The test is hybrid: AI agents in the roles of people with disabilities use the site in a real browser and record the barriers they run into. Our expert checks that the findings are correct and adds to them where needed before the report is handed over.

How it differs from an automated test

An automated test checks the page code against rules and finds only some of the barriers; many show up only in real use. A simple AI test, in turn, often just summarises the results of tools or judges a screenshot. The hybrid test is different:

  • It really uses the site, not just scans it. The agents take on the roles of people with disabilities and work with the site as blind, low-vision, deaf and physically disabled users: they operate it with the keyboard, open menus, fill in forms, trigger error messages, enlarge the text, change the contrast, look for captions and transcripts of video and audio, and record the barriers they run into.
  • Every finding is verified. The agents reproduce it on the live site and back it with a screenshot or a measurement.
  • A real screen reader. In the add-on, an agent in the role of a blind user operates the NVDA screen reader with real keystrokes and records what it says, not just what is in the code.
  • Findings you can act on. Each one has a description with evidence, the impact on users, its severity, references to WCAG 2.2 and EN 301 549 and a specific fix.
  • Our methodology and 20 years of practice behind it. The procedures and checks come from our manual evaluations, and our expert checks the findings and adds to them before handover.

What the hybrid test does not replace

It is a simulation: it does not replace testing with people with disabilities, which we offer as part of UX testing and as an add-on to the in-depth evaluation. It differs from the in-depth evaluation in who looks for the findings: here it is a team of AI agents following our methodology and simulated scenarios. In the in-depth evaluation, a team of our experts goes through your website. In the hybrid test, an expert checks that the findings are correct and adds to them where needed. The main difference is the guarantee of completeness.

Sample outputs

For our friends at Slovensko.Digital we ran the hybrid test on two services during HackDay. This is what the outputs of a hybrid test look like (they are in Slovak):

  • Autogram Portal: a web application for signing documents electronically (public and signed-in parts, embedded signing, e-mails, generated PDFs),
  • priznanie.digital: a tax return wizard for the self-employed, also tested with the NVDA screen reader.

In Autogram Portal, most of the findings (40 of 46) were fixed within the same day. The more precise the input, the more precise the output. The output can also be a machine-readable file of instructions for your own agent, with which you can remove most of the barriers found from your code automatically!