Accessibility Paradox: Inclusion vs. Exclusion

• Accessibility Paradox is defined as the contradiction where systems claim universal access while practically enforcing exclusion.
• Empirical evidence from web, mobile, and workplace studies shows that accessibility features often conflict with productivity, legal, and design imperatives.
• Research indicates that as technologies evolve, adaptable, user-centered solutions are essential to overcome gaps between formal compliance and actual usability.

The accessibility paradox denotes a recurring contradiction in which systems, institutions, or organizations claim to expand access while simultaneously reproducing exclusion. In its explicit organizational formulation, Marathe and Piper define the term as “the inherent tension between the productivity- and profit-driven nature of tech companies and their desire to hire and retain disabled workers” (Marathe et al., 25 Aug 2025). Across adjacent literatures on web accessibility, mobile computing, public infrastructure, and education, the same structure appears in related forms: societies endorse equal opportunity and nondiscrimination, yet disabled people still encounter websites, tools, and institutional processes that they cannot independently perceive, understand, navigate, interact with, or use for full participation (Qadri et al., 2011).

1. Conceptual formation and early articulation

Qadri and Banday’s account of web accessibility provides an early conceptual basis for the paradox. They frame accessibility as the degree to which websites and their underlying technologies are “barrier free to all users of information,” and, drawing on W3C/WAI, emphasize that accessible web systems must allow disabled users to “perceive, understand, navigate and interact with the web.” Their paper treats web accessibility as societal, ethical, legal, and technical at once: inaccessible sites deepen the digital divide, conflict with equal opportunity and nondiscrimination, and persist despite legislation, standards, and evaluation tools (Qadri et al., 2011).

That formulation already contains several tensions later generalized under the label “accessibility paradox.” The internet is presented as universal and transformative, yet it can also create a digital divide for disabled users. Laws, standards, and tools exist, yet accessibility remains a continuing challenge. Accessibility is formally recognized, but recognition has outpaced implementation. Qadri and Banday also show why the problem is multi-dimensional rather than reducible to one impairment category or one interface defect: WCAG-aligned disability categories include visual disabilities, auditory disabilities, motor disabilities, and cognitive problems, while assistive technologies such as screen readers, magnifiers, closed captioning, and alternative keyboards succeed only if websites are authored in ways those tools can interpret (Qadri et al., 2011).

Later systems work sharpened the same contradiction. A mobile accessibility study argued that mobile devices had begun to include accessibility features, but these “inclusive efforts still fall short,” mainly because current mobile operating systems lack interoperability, extensibility, and system-wide support for compounded accessibility components (Nicolau et al., 2014). A web science survey made a parallel point for the modern web: as representations became increasingly dynamic, JavaScript- and Ajax-driven, accessibility ceased to be only a checklist problem and became a systems problem of discovering, crawling, and evaluating deferred client-side states (Bostic et al., 2019). This suggests that the paradox is intensified by technological evolution itself: the richer and more interactive the infrastructure becomes, the harder it may be to guarantee that access remains robust.

2. Workplace and software-practice formulations

Marathe and Piper’s study gives the paradox its most explicit organizational name. Based on semi-structured interviews with 20 blind and low-vision professionals in U.S. technology companies, they show that accessibility is undermined not only by isolated inaccessible tools but by a persistent misalignment between disability inclusion and the productivity, speed, standardization, security, and profit imperatives of technology firms. Their participants ranged from roughly 20 to 60 years old, had less than 1 year to 33 years of industry experience, and worked across Big Tech, midsized firms, and startups. The paradox manifested through four interrelated themes: digital infrastructure, accommodations processes and policies, ability assumptions, and competing priorities (Marathe et al., 25 Aug 2025).

The empirical detail is infrastructural rather than symbolic. Participants described inaccessible job application portals, inaccessible or partially accessible offer letters and background-check forms, inaccessible onboarding materials, and persistent barriers in collaboration software, productivity tools, chat systems, project-management systems, performance-evaluation platforms, HR systems, payroll portals, benefits systems, and timekeeping systems. Accommodation processes often signaled formal inclusion while delaying practical access: one participant reported waiting six months to get JAWS approved, and another waited 9 to 12 months for security and privacy assessments of needed software. Workarounds could enable task completion while reducing autonomy, privacy, anonymity, or status; a participant told to send “anonymous” meeting input through a mediator responded, “No, everyone else can be anonymous, but I can’t” (Marathe et al., 25 Aug 2025).

Broader software-practice research supports this organizational reading. A mixed-methods study based on 15 interviews and 365 survey respondents from 26 countries grouped practitioner views into 44 statements across eight topics and found that accessibility is acknowledged but often addressed “in a simple way and with short term goals.” Only about 30% of respondents had direct accessibility-related work experience, and statistically significant gaps appeared between direct and indirect experience groups, especially on whether accessibility belongs in all projects, whether it is intertwined with multiple activities, and whether it is wrongly treated as non-core or mainly front-end work (Bi et al., 2021). A separate mixed-methods survey of 110 mobile app developers across 43 countries reported that about 92.72% rated accessibility at least moderately important, yet only nine participants said accessibility was prioritized from the start through requirements, feasibility, or design reviews, and only six said they tested at every stage of the SDLC. Accessibility testing most often occurred in development, testing/QA, or end-of-development stages, and user testing with people with disabilities was rare (Indika et al., 20 Jan 2026). Together these studies indicate that the paradox is sustained by socio-technical organization: accessibility is ethically endorsed but processually deferred.

