Your Assessment Says the Candidate Failed. What If You Actually Tested Their Disability?
A candidate takes your pre-employment assessment.
They score poorly.
Your system rejects them.
Clean process.
Except there is one question your dashboard probably cannot answer:
Did they fail the assessment—or did the assessment fail them?
New research presented at the 2026 ACM CHI Conference on Human Factors in Computing Systems makes that question difficult to ignore.
Researchers recruited 17 disabled U.S. workers and job seekers and asked them to attempt simulated versions of commonly used digital hiring assessments, including cognitive tests, personality assessments, gamified tests, emotional-intelligence tests and an AI-scored video interview.
Nine of the 17 participants—53%—were unable to complete at least one assessment.
Another seven participants—41%—scored below the fifth percentile on at least one test. DOI
Those percentages should not be treated as population estimates.
This was a small qualitative study intentionally designed to understand people's experiences, not estimate the percentage of disabled candidates affected across the U.S. workforce.
But what happened inside the study matters.
Participants encountered tests they could not see.
Interfaces they could not navigate.
Games that introduced unnecessary cognitive load.
Communication formats that did not accommodate how they communicated.
And assessments where the limitation influencing the score was not necessarily the capability the employer intended to measure.
One participant summarized the problem remarkably well:
“It penalizes you for processing information differently.” DOI
For Talent Acquisition, that creates a problem much larger than accessibility.
Because if a hiring assessment measures something other than what you think it measures, you do not merely have a candidate-experience problem.
You have a decision-quality problem.
The labor-market context makes this impossible to dismiss
In March, the U.S. Bureau of Labor Statistics released its annual disability employment report.
In 2025, the employment-population ratio for people with disabilities was 22.8%, compared with 65.2% for people without disabilities.
Age explains part of that enormous difference because half of people with disabilities in the BLS population were 65 or older.
But not all of it.
Among people with disabilities ages 16–64, only 38.1% were employed.
And across every age group measured, people with disabilities were substantially less likely to be employed than people without disabilities. Bureau of Labor Statistics
Among people participating in the labor market, the unemployment rate tells another part of the story.
It was 8.3% for people with disabilities in 2025.
For people without disabilities:
4.1%. Bureau of Labor Statistics
Again, these statistics do not tell us why any particular person is unemployed.
They cannot establish how much of the gap results from discrimination, accessibility, occupational differences, education, health, labor-force composition or other factors.
BLS itself cautions that the disability population differs demographically from the nondisabled population.
But the disparity is real.
Which means every unnecessary barrier built into hiring deserves scrutiny.
Especially barriers organizations may not realize exist.
Accessibility is usually treated as an accommodation problem
Most hiring organizations approach disability accessibility like this:
Standard process → candidate requests accommodation → organization modifies process.
Sometimes that is exactly how it should work.
The Americans with Disabilities Act generally requires covered employers to provide reasonable accommodations during the application and hiring process for qualified applicants with disabilities unless doing so would impose an undue hardship. ADA.gov
But there is a hidden assumption inside the standard model:
The candidate has to know they need an accommodation before encountering the barrier.
Consider a candidate receiving this email:
Congratulations! The next step is a 30-minute online assessment. Please complete it by Friday.
What exactly should they request?
Additional time?
Screen-reader compatibility?
Captions?
A different communication modality?
Keyboard navigation?
Reduced visual complexity?
An alternative assessment?
They may have no idea what the assessment contains.
The Department of Justice's guidance on algorithmic and technology-enabled hiring explicitly addresses this problem.
Employers using hiring technology may need to tell applicants what technology is being used and how they will be evaluated, provide enough information for applicants to determine whether they need an accommodation, and maintain clear procedures for requesting one. ADA.gov
That is a subtle but important shift.
Accessibility cannot depend entirely on candidates successfully predicting barriers inside a system they have never seen.
The 2026 study shows exactly how that prediction can fail
The CHI researchers used two commercial test-preparation platforms designed to simulate assessments used in real hiring.
Participants completed up to eight assessments across two sessions.
The researchers intentionally included people with different disabilities.
Among the 17 participants:
seven reported cognitive disabilities,
seven vision disabilities,
six ambulatory disabilities,
five independent-living disabilities,
two hearing disabilities,
and others identified conditions including autism, brain injury, multiple sclerosis, chronic pain and epilepsy.
Many participants had multiple disabilities. DOI
Only one of the two assessment platforms offered an accommodation option:
25% additional time.
