repetitive, is it referential, is it efficient. A screen reader user wants the purpose of the image, not a visual description. Someone with low vision, on the other hand, gets more from a large-print caption than from alt text.

Keyboard and focus

Every interactive element has to be reachable from the keyboard, and the focus indicator has to survive. The article on keyboard access contrasts sequential tabbing with the direct access a pointer gives you. Its hard rule is that a visible focus indicator may be replaced but never removed. Widgets built from div and span elements need extra work, and so do dropdown menus. A page with more than about a hundred menu items should offer a skip link past the navigation.

Font size belongs to the reader

A fixed pixel size fights whoever is reading, so the units that matter are relative. The article on letting users control font size reports that relative values such as 120% and 90% hold up better than absolute ones. It sets a floor of ten points, and twelve points for a senior audience. Test at 10, 12 and 14 points first, then push to 18 and 24 to see where the layout breaks. Text baked into an image can’t be resized at all, which is one more reason to keep words out of pictures.

A screen reader on mobile isn’t a checklist

The study of screen reader users on mobile shows how different the experience is there. On a phone, access is sequential and skimming is hard. A heading must be coded as a heading, which larger text alone isn’t, and a label should put the useful words first. Third-party accessibility menus were ignored by the participants, who called them redundant and restrictive.

A conformance test catches none of that. It passes a site whose headings are styled paragraphs, and it passes a site with a widget bolted on top. Accessibility is not enough argues that treating accessibility as a scoreboard misses the point. The alternative is small and cheap: run a short session with four or five employees who have a disability, one hour each, on the assistive technology they already use.

The same piece reminds us that most people over 45 have reduced vision and depend on resizable type, which ties the font question to the conformance question (NN/g).

A round of five short sessions costs less than a sprint, and it surfaces failures that no rulebook names. It is the same sample size that a plain usability test relies on.

Where the guideline isn’t enough

NN/g’s own articles don’t always agree, and it is better to say so than to smooth it over. The 1996 article on accessible design for users with disabilities treats a literal image description as close to useless. NN/g now flags that text as historical context. The 2024 pieces on alt text open the door to describing the message when the image carries something the surrounding text doesn’t.

An older argument holds that clearing the high-priority checklist can still leave a person unable to use the site, which runs against reading conformance as the finish line.

The two alt text articles want an empty alt when an image repeats nearby text, and that collides with the widespread habit of filling alt on every image.

The chart article uses contrast to steer attention, while every other piece uses the word to guarantee that people can read. Same word, two different jobs, and no note in the guidelines brings them together.

None of this is a reason to skip the mechanical checks. It is a reason to stop treating a passing audit as proof that the page works.

Sources

More articles

See all articles