3. Compliance, usability, and the limits of automated assurance

One major form of the accessibility paradox is the gap between formal compliance and lived usability. A care-driven accessibility paper argues that accessibility work is too often reduced to checklist conformance, legal risk management, or passing audits. It proposes moving “beyond compliance” toward user autonomy, treating standards as a baseline rather than a ceiling. In that formulation, compliance is a baseline of care, while autonomy is enhanced care. The paper’s “Comfort Mode” framework therefore makes contrast, typography, motion, scaling, spacing, and related properties user-adjustable through progressive disclosure, precisely because accessibility exists “on a spectrum, not a checkbox” (R, 12 Jun 2025).

The paradox becomes especially visible in authoring environments. A study of blind authors on CMS-based websites evaluated WordPress 6.7.1 using WAVE, manual usability testing, and expert analysis through Barrier Walkthrough. The authors found that block-based interfaces were often marked as accessible by automated tools yet produced critical usability issues during manual evaluation. Automated evaluation yielded a median of 14 errors and 22 alerts per page, while manual testing identified 4 critical, 5 significant, and 2 minor severities. Correlation between automated and manual results was limited and uneven: for critical severity the reported Pearson correlation was $r = 0.69$, but for significant severity it was only $r = -0.14$. The most problematic areas were the block editor and design customization, whereas simpler sequential tasks such as login and general settings performed better (Vera-Amaro et al., 20 Aug 2025).

This divergence is consistent with broader web-science work on accessibility assessment. Dynamic web applications increasingly produce client-side states only after execution and interaction, so automated tools remain strongest on inspectable syntactic failures while complex interactive applications still require heavy manual testing. The web-science survey on accessibility, user modeling, and crawling states that current web accessibility research is still focused on increasing the percentage of automatically testable standards and still relies heavily upon manual testing for complex interactive applications (Bostic et al., 2019). The article on blind CMS authors reaches the same conclusion from another direction: effective interaction requires semantic clarity, predictable navigation, meaningful feedback, and workflows aligned with screen readers, none of which can be inferred reliably from automated conformance signals alone (Vera-Amaro et al., 20 Aug 2025).

4. Conflicting access needs, collective access, and capability-space models

A second major form of the paradox arises when improving access for one group constrains access for another. A six-month qualitative field study of a mixed-ability nonprofit asked directly: “When an accessibility feature or practice benefits some people while constraining others, how should designers navigate these trade-offs?” The organization’s communication ecology included Slack, WhatsApp, text messaging, phone calls, TeamSnap, website calendars, social media, and direct messages on personal accounts. Texting was comparatively accessible and familiar for many blind and low-vision members, but it did not scale well organizationally. Slack and TeamSnap supported structured coordination, segmented channels, files, and workflows, but were often more visually oriented, more complex to learn, or less usable for blind members. The authors conclude that there exists no single “technology” to meet the multiplicity of stakeholder needs and preferences, and that access conflicts are not just technical problems but “generative” sites for accountability and repair (Tang et al., 11 May 2026).

This line of work rejects the assumption that accessibility can always be increased monotonically for all users at once. The mixed-ability nonprofit study explicitly distinguishes designing for “everyone” from designing for collective access: the latter is ongoing, local, relational, and contingent rather than universal and static (Tang et al., 11 May 2026). The autonomy-centered “Comfort Mode” paper reaches a related conclusion from personalization rather than organizational ethnography: standardized accessible defaults remain necessary, but no single default can fit all users, so user-controlled customization becomes the practical response to conflicting needs (R, 12 Jun 2025). In both cases, the paradox lies in the failure of singular solutions.

The most formal generalization of this trade-off structure appears in the “Accessibility Capability Boundary” (ACB). That framework models accessibility not as a binary property but as a dynamic capability space over users, systems, and environments. Its constraint vector is

$$\boldsymbol{\kappa} = (L_d, L_c, D_i, P_o, C_x, A_d, A_c, L_z),$$

with components for deployment latency, cognitive load, infrastructure dependency, offline persistence, interaction complexity, adaptability, assistive compatibility, and localization coverage. Accessibility utility is defined as

$$U(\mathcal{S}, \mathcal{U}, \mathcal{E}) = \frac{1}{Z} \sum_i w_i \cdot \phi_i(\kappa_i, \mathcal{U}, \mathcal{E}),$$

and the boundary of a system class $\Sigma$ is

$$\text{ACB}(\Sigma) = \{(\mathcal{U}, \mathcal{E}) \mid \exists \mathcal{S} \in \Sigma : U(\mathcal{S}, \mathcal{U}, \mathcal{E}) \geq \theta\}.$$

The same paper argues that AI-generated browser-native systems can shift this boundary outward by lowering deployment friction and increasing adaptability, but without eliminating verification difficulty, hardware-access constraints, browser inconsistencies, or hallucination risk (Jahangir et al., 19 May 2026). This suggests a formal restatement of the paradox: accessibility improvements are often movements along a frontier of non-dominating trade-offs rather than unconditional gains.

