AI resume-screening tools frequently misrepresent blind and low-vision applicants and create dehumanizing interactions; affected job seekers respond by using assistive workarounds, sharing AI literacy in peer networks, and sometimes refusing to engage with automated systems. Design changes that center disability perspectives and acknowledge interdependent job-search practices are needed to avoid creating new systemic barriers.

Main Finding

AI-driven hiring systems often misrepresent blind and low-vision (BLV) job seekers and produce dehumanizing, contextless interactions. In response, BLV individuals engage in "strategic counter-navigation" (using alternative tools, résumé engineering, and peer-shared AI literacy) and sometimes "strategic refusal" (opting out of systems) to reclaim agency. The authors argue the job search for BLV people is best understood through an interdependence lens—access is produced through social and technical relationships—rather than an individualistic independence model.

Key Points

• Accessibility and screening problems

  • Many job platforms remain inaccessible to screen readers (form controls without labels, CAPTCHAs, timing out), preventing independent application completion.
  • résumé parsers and AI screeners can misinterpret or devalue disability-related achievements and nonstandard career trajectories, filtering out qualified BLV candidates before human review.
  • AI tools produce impersonal, contextless interactions that feel dehumanizing to applicants.

• BLV job-seekers' strategies

  • Strategic counter-navigation: deliberate use of alternate tools to bypass algorithmic filters (e.g., tailoring resume formatting, using plain-text versions), building and leveraging peer networks to share knowledge about which systems work, and using sighted allies where needed.
  • Strategic refusal: choosing not to engage with certain platforms/organizations that rely on inaccessible or opaque AI systems.
  • Social media and personal branding: BLV candidates rely on platforms like LinkedIn but face accessibility gaps; they also use networks to make up for opaque AI decision-making.

• Conceptual contribution

  • The paper applies an interdependence framework, emphasizing that access arises from relationships among people, assistive technologies, hiring platforms, and employers—not simply from an individual's skills.

• Design recommendations (high-level)

  • Build hiring tools that center disability perspectives, enable collaborative/assisted application workflows, increase parser transparency and robustness to nonstandard résumés, and intentionally support informed counter-navigation.

Data & Methods

• Participants: 17 self-identified blind or low-vision adults actively searching for jobs (7 employed, 10 unemployed). Varied use of social media for job search (daily to never), and varied recent application counts (0 to 20+).
• Recruitment: community partners, mailing lists for screen-reader users, social media flyers, and snowball sampling.
• Procedure: 17 remote semi-structured interviews (Zoom) conducted June–September 2024; average length 63 minutes; participants compensated (~US$30/hr).
• Analysis: reflexive thematic analysis—open coding with axial coding to synthesize higher-level themes; quotes anonymized (P1–P17).
• Key empirical insight derived from participants’ reported experiences rather than quantitative outcome measures.

Implications for AI Economics

• Labor market inefficiencies and talent loss

  • Automated pre-screening that misjudges BLV candidates creates search frictions and selection errors: employers potentially miss qualified workers, increasing matching inefficiency and welfare loss relative to inclusive hiring processes.
  • The high unemployment/not-employed rate for BLV individuals (the paper cites ~55% not employed; among active job seekers unemployment ~8%, roughly double non-disabled peers) suggests substantial unused labor supply that opaque AI screening may exacerbate.

• Externalities and hidden transaction costs

  • Accessibility failures shift costs onto BLV job seekers and their social networks (time spent on counter-navigation, reliance on sighted assistance, peer knowledge-sharing). These are negative externalities not priced into AI vendors’ or employers’ decisions.
  • Peer networks and counter-navigation act as informal markets for information and services (e.g., résumé templates, platform-specific heuristics), indicating unmet demand that commercial or non-profit providers could address.

• Incentive misalignment for AI vendors and employers

  • Vendors optimizing for scale or speed have limited incentive to accommodate fringe use-cases (e.g., BLV-friendly parsers), causing a market failure where socially valuable accessibility investments are under-provided.
  • Employers face trade-offs between process efficiency and inclusive matching; absent regulation or reputation costs, efficiency may be prioritized, worsening exclusion.

• Policy and regulatory levers

  • Requiring accessibility standards (WCAG adherence), algorithmic impact assessments, and disclosure of automated decision-making in hiring can internalize social costs and reduce selection bias.
  • Auditing résumé parsers and AI screeners for disability-related biases and requiring remediation would help align private incentives with social welfare.

• Market opportunities and interventions

  • Opportunity for specialized markets: accessibility-first applicant tracking systems (ATS), résumé-conversion tools robust to assistive-technology outputs, and verified alternative application channels (human-reviewed submissions) could emerge.
  • Subsidies, certification, or procurement preferences for accessible hiring platforms could correct market failures and improve matching efficiency.

• Research and measurement gaps that matter for economics

  • Need for quantitative estimates of: (a) how much automated filtering reduces BLV candidate progression at each funnel stage, (b) productivity costs to employers from missed matches, and (c) costs borne by BLV applicants for counter-navigation.
  • Measuring welfare impacts and cost-effectiveness of interventions (e.g., mandatory audits, accessible ATS) is essential to design policy with clear economic trade-offs.

• Broader normative considerations

  • The interdependence framing implies that welfare-enhancing interventions should not only aim to "teach" BLV applicants to game systems but also to redesign systems to reduce dependence on costly informal workarounds.
  • Economically efficient and equitable outcomes require shifting burdens away from disadvantaged job seekers and toward platform designers and employers through incentives, regulation, or market design.

Shortcomings and open questions for economic researchers - This study is qualitative and based on 17 interviews; generalization and causal quantification are needed. - Open empirical questions: prevalence of parser errors across platforms, impact on hiring rates and wages, employer cost-benefit of accessible design, and the elasticity of employer adoption in response to regulation or certification.

Overall, the paper documents a concrete mechanism—AI-mediated screening and inaccessible interfaces—that can generate search frictions and distributional harms in the labor market for BLV workers. From an AI economics perspective, addressing these harms requires correcting incentive misalignments (through regulation, procurement, or market innovation), better measurement of welfare effects, and designing hiring systems that internalize accessibility costs.