The other offered none.
Participants frequently assumed that if accessibility options were not presented, they were unavailable. DOI
And this is where the problem becomes operationally interesting.
More time does not make a visual test accessible to someone who cannot see it.
More time does not translate English into American Sign Language.
More time does not necessarily make an interface compatible with assistive technology.
More time does not reduce irrelevant sensory complexity.
Accommodation is not synonymous with extra time.
Yet hiring systems can behave as though it is.
One test asked candidates to identify emotions from photographs
Think about what the employer believes it is measuring.
Emotional intelligence.
Now imagine the assessment consists entirely of photographs.
Candidates must look at faces and identify emotions such as:
anger,
contempt,
disgust,
surprise,
fear,
or happiness.
Two of the 17 disabled participants in the 2026 study could not complete the test at all.
Among the 15 who could, the average accuracy score was 31.2 out of 100. DOI
What does that score mean?
Maybe the participant has difficulty recognizing facial emotion.
Maybe the assessment is poorly constructed.
Maybe the images are inaccessible.
Maybe a visual impairment affected performance.
Maybe cultural or neurocognitive differences matter.
Maybe some combination of those things occurred.
The study was not designed to establish psychometric validity across the broader population.
But it exposes the question every employer should be able to answer before deploying an assessment:
What exactly causes someone to score poorly?
If the answer includes characteristics unrelated to successful job performance, the score becomes much harder to defend as evidence.
This is called construct contamination
You do not need a PhD in industrial-organizational psychology to understand the idea.
Suppose you want to test whether someone can perform Skill A.
Your assessment requires:
Skill A
plus Skill B
plus Ability C
plus navigating Interface D.
The candidate performs poorly.
Which component failed?
If Skill B, Ability C or Interface D is irrelevant to the job, those things have contaminated your measurement of Skill A.
Accessibility failures can therefore become measurement failures.
The Department of Justice gives a straightforward example.
Imagine a school district evaluating candidates using a computer-based test requiring vision.
A visually impaired applicant performs poorly.
But vision is not necessary for performing the staff-assistant job.
If the test screens that person out because of their disability rather than because they lack the actual job skill, the employer may need to provide an accessible test or another adjustment. ADA.gov
That is not simply:
Be nicer to disabled candidates.
It is:
Measure the thing you claim to be measuring.
Every TA leader should care about that.
Gamification creates another version of the problem
Several participants in the 2026 study particularly struggled with gamified assessments.
One example involved inflating a digital balloon while trying to avoid bursting it, intended to measure risk-taking behavior.
Participants questioned why the game itself was necessary.
They described visual distraction, additional cognitive load and unnecessary complexity introduced by the format. DOI
This does not prove gamified assessments are universally invalid or discriminatory.
The study does not support that conclusion.
But it suggests an excellent procurement question:
Why does this assessment need to be a game?
If the construct could be measured through a simpler, more accessible format, what does gamification add?
If the answer is:
engagement,
candidate novelty,
vendor differentiation,
or "it's more fun,"
then TA should ask whether those benefits justify additional variance in how candidates interact with the measurement.
Hiring technology often receives a pass because it looks sophisticated.
Sophistication is not validity.
Then there is the AI video interview
This is where today's topic could easily collapse into our first Journal article.
It won't.
We have already examined AI interview transparency and candidate trust.
The disability question is different.
The DOJ specifically warns about facial- and voice-analysis technologies used in hiring.
An applicant with autism or a speech impairment, for example, could be screened out by a system analyzing facial expression or voice even when those characteristics are unrelated to their ability to perform the job. ADA.gov
This creates a problem traditional bias testing can miss.
Race and gender analyses often involve comparatively familiar demographic categories.
Disability is extraordinarily heterogeneous.
Vision.
Hearing.
Mobility.
Cognition.
Speech.
Neurological conditions.
Mental health conditions.
Chronic illness.
Multiple disabilities occurring simultaneously.
A tool may be accessible to one disabled candidate and unusable by another.
DOJ guidance explicitly notes that hiring technologies can affect different disabilities in different ways. ADA.gov
So:
"Our vendor tested for bias."
is not enough information.
TA needs to know:
Tested for what?
Vendor assurance is not employer governance
This may be one of the most important lessons for modern TA technology procurement.
An assessment vendor tells you:
Our tool is validated.
Our system reduces bias.
Our assessment is accessible.
Our AI is responsible.
Wonderful.