5. Civic infrastructure, education, and geopolitical variation

In public information systems, the paradox appears when transparency mechanisms exclude the public they claim to serve. An audit of six U.S. government dashboard ecosystems found that many charts failed basic discoverability for screen readers or depended on hover, blocking keyboard and screen-reader use. The study checked five items: discoverability of key metrics by assistive technology, keyboard access without mouse hover, clear semantic labels, short plain-language status and trend notes, and machine-readable tables or CSVs mirroring the visual metrics. Its key structural finding was “urgency inversion”: faster, operational dashboards tended to provide fewer accessible affordances than slower accountability dashboards. The result is a temporal asymmetry in civic access: the information that matters most for immediate action is often least accessible to blind and low-vision residents (Acharya, 10 Nov 2025).

Swiss public administration shows the same contradiction at a governance level. A Design Science Research study examined the implementation gap between accessibility legislation and practical implementation for visually impaired citizens and proposed the Inclusive Public Administration Framework, integrating WCAG with the HERMES project management methodology. The study identified resource constraints, fragmented policy enforcement, limited technical expertise, and weak role clarity as barriers. In its evaluation, 60% of respondents rated policy and process alignment as good or excellent, 80% rated technical guidance as good, and 60% rated future adaptability as good, while 40% wanted clearer role delineation and stakeholder responsibilities (Werren et al., 12 Mar 2025). The Swiss case is therefore not a lack-of-standards problem; it is a translation problem between law, standards, and operational delivery.

Cross-regional evidence from the Global South adds a geopolitical dimension. A large-scale comparative study evaluated 100,000 websites from 10 countries in mobile emulation mode and found that countries with stricter accessibility regulations and enforcement tended to adhere better to WCAG. Yet blind and low-vision users were disproportionately affected: only 40.89% of websites met critical accessibility standards for blind users. The most common failures included color contrast, link names, image alt text, meta-viewport settings, button names, labels, and missing or weak ARIA descriptions. The paper connects these outcomes to mobile dependence, slow or unreliable networks, weak enforcement, and limited accessibility expertise (Bhuiyan et al., 28 Jan 2025). This makes the paradox distributional as well as technical: the contexts most dependent on mobile web access are also contexts where accessible implementation is often weakest.

Educational institutions reproduce similar contradictions in another register. An autoethnography of a blind student’s higher-education trajectory in China conceptualizes access itself as a contradictory construct. In a specialized university setting, the environment was relatively navigable because blind students were expected, but the same arrangement narrowed major choice and future opportunities. In a mainstream university, formal inclusion required constant adaptation to sighted norms, ad hoc negotiation, and dependence on unstable accommodations. The authors conclude that accessibility should be understood as “an ongoing, exploratory practice within ableist structures” rather than a final solution to ableism (Zhang et al., 17 Feb 2026). This suggests that the paradox can arise through either segregation or assimilation: access may be granted, but only on terms that reproduce marginality in another form.

6. Distinct technical usages outside disability-access scholarship

In a distinct non-HCI usage, “accessibility” refers to information or state accessibility rather than disability access. A 2025 paper on Loschmidt’s paradox argues that macroscopic irreversibility arises because information accessibility is geometrically bounded even when microscopic dynamics are time-reversible. In that account, Lyapunov exponents occur in paired spectra $\{\lambda_i,-\lambda_i\}$, stable manifolds contract below the thermal de Broglie resolution

$\lambda = \frac{\hbar}{\sqrt{m k_B T}},$

and the information-loss rate is

$$h_{KS} = \frac{1}{2}\sum_{\text{all } i} |\lambda_i|.$$

The paper’s claim is that observers in either time direction only access expanding manifolds operationally, because the information needed for exact reversal is compressed below measurability (Wolfson, 5 Nov 2025).

A related but separate thermodynamic argument relocates the Gibbs paradox from indistinguishability to spatial accessibility. Guo proposes that collisions reduce the instantaneous accessible physical space of a molecule and introduces an additional factor

$e^{-v_i/v_0}$

into the one-particle partition function, where $r = -0.14$0 is a molecule-associated volume and $r = -0.14$1 is the average volume per molecule. On that basis, the paper derives an extensive entropy without invoking the usual $r = -0.14$2 correction (Guo, 2021). These physical-sciences usages are terminologically distinct from disability-access research, but they share a similar logical form: formally possible states or trajectories exist, yet operational accessibility is bounded. This suggests a broader abstract pattern across domains in which the paradox names a gap between nominal possibility and effective reachability.