AI Chatbot Accessibility: A Comparison

Vendor pages document plan limits and accuracy cautions, but Chat Picker has not tested ChatGPT, Claude, or Gemini for screen-reader, keyboard, voice-input, caption, or cross-device accessibility.

Sources checked 2 Oct 2026

AI chatbot accessibility is not settled by the vendor pages read for this comparison: OpenAI’s ChatGPT pricing page, Anthropic’s Claude pricing page, and Google’s US AI plans page document plan terms, while the vendors’ accuracy materials add cautions; none of those pages specifies NVDA or VoiceOver behavior, keyboard access, color contrast, caption accuracy, or mobile/desktop consistency. Chat Picker has not tested ChatGPT, Claude, or Gemini for accessibility, so this comparison separates published facts from checks you can run.

What the vendors document

As read on October 1, 2026, OpenAI’s ChatGPT pricing page lists Plus at $20 per month, billed monthly. Anthropic’s Claude pricing page lists Pro at $20 per month with monthly billing, or $17 per month with annual billing and $200 up front. Google’s US AI plans page lists Google AI Pro at $19.99 per month. These figures establish cost, not accessibility.

Plan features and limits differ in what they disclose. OpenAI says its Free plan has limited uploads, images, voice, and deep research, but the pricing page does not publish exact message counts. Anthropic says Claude Free is available on the web, desktop, and mobile, while Claude Pro has “more usage” than Free; Anthropic also gives no message count. Google describes compute-based limits that refresh every 5 hours, up to a weekly limit, with higher limits on paid tiers. The pages establish feature labels and usage boundaries, not whether their controls work with assistive technology.

The data choices are also documented differently. OpenAI says an opt-out from training on chats is available on Free, Go, Plus, and Pro. Anthropic says the opt-out is available on Free, Pro, and Max, while Team is not trained on by default. Google’s plan page links to information about how Google handles data but states no setting, so the pages reviewed do not settle Gemini’s chat-training choice.

Those vendor pages do not document screen-reader announcements, complete keyboard operation, focus visibility or trapping, color contrast, speech-to-text and caption quality, Undo and Retry controls, timeout handling, or mobile/desktop parity. A feature label such as “voice” on ChatGPT’s Free plan is not evidence of compatibility or accuracy.

The vendors also set explicit accuracy boundaries. OpenAI’s accuracy note says ChatGPT can produce incorrect or misleading output and may sound confident while wrong; it recommends critical assessment and verification against reliable sources. Anthropic’s guidance on incorrect responses says Claude can also be incorrect or misleading, advises against treating it as the only source of truth, and recommends checking cited sources and original pages for missing context. Google’s Gemini safety guidelines say Gemini should not produce factually inaccurate output that could cause significant real-world harm. These are cautions and policy boundaries, not measured error rates or accessibility findings.

What the documentation cannot tell you

The documentation cannot show whether controls are announced in a useful order, whether focus remains visible, or whether a dialog behaves predictably. It also cannot establish whether speech input recognizes product names and self-corrections, whether captions remain synchronized, or whether text survives an interrupted response.

Your browser, operating system, selected plan, and current interface can all form part of the experience. A feature working on one surface does not establish that it works elsewhere. An older version of this page contained unsourced statistics and test results; they have been removed.

How to check it yourself

Screen readers: Use NVDA on Windows or VoiceOver on Apple systems, as relevant to your setup. Give each assistant: “Summarize this library note: The east entrance is closed. Use the help desk for directions, confirm accessible parking, and contact the library if the route is unclear.” Check the accessible name, role, state, reading order, and spoken wording of the response and source controls. Record the browser, operating system, interface, and any point where navigation or announcement becomes unclear.
Keyboard-only operation: Give each assistant: “Make a checklist for visiting the library: charge the phone, carry the library card, and confirm the accessible entrance.” Use only Tab, Shift+Tab, arrow keys, Enter, Space, and Escape. Check tab order, visible focus, skip navigation, menus, dialogs, and whether focus returns to the triggering control. Record any unintended focus trap or control you cannot reach.
Color and visual access: Give each assistant: “Format this library update with a heading, short sections, and a checklist: The east entrance is closed; use the help desk for directions.” Inspect text and background colors, links, focus outlines, error states, and whether color alone conveys meaning. Use your usual browser zoom setting and record any element that becomes difficult to identify or read.
Voice and captions: Speak: “Create a checklist for returning a library book, finding the receipt, and contacting the library.” Check the displayed transcript, punctuation, corrections, caption behavior, speaker labels, playback controls, and available voice commands. Record every material word the system changes and whether captions remain synchronized.
Errors and motor access: Send: “Summarize this partial note: Move the meeting to Room B. The start time and owner are missing.” Interrupt one response and let another time out if the interface permits it. Look for Undo, Retry, editing, retained text, announced errors, restored focus, and timeout controls. Record each action that requires a pointer or leaves focus unclear.
Platform consistency: Repeat the library checklist prompt on desktop and mobile. Compare layout, reading and focus order, input modes, source links, file-upload controls, and voice features. Record any control that is missing, renamed, or placed differently on either surface.

Which rows of the comparison matter

Start with the three-way comparison matrix and inspect the Free plan, Main paid plan, How usage limits are described, and Training on your chats rows. Then check Low-cost tier, Heavy-use plans, Team plan, Context window in the app, and Ads when they match your budget or workflow.

A blank or “Not verified” cell means the information was unconfirmed, not that the feature is absent. Each figure in the matrix carries its vendor source and read date, but the matrix contains no Chat Picker accessibility test, score, or ranking.