The Five Accessibility Failures I Find on Almost Every Site
After 300-plus WCAG audits, the same five issues account for most of what I find: missing or useless alt text, broken heading hierarchy, unlabeled form fields, links that say "click here," and color contrast that fails at 4.5:1. Each one has a fix that takes minutes. Together they are most of your findings report.
There are 217 days until April 26, 2027. If you are a public entity with a population over 50,000, that is your Title II deadline for WCAG 2.1 Level AA. Under 50,000 and special districts have until April 26, 2028. Fixing these five will not get you to conformance. Not fixing them guarantees you will not get there.
Which five, and how bad is each one?
| Failure | Success criterion | Who it hurts | Automatable? |
|---|---|---|---|
| 1. Missing or useless alt text | 1.1.1 Non-text Content | Screen reader users | Presence yes, quality no |
| 2. Broken heading hierarchy | 1.3.1 Info and Relationships | Screen reader navigation | Yes |
| 3. Unlabeled form fields | 3.3.2 Labels or Instructions, 4.1.2 Name, Role, Value | Anyone using assistive tech | Yes |
| 4. Vague link text | 2.4.4 Link Purpose (In Context) | Link-list navigation | Yes |
| 5. Contrast below 4.5:1 | 1.4.3 Contrast (Minimum) | Low vision, everyone outdoors | Yes |
Four of the five a scanner can find for you. Number one is the one that needs a person, and number one is the most common finding I have. That is not a coincidence.
Why is alt text still the number one issue?
Because it is the only one on the list that requires a judgment call, and judgment does not scale through a checkbox.
At the academic medical center where I work, alt text is documented as the top issue we see from content editors. Not because the editors are careless. Because "describe this image" is a genuinely hard instruction when you are publishing your fourth page of the afternoon.
Here is the decision tree I teach. It takes about ninety seconds to learn.
- Is the image decorative? Does it add nothing the surrounding text does not already say? Give it an empty alt attribute, alt="". Do not delete the attribute. Do not write "decorative image." Empty and intentional.
- Is it informative? Does it convey content? Describe the content, not the picture. Not "photo of a woman at a desk." Write what the reader needs: what is happening, or what the image proves.
- Is it functional? Is it inside a link or a button? Describe the destination or the action, not the graphic. A magnifying glass icon in a search button is alt="Search", not alt="magnifying glass".
- Is it complex? Charts, diagrams, infographics. Short alt names the thing, and a longer description sits in the surrounding text or a linked page. If your alt text is three sentences long, you needed a caption.
The test I use when I am stuck: read the page out loud, substituting your alt text for the image. If the sentence still works, you are done. If it lands with a thud, rewrite it.
What is actually wrong with a heading that skips a level?
Screen reader users navigate by heading. It is the single most common way to move through a page, and skipping from H2 to H4 tells that user a section is nested inside something that does not exist.
The failure is almost never intentional. It is a style choice. Someone wanted smaller text, picked H4, and the visual result looked right.
| Level | Heading |
|---|---|
| H1 | Patient Resources |
| H4 | Before Your Visit |
| H4 | Parking |
| H2 | Billing |
| Level | Heading |
|---|---|
| H1 | Patient Resources |
| H2 | Before Your Visit |
| H2 | Parking |
| H2 | Billing |
Same page. Same look, once you style H2 to be the size you wanted. Completely different structure to anyone not looking at it.
One H1 per page, describing the page. Headings descend in order. You can skip back up freely, you just cannot skip down.
Why can't a placeholder be the label?
Because it disappears the moment someone types, and because assistive tech treats it as a hint rather than a name.
The placeholder-as-label pattern is everywhere: a search box with the word "Search" printed inside it and no label element anywhere in the markup. It looks clean. It fails for a screen reader user, for anyone using voice control (there is no name to speak), and for the person with a cognitive disability who forgot what the field wanted the second they started typing.
Every input gets a real label element with a for attribute pointing at the input's id. If the design cannot tolerate visible labels, the design is wrong, but you can at least use a visually hidden label class. Do not use aria-label as your first choice. Use it when there is genuinely no visible text to associate.
Why is "click here" a real barrier?
Because screen readers can list every link on a page, out of context, as a navigation shortcut. On a page with forty "read more" links, that list is forty identical entries.
The fix is to write the link text as the destination. "Read more" becomes "Read the 2026 accessibility policy." "Click here" becomes "Download the visitor parking map."
This one has a bonus: it is also the single easiest SEO improvement on most sites, and the AI crawlers reading your pages use the same signal. Descriptive link text is the rare fix that pays three ways.
Where does contrast actually fail?
Placeholder text. Designers lighten it to signal that it is a hint. Grey on white lands around 2.8:1. It needs 4.5:1 like any other text.
Disabled buttons. People assume disabled controls are exempt. The exemption in WCAG covers inactive user interface components, but if your disabled button is communicating information the user needs, in practice you should meet 4.5:1 anyway. Auditors argue about this one. Users do not.
Body text needs 4.5:1. Large text (18.66px bold or 24px regular and up) needs 3:1. Non-text elements like icons and input borders need 3:1 under 1.4.11.
How do I find all five on my own site in an afternoon?
- Run an automated scan across your top 100 pages. Contrast, headings, labels, link text, and missing alt attributes all surface here. eiSEO does this, and so do the tools we use at work: Blackboard Ally for Websites, and WAVE. The tool matters less than actually running one.
- Sort the findings by template, not by page. One broken header template is one fix, not four hundred findings.
- Read every alt attribute a human wrote. This is the part the scanner cannot do. Pull them into a spreadsheet and read them as a list. Bad alt text is obvious in a list and invisible on a page.
- Keyboard the top three tasks. Tab through them, no mouse. This finds things no scanner will.
- Fix templates first, then the long tail. The order is the whole strategy.
That is one afternoon and it will tell you more about your real exposure than any score will.
One caution before you start. These five are the most common findings, not the most serious ones. A keyboard trap in a checkout flow locks a person out entirely and will not appear on this list. Fix the five because they are cheap and they are everywhere. Do not mistake finishing them for being done.
Key points
- Five failures account for most of what 300-plus audits turn up, and four of them are machine-detectable.
- Alt text is the most common finding because it is the only one requiring human judgment. Learn the four-way decision tree.
- Heading hierarchy breaks when people choose headings for size instead of structure.
- Placeholders are not labels, and "click here" is a real navigation barrier, not a style preference.
- Fix by template rather than by page, and always read a list of your alt text with your own eyes.
Scan your first 100 pages free with eiSEO, then come back and read the alt-text section again with your results open. The scanner will hand you four of these. The fifth one is yours. If you want to see how a program handles this at scale, ours is documented publicly on the UAMS web accessibility page, including the annual training requirement and the 80% passing quiz for every CMS and calendar editor.
See you next Monday.
I manage web for an academic medical center. Everything I write about that work comes from what is already public. Opinions are mine.