[
    {
        "role": "user",
        "content": "I%20am%20having%20trouble%20with%20copilot%20and%20accessibility%20using%20jaws.%20Chatgpt%20seems%20better%2C%20am%20I%20missing%20something%3F"
    },
    {
        "role": "assistant",
        "content": "You're definitely not alone in feeling like ChatGPT often works better with JAWS (a screen reader) than Copilot. There *are* documented accessibility issues and differences in how each platform handles screen readers. Below I\u2019ll break down what\u2019s known, what specifically might be causing trouble, and some suggestions or workarounds. If you want, I can also dig into the latest patches or updates specific to your version of Copilot.\n\n---\n\n## What Microsoft says about Copilot + JAWS\n\nHere\u2019s what's officially documented for Copilot:\n\n- **