A stairlift doesn’t repair a staircase, just as a screen reader doesn’t rewrite a broken website.

A stairlift and a screen reader solve the same problem: closing the gap between a body and an environment that wasn’t designed for it. Assistive technology now spans everything from a chairlift bolted to a staircase to an AI captioning feature running inside a phone, and the same auditing logic applies to both.

The phone in your pocket is now the highest-adoption assistive device

You don’t have to buy this hardware. It ships free, preinstalled, on every iPhone and every Android sold. The feature arrives the moment you update your software, not the moment you’re diagnosed, prescribed, or fitted, and that changes who gets access, and how fast.

iOS and Android now ship comparable built-in kits for screen readers on mobile and beyond, just under different names:

Feature categoryiOSAndroid
Screen readingVoiceOverTalkBack
MagnificationMagnifierMagnification
Sound alertsSound RecognitionSound Amplifier, Live Caption
Speech on someone’s behalfLive Speech, Personal VoiceSelect to Speak

A dedicated screen magnifier used to be a piece of medical equipment, ordered and paid for on its own. That function is now a toggle in a settings menu that arrived with a free software update. The device didn’t get cheaper so much as it disappeared into hardware people already owned.

Nobody has to ask a doctor’s permission to turn a feature on, and nobody appears in a government equipment ledger for doing it. The improvement arrives quietly, at the speed of an OS update.

Live AI inference replaces captioners and rigid text prediction

Auto-captioning used to mean a stenographer typing live, or captions added after the fact in editing, but not anymore. Live automatic speech recognition generates captions straight from unscripted audio, on-device or in the cloud, with no person relaying the words. Predictive text once meant a fixed dictionary guessing the next word; generative models now draft full sentences from a fragment or a spoken prompt.

Live captioning runs on word error rate

Every automatic speech recognition system gets scored the same way: transcribe a sample, and count what it got wrong. The ACM Digital Library calls that count the Word Error Rate, the share of errors in a machine transcript measured against a correct sample.

Predictive and generative text cut keystrokes

Apple’s predictive text and Google’s Gboard suggestions save keystrokes for users with limited motor control. Generative drafting tools go further, turning a short prompt or dictation into a finished message for someone with dysgraphia or limited typing stamina.

The same shift finally gives screen reader users a way to fix decades of unlabelled images. The Be My AI tool from the Be My Eyes app, for example, now generates a description on the spot, turning a blank alt tag into text a screen reader can read aloud.

The real test for AI accessibility features

Question: Does your new feature still work when a screen reader is running underneath it? 

Answer: Turn on VoiceOver, or NVDA, or Dragon, and watch what happens. 

A structural feature keeps functioning because it’s wired into the page’s markup. A cosmetic one falls apart because it was built for a sighted demo, not the tool a real user has open. The distinction becomes obvious within seconds once a screen reader is turned on.

Company demos run in controlled conditions, with a mouse, a monitor, and often no assistive tech in the loop, unlike real usage. The only fair test is whether the feature survives contact with a screen reader. A chat widget that looks helpful can still trap keyboard focus the moment a screen reader user tabs into it. 

An AI feature layered onto a broken interface doesn’t fix the interface. It just adds a second thing that also doesn’t work.

AI-generated alt text is the clearest case.

Put the description inside a div with no alt attribute path, and it never reaches the accessibility tree. VoiceOver reads nothing, regardless of how good the caption is. The markup fails before the AI layer even engages.

DOM order causes the same collapse. When elements are coded out of visual sequence, a screen reader announces them out of order no matter how accurate any AI description is layered on top. Screen reader users can spot how AI alt text falls apart under a real screen reader. The feature reads well in a press release. It reads as noise in NVDA.

Where AT funding comes from

Most AT funding doesn’t come from families paying out of pocket. Medicaid’s Home and Community-Based Services waivers actually cover most of what reaches users in the US. As of September 2025, 228 of the 263 HCBS programs reviewed covered at least one category of assistive technology.

What medicaid and medicare typically cover

Coverage runs through durable medical equipment rules, and the bar is narrower than most families assume: a device has to be durable, tied to a documented medical need, and prescribed for a qualifying condition, per CMS guidance.

Medicare Advantage and Cost Plans can add extra vision, hearing, and dental benefits beyond Original Medicare. These extra benefits count as a supplemental category, not a guarantee of AT device reimbursement.

What still falls to the family

Buy a general-purpose device that does the same job as prescribed equipment, and it usually falls outside these programs. Built-in phone accessibility features like VoiceOver and TalkBack turn on directly in settings, free and immediate. The remaining cost question shifts from affording a device to affording the phone and the data plan running it. Either way, the mechanism doesn’t change. Something still has to close the space between a body and the environment it’s using, whether that’s via a state waiver or a free OS update.

ADA covers the software you work from

The ADA measures a remote laptop by the same bar it applies to a blocked doorway. A video-call platform or internal portal a screen reader can’t parse triggers it, as would a custom dropdown that traps keyboard focus on an internal HR portal that blocks an employee running NVDA from filing a timesheet. 

