Knowledge base article
Automatic and manual testing.
What a scanner can establish, what a human needs to assess, and why you need both.
Updated on 1 September 2026
In short
- An automated scan finds approximately 30 to 50 percent of real accessibility issues.
- Of the 55 WCAG criteria we track, the scanner touches 31 automatically or partially.
- The rest requires human judgement: whether a link text makes sense in its context, whether an alternative text really describes the image.
- You need both. The scan tracks it, the human judges and decides.
What a scan does and does not find.
An automated scan is fast, consistent and can recheck any page at any time, but it does not find everything. Research from 2017 by the UK's Government Digital Service put the best single tool at 41 percent of planted errors on a test page. Deque, the maker of the scan engine Wexlo uses, itself reports 57 percent, but then counts individual instances rather than types of problem. Contrast errors repeat dozens of times on one page, a keyboard trap only once.
Wexlo therefore uses the lower estimate: automated scanning finds about 30 to 50 percent of real accessibility problems.
The line sometimes runs within a single criterion.
What a machine does well
- Measurable values such as contrast ratios
- Missing technical labels and attributes
- Detectable code and ARIA patterns
What a human needs to assess
- Whether alt text accurately describes the image
- Whether the reading order in context is logical
- Whether instructions are understandable
- Whether the site is genuinely usable with a screen reader
Contrast (WCAG 1.4.3) is deliberately marked "partially" automated in our own criteria map: the contrast engine cannot always determine a background with a decorative image and then gives no verdict. Of the 55 WCAG 2.2 A/AA criteria, our scanner tests 31 automatically or partially automatically.
Where the line is.
Of the 55 WCAG 2.2 A/AA criteria, our scanner tests 31 automatically or partly automatically. The rest need a human.
31
A machine can test it
Automatic or partly automatic, such as contrast values and missing labels.
24
Only a human can assess it
Such as whether alt text is accurate, whether the reading order is logical, or whether the site works with a screen reader.
31 of 55 is about how many criteria we can touch, not about how many problems we find. The latter is lower, around 30 to 50 percent, because within a single criterion a scan can still miss something.
Sources
- Government Digital Service, 2017
Comparative study of automated accessibility tools, 143 planted errors.
- Deque, axe-core
Own coverage claim of 57 percent, counted per instance rather than per type of problem.
- Wexlo criterion map
55 criteria in total, 31 machine-tested, 1.4.3 marked partial. Our own map, checkable criterion by criterion.
Test one page of your own site.
Free and without an account. Rather ask a question first? Use the contact form.