How the scan works
Four kinds of checks. What the scan is not sure about, it leaves for a human to review.
Four kinds of checks
Fixed rules (axe-core)
A public rule set tests fixed facts, such as colour contrast or an image without alternative text. The same page gives the same result.
Measurements in a real browser
We open the page and use it like a visitor would: with the keyboard, on a narrow screen, with adjusted text spacing. The result is a measured value against a threshold, for example 4.5:1.
Checks with an AI assessment
Some questions cannot be measured, such as whether alt text describes what is in the image. A script finds candidates and collects the evidence: the page, screenshots and measured values. An AI model assesses each candidate separately against a written guideline: passes, fails or cannot be determined. Without evidence the outcome is cannot be determined and it goes to the manual review.
Presence only
For video, audio and things only a human can judge, the scan reports that there is something to look at. It does not say good or bad.
How to read the report
- Finding
- Reported by a check.
- Manual review
- Possibly a problem, but the scan cannot determine it with certainty. A human has to look. The reason is stated.
- Passes
- What we checked, passes. That says nothing about what we do not check.
What the scan covers
Besides axe we use 27 test programs of our own with a total of 107 kinds of checks. They touch 37 of the 55 WCAG 2.2 success criteria at level A and AA, some of them partially. For the other 18 criteria we have no automatic check, mostly because they need human judgement.
What the scan does not do
- Assess PDF documents.
- Assess the content of videos, such as captions and audio description.
- Assess how a screen reader reads out the page.
- Replace a manual audit. Without that audit the scan is no basis for an accessibility statement.
How good is the scan? Tested internally
We compared the scan with public audits by independent auditors on eight websites we did not know. We recorded the outcome before we changed anything.
The scan found about two thirds of what the auditors reported, on the right page and with the right kind of check. The coverage is twice as much as traditional scanners such as axe.
This is our own measurement, not an independent test. Problems from the reports that had already been fixed on the live website did not count. We keep measuring and will update this figure when it changes.
Questions about the scan?
Get in touch, or try the free quickscan first.