AI Hiring Tools Need an Accessible Way Around the Test

A candidate can do the job and still struggle with the software used to apply for it. A timed assessment may penalize someone who uses assistive technology. A video interview tool may treat an unusual speech pattern as a weak response. A game intended to measure concentration may instead measure how comfortably someone can use its interface.

That is the problem with treating AI hiring assessments as neutral gatekeepers: the system can confuse difficulty taking the test with difficulty doing the work. An employer needs to know which one it is measuring before rejecting anyone.

Consistent scoring is not the same as a fair test

Giving every applicant the same assessment sounds fair. But identical conditions do not necessarily produce a meaningful comparison. If a role involves careful analysis with flexible deadlines, a rapid-response game needs a convincing explanation for why speed predicts success in that job.

The same scrutiny applies to automated video interview scoring. What, precisely, is being evaluated: the substance of an answer, speaking speed, facial movement, or something else? Those are different signals. An employer should not assume that a vendor’s overall accuracy claim establishes that every signal is relevant to the role or appropriate for applicants with disabilities.

This is not merely a hypothetical concern. In its 2022 guidance on disability and algorithmic hiring tools, the U.S. Equal Employment Opportunity Commission described how these systems can screen out qualified people with disabilities. It also explained that employers must consider reasonable accommodations. That guidance concerns U.S. law, but the design problem reaches well beyond one jurisdiction.

The alternative must exist before the rejection

A small accessibility link at the bottom of a rejection email arrives too late. Applicants need to understand the assessment format and how to request help before the clock starts.

A credible process would include:

• Advance information: explain whether the assessment involves timed responses, audio, video, mouse movements, or other interaction requirements.
• A clear accommodation route: make requests for reasonable accommodations in hiring possible without forcing candidates to explain sensitive circumstances to an automated chatbot.
• An equivalent assessment: provide another way to demonstrate the relevant ability where appropriate, rather than simply waiving scrutiny or sending the applicant through a harder process.
• Accountable review: offer human review of hiring decisions by someone who can reconsider an outcome, not merely repeat the score.

Employers also need to ask vendors about job-related assessment validation: what evidence supports the test for this particular role, what limitations are known, and how accessibility was evaluated. Buying a tool does not outsource responsibility for deciding whom it excludes.

There is a privacy boundary here, too. Applicant disability information collected to arrange an accommodation should not quietly become another input to a candidate-ranking model.

The strongest case for automation is that it can reduce repetitive work and help employers assess people more consistently. But accessible recruitment processes need a way to recognize when the assessment itself is the obstacle. A candidate should not have to pass an inaccessible test to earn the right to explain that it was inaccessible.