The EEOC doesn’t treat a home office as a loophole. Remote tools face the same accommodation duty as anything inside the building.

Employers and vendors get measured against one bar: Section 508 for federal work, WCAG 2.2 for the rest.

A working accommodation process starts with a name: an unlabelled dashboard button, a caption feed dropping every third word. IT or HR checks the claim against WCAG 2.2, ships a fix, and documents it, so the same wall doesn’t stop the next employee who hits it.

Aging turns most people into AT users

Hearing and vision loss catch up with most people eventually. Presbycusis, age-related hearing loss, and age-related macular degeneration are two of the most common consequences of getting older, tracked by the National Institute on Deafness and Other Communication Disorders and the National Eye Institute. Most people meet one or the other eventually.

Nobody books a specialist to turn on Live Caption. It usually starts smaller: the TV volume creeps up past what the rest of the household finds comfortable, someone mentions it, and a grown kid turns on captions instead of arguing about the volume again. The setting was already sitting in the phone’s accessibility menu, unused, because nobody had a reason to go looking for it before. An older adult raising the font size on a text message is reaching for the same toggle a younger user with impaired vision already runs every day, just found by a different route: trial and error at the kitchen table instead of a referral.

No diagnosis required, just a thumb on a slider.

Design for disability and design for age-related decline converge on one toolkit including captions, adjustable text, and voice control. Build it once, and both groups inherit it for free, futureproofed.

What a website accessibility audit tests

WCAG Section 4’s screen-reader test is one of four checks a full audit runs against a live page, not against a written standard. Each check asks the same underlying question in a different form: does this actually work for the tool a real visitor has open?

  • Screen reader compatibility: does a screen reader read every control in order?
  • Keyboard-only navigation: does every action work without a mouse?
  • Color contrast: does text hold up on a low-vision display?
  • Caption accuracy: do captions match what’s actually spoken?

Together, the four checks show whether a feature is structural or just decoration. Run a free accessibility check against a live page and get the answer in minutes, real or for show.

The stairlift and screen reader now converge

A modern stairlift comes with an app and a voice command alongside the rail and the motor. A modern screen reader runs on a language model guessing at meaning instead of a fixed rule set. Both now run on software first, hardware second.

The next accessibility problem shows up as a phone, website, or workplace tool that quietly breaks the AI features already built into it. That’s what audits now catch.

The environment still isn’t built for the people who have to use it. It moved from the rail on the wall to the code inside the app.

Where it will move next remains to be seen.

Frequently asked questions

How do you tell if an AI accessibility feature works?

Turn on a screen reader like JAWS, VoiceOver, NVDA, or Dragon and see if the feature still functions. A structural feature will keep working, because it’s wired into the page’s code.

Does Medicaid cover assistive technology like AAC devices or tablets?

Medicaid’s Home and Community-Based Services waivers cover most assistive technology that reaches users, with 228 of 263 programs reviewed covering at least one category as of September 2025. Coverage requires durability, documented medical need, and a prescription for a qualifying condition. A general-purpose tablet running similar software usually doesn’t qualify.

Are employers required to make remote software accessible under the ADA?

Yes. The EEOC applies the same accommodation duty to remote tools as to anything inside a physical building.

References

  1. Broder, C. (2026, April 21). State medicaid coverage of assistive technology for adults using home- and community-based services – MACPAC. Macpac. https://www.macpac.gov/publication/state-medicaid-coverage-of-assistive-technology-for-adults-using-home-and-community-based-services/ 
  2. Kuhn, K., Kersken, V., Reuter, B., Egger, N., & Zimmermann, G. (2023). Measuring the Accuracy of Automatic Speech Recognition Solutions. ACM Transactions on Accessible Computing, 16(4). https://doi.org/10.1145/3636513 
  3. Medicare. (2019). Durable medical equipment coverage. Medicare.Gov. https://www.medicare.gov/coverage/durable-medical-equipment-dme-coverage 
  4. Medicare & you. (2024). https://www.medicare.gov/publications/10050-Medicare-and-You.pdf 
  5. National Academies of Sciences, E., Division, H. M., Services, B. H. C., Impairments, C. U. S. A. P. T. E. R. E., Flaubert, J. L., Spicer, C. M., & Jette, A. M. (2017). Coverage for Relevant Products and Technologies. National Academies Press (US). https://www.ncbi.nlm.nih.gov/books/NBK453296/ 
  6. National Institute on Deafness and other communication disorders (NIDCD). (2025, January 21). National Institutes of Health (NIH). https://www.nih.gov/about-nih/nih-almanac/national-institute-deafness-other-communication-disorders-nidcd 
  7. U.S. Access Board – W3C publishes working draft of the W3C accessibility guidelines (WCAG) 3.0. (2021). AccessBoard. https://www.access-board.gov/news/2021/01/21/w3c-publishes-working-draft-of-the-web-content-accessibility-guidelines-wcag-3-0/ 
  8. Your coverage options | medicare. (n.d.). Www.Medicare.Gov. Retrieved September 1, 2026, from https://www.medicare.gov/basics/get-started-with-medicare/get-more-coverage/your-coverage-options