Natural language opens some doors and closes others.
On one hand it lowers barriers: for deaf people a conversational system opens access to otherwise frustrating services. For those with cognitive difficulties, a step-by-step conversational guide is more flexible than a rigid wizard; for those moving between languages and cultures, register alignment reduces misunderstandings; for those with dyslexia or dysorthography, speaking instead of writing changes the quality of access.
On the other it raises new ones. Prompt-based interfaces load the formulation of precise requests onto the user, and the articulation barrier hits precisely the categories chat promises to include: the study of about 80 deaf and hard-of-hearing people shows that, for those with sign language as their mother tongue, formulating effective prompts in written English is the main difficulty of use — “as a native ASL user it takes time to build the right prompt, because the channel is only text”. Automatic adaptation does not compensate (see The conversational interface: models only respond to explicit requests, which those with lower literacy struggle to produce). For blind and low-vision users conversational interfaces hardly replace screen readers and mature standards; the study by Adnin and Das on 19 blind users (ASSETS ‘24, Northeastern University) shows that inaccessibilities — unlabeled buttons, unannounced feedback — induce erroneous mental models of the system, which distort trust calibration.
On the regulatory level the framework exists but the tools do not. The European Accessibility Act is in force and refers to WCAG 2.1 AA via EN 301 549. WCAG 3.0 is still a Working Draft (January 2026), with Candidate Recommendation expected for end of 2027 and final standard no earlier than 2028; it will change name to W3C Accessibility Guidelines, extend scope to apps, XR, and operating systems, replace pass/fail with bronze/silver/gold levels, and finally address cognitive accessibility. It explicitly treats AI, via assertions: declaring to have a documented human review process when AI generates or alters content, and to train models on representative and non-biased disability information. These are governance controls, not properties verifiable in the DOM: accessibility moves from technical testing to organizational maturity. Meanwhile the coverage of commercial tools for voice, XR, and AI content is practically zero in 2026: no platform offers turnkey compliance, at most expert manual review.
The answer is not choosing between chat and GUI, but designing hybrid interfaces: conversational and graphic elements, guided paths and direct commands, textual, vocal, and visual inputs — each user chooses the mode suited to the context and the moment.
Effect on the professions
- UX Researcher — Intentionally broaden the perimeter of inclusive research: low AI literacy, articulation difficulties, high cognitive pressure, different cultural and linguistic backgrounds — from the start, not as a later integration. → Useful resources: “We do use it, but not how hearing people think”; “I look at it as the king of knowledge” (ASSETS ‘24)
- Product / UX Designer — The hybrid is not adding a chat to the UI: it requires rethinking the interaction architecture — multiple coherent paths, smooth transitions between modes, no dead ends.
- Content Strategist — Content to be queried via AI (dense, explicit structure) and content to be read (narrative): the coherence between the two must be designed without sacrificing either.
- Front-end Developer — WCAG 3.0 does not yet cover conversational interfaces in a usable way, but WCAG 2.2 already offers criteria applicable to a chat: 4.1.3 Status Messages (new responses must be announced via
aria-live), 2.4.3 Focus Order (focus must remain logical between input field and history), Accessible Authentication, Consistent Help. Automated testing is not enough: manual verification with NVDA or JAWS is needed. - Accessibility Specialist — With the EAA in force, WCAG 3.0 coming, and zero tool coverage on AI interfaces, this is the figure that must build internally what the market does not offer: manual testing protocols for conversation, acceptance criteria for generative components, and the documented processes WCAG 3.0 will ask to declare.