Now show the evidence.
What population was included?
Which disabilities were represented?
What accessibility standards were tested?
Does the interface work with screen readers?
Can it be completed without a mouse?
Are videos captioned?
Does it require speech?
Does it require visual interpretation?
Are time limits necessary to the construct?
What happens when a candidate requests an alternative?
Does the alternative measure the same construct?
How is an accommodation recorded?
Can accommodation status influence downstream reviewers?
Can the system identify candidates whose score may be invalid because of an accessibility failure?
What happens after a vendor updates the product?
The DOJ recommends evaluating hiring technologies before use and regularly while they remain in use for the possibility that they screen out qualified people with disabilities. ADA.gov
Procurement is therefore not the end of assessment governance.
It is the beginning.
Accessibility should not require disclosure to become visible
There is another tension TA leaders need to understand.
The accommodation process often requires candidates to disclose something.
But applicants may have very rational reasons not to want to disclose a disability early in hiring.
And before a conditional job offer, employers face important restrictions around disability-related questions and medical examinations.
The U.S. Equal Employment Opportunity Commission explains that employers generally cannot ask applicants before an offer whether they have a disability or ask questions likely to reveal one, although employers may ask certain questions about accommodations under defined circumstances. EEOC
The result is a design challenge:
How do you make accommodation available without requiring candidates to reveal more than necessary?
The answer is not to ask everyone:
"What disability do you have?"
It is to create a clear, low-friction route that says:
If you need an adjustment to participate in any part of this hiring process, here is how to request one.
Then actually make that route work.
The accommodation request should not disappear into a generic HR inbox.
The recruiter should know what happens next.
The candidate should know what happens next.
And requesting help should not become evidence against the candidate.
We have a recent example of exactly what not to do
In September 2025, the EEOC sued staffing company Dexian DISYS over allegations involving a deaf job candidate.
According to the EEOC's complaint, a recruiter had been pursuing the candidate for an accounting analyst position.
The candidate disclosed that they were hard of hearing and needed an accommodation for the interview.
According to the agency:
the recruiting effort stopped.
The recruiter allegedly ceased responding even after the candidate asked whether disability discrimination was occurring. EEOC
Those are allegations in an EEOC lawsuit, not a final judicial finding, and they should be described accordingly.
But the case illustrates why accommodation cannot merely exist as policy language.
The actual hiring workflow needs to survive the moment someone uses it.
A company can have an accommodation statement on every job posting.
If the recruiter does not know what to do when someone requests one, the candidate experiences the recruiter.
Not the statement.
The accessibility statement is not the accessibility process
You've seen the paragraph:
If you require a reasonable accommodation during the application process, please contact...
Good.
Keep it.
But test what happens next.
Send a request.
Who receives it?
How quickly?
Does that person understand the hiring technology being used?
Can they pause an assessment deadline?
Can they issue an accessible alternative?
Who determines whether the alternative is equivalent?
Does the recruiter know the candidate is waiting?
Does the hiring manager need to know why the process changed?
Probably not.
Does the ATS preserve sensitive information appropriately?
What happens if the request arrives 20 minutes before the assessment deadline?
What happens after hours?
What happens if the candidate cannot access the accommodation-request mechanism itself?
Accessibility is an operational chain.
The statement is one link.
Accommodation requests are also telling you where the system breaks
Here is where this becomes useful beyond compliance.
Organizations often treat accommodation requests as individual exceptions.
Candidate A needed captions.
Candidate B needed extra time.
Candidate C needed another assessment format.
Solve each request.
Close the ticket.
Move on.
But aggregated appropriately, those requests contain system information.
If candidates repeatedly request extra time on the same assessment, investigate why.
If screen-reader users repeatedly cannot access the same platform, investigate the platform.
If candidates consistently require an alternative to the same video assessment, investigate the assessment.
If one stage generates substantially more accommodation requests than every other stage, investigate the stage.
Do not expose individual disability information to people who do not need it.
Do not turn accommodation records into candidate-performance data.
But operational patterns can reveal design failures.
An exception repeated often enough is no longer an exception.
It is evidence.
Your "completion rate" may be lying to you
Suppose your assessment dashboard says:
Completion rate: 92%.
Looks excellent.
But what happened to the other 8%?
Did they lose interest?
Accept another offer?
Decide the assessment was too long?
Encounter technical problems?
Need an accommodation?
Try to request one and give up?
Were they unable to use the platform?
Did the assessment itself cause the withdrawal?
Your completion rate cannot tell you.
This is a recurring problem in hiring analytics.
We measure what the system records.
Then we accidentally treat the recorded outcome as the explanation.
Candidate withdrew is an outcome.
It is not a cause.
Assessment incomplete is an outcome.
It is not a cause.
If accessibility problems are buried inside ordinary candidate attrition, organizations may never see them.
There is a second metric hiding behind completion
Even candidates who finish an assessment may not have received an equivalent opportunity to demonstrate capability.
Remember the 2026 study.
Nine participants could not complete at least one assessment.
But seven others scored below the fifth percentile on at least one test. DOI
Completion therefore cannot be your only accessibility signal.
Someone can successfully click Submit on an inaccessible assessment.
That does not make the resulting score valid.
This is why TA needs to distinguish:
technical completion
from
valid measurement.
The first asks:
Did the candidate finish?
The second asks:
Did the process give them a fair opportunity to demonstrate the job-relevant construct?
Those are not remotely the same question.
What does accessibility cost?
This is where I am deliberately not going to manufacture an ROI calculation.
There is no credible universal number for:
the financial cost of an inaccessible hiring assessment.
You could build a modeled scenario.
Candidate volume × assessment failure × qualified-candidate rate × replacement sourcing cost × vacancy cost.
The arithmetic would work.
The assumptions would dominate the answer.
That would create a very impressive-looking number with very weak meaning.
We do have credible evidence of labor-market disparities.
We have legal obligations.
We have qualitative evidence showing candidates can be prevented from completing assessments.
We have regulatory guidance explaining how hiring technologies can screen out qualified disabled applicants.
That is enough to justify investigation.
PathPair does not need to turn every human problem into a fake dollar sign to make it matter.
Accessibility is not charity
This framing needs to die.
An accessible hiring process is not:
giving disabled candidates an advantage.
The Department of Labor describes reasonable accommodation as a modification enabling a person with a disability to have an equal opportunity to apply for a job and perform its essential functions. Department of Labor
Equal opportunity does not always require identical treatment.
Imagine two candidates taking a test intended to measure written analysis.
One candidate uses a screen reader.
Allowing the screen reader does not give them an unfair advantage.
It removes a barrier unrelated to written analysis.
The correct comparison is not:
Did everyone experience the identical interface?
It is:
Did everyone have a meaningful opportunity to demonstrate the capability we intended to assess?
That distinction should sit at the center of assessment design.
And sometimes accessibility improves the process for everyone
This is one of the quieter benefits.
Clear instructions help candidates with cognitive disabilities.
They also help everyone else.
Captions help deaf and hard-of-hearing candidates.
They also help candidates in noisy environments.
Keyboard navigation helps candidates who cannot use a mouse.
It also helps power users and people dealing with temporary injuries.
Transparent assessment expectations help candidates decide whether they need accommodations.
They also reduce uncertainty for everyone.
Simplifying unnecessary interface complexity helps neurodivergent candidates.
It can also make the assessment easier to understand for the entire applicant population.
Universal design does not eliminate the need for individual accommodation.
But good baseline design reduces how often people must ask the organization to repair the process one candidate at a time.
TA should audit the assessment before auditing the candidate
Take one assessment currently used in your hiring process.
Not the vendor documentation.
The actual candidate experience.
Complete it yourself.
Then answer:
What job requirement does this assessment measure?
Not:
"What does the vendor say it predicts?"
What requirement in this job requires this construct?
What abilities does the interface itself require?
Vision?
Hearing?
Speech?
Fine motor control?
Fast reading?
Memory?
Rapid task switching?
Facial interpretation?
Sustained attention?
Which of those abilities are actually necessary for the job?
If an ability is required by the assessment but not the work, investigate.
What accommodations are available?
Try requesting one.
Do not merely confirm that a policy exists.
What happens to the candidate's score after accommodation?
Is it still comparable?
Does the vendor know?
Does the recruiter know?
Does the hiring manager see accommodation information they do not need?
What happens if the candidate cannot complete the tool at all?
Is there an alternative?
Does the alternative evaluate the same requirement?
Who owns the exception?
If the answer is unclear, fix that before the next candidate discovers it for you.
Stop treating assessment accessibility as somebody else's job
Recruiting says:
The vendor handles accessibility.
The vendor says:
The employer handles accommodations.
Legal says:
Talk to HR.
HR says:
Talk to Recruiting.
Recruiting says:
The candidate should contact the accommodation inbox.
The candidate just wanted to apply for a job.
This is exactly how system failures become human experiences.
Someone needs end-to-end ownership.
Not ownership of the disability.
Not ownership of the medical information.
Ownership of the candidate's ability to move through the process.
That is a hiring-operations responsibility.
The question is bigger than disability
There is a reason this belongs in a publication about hiring experience rather than solely a compliance newsletter.
Accessibility exposes a fundamental truth about selection systems:
Every hiring process makes assumptions about the human being moving through it.
They can see this.
They can hear this.
They have a laptop.
They have high-speed internet.
They can sit uninterrupted for 45 minutes.
They process information quickly.
They communicate verbally.
They understand this interface.
They can travel here.
They can complete this at 2 p.m.
They can perform under this particular artificial constraint.
Sometimes those assumptions are necessary.
A job has requirements.
Sometimes they are accidental.
Hiring Experience Engineering requires knowing the difference.
The best candidate assessment does not identify the best test-taker
It identifies evidence relevant to the work.
That sounds embarrassingly obvious.
Yet the 2026 study shows how quickly the two can diverge.
A participant with a brain-injury-related disability described the fear that assessments reward whoever can "beat the test" rather than whoever can best perform the job. DOI
That is a qualitative observation from one participant.
It is not proof that digital assessments generally select bad employees.
But it is one hell of a question for TA leaders to carry into the next vendor meeting:
Are we identifying the person most capable of doing the job—or the person most capable of surviving our hiring process?
Because those people are not necessarily the same.
And if your process cannot distinguish between them, accessibility is not sitting at the edge of your hiring strategy.
It is sitting directly inside decision quality.
If you want to test this inside your organization
Don't begin with a giant accessibility transformation.
Choose one high-volume role.
Map every candidate-facing interaction from application through offer.
Identify every place where the process assumes a particular sensory, cognitive, physical or communication ability.
Then look at the assessment stage especially closely.
Ask:
What are we measuring?
What else does the candidate have to be able to do merely to complete the measurement?
Where can they request an adjustment?
Who owns that request?
And what happens to the hiring decision when the standard process does not work?
If those questions produce uncomfortable answers, you have found something worth investigating.
That is the purpose of PathPair's complimentary Bridge Assessment: determine whether enough evidence exists to justify deeper examination of the hiring experience before assuming a larger engagement is necessary.
Because accessibility should not begin after someone says:
"I couldn't complete your process."
By then, the system has already told you something.
SOURCES & FURTHER READING
“Disqualified by Disability: The Exclusion of Disabled Workers by Digitized Hiring Assessments” — ACM CHI Conference on Human Factors in Computing Systems, 2026.
The central new research for this edition. Seventeen disabled U.S. participants attempted simulated cognitive, personality, gamified and AI-enabled assessments before completing qualitative interviews about their experiences. DOI
Read the 2026 CHI research
U.S. Bureau of Labor Statistics — People with a Disability: Labor Force Characteristics, 2025 — March 3, 2026.
Primary federal source for current disability employment, unemployment, occupation and labor-force data. Bureau of Labor Statistics
Read the BLS disability employment report
U.S. Department of Justice — Algorithms, Artificial Intelligence, and Disability Discrimination in Hiring.
Primary federal guidance explaining how hiring technology can unlawfully screen out qualified people with disabilities and how employers should address accommodations, accessibility and evaluation. ADA.gov
Read DOJ hiring-technology guidance
U.S. Equal Employment Opportunity Commission — What Can't I Ask When Hiring?
Primary federal guidance covering pre-offer disability questions and circumstances involving accommodation inquiries. EEOC
Read EEOC hiring guidance
U.S. Department of Labor — Opening Doors to All Candidates: Tips for Ensuring Access for Applicants with Disabilities.
Employer guidance covering accessible applications, interviews, accommodation processes and physical access. Department of Labor
Read DOL accessibility guidance
U.S. Department of Labor — Accommodations.
Overview of reasonable accommodation and equal opportunity in the application process under Title I of the ADA. Department of Labor
Read DOL accommodation guidance
EEOC v. Digital Intelligence Systems / Dexian DISYS — EEOC announcement, September 26, 2025.
Recent enforcement example alleging recruiting ceased after a deaf candidate requested an interview accommodation. The allegations had not, in the cited announcement, been established as findings of liability. EEOC
Read the EEOC case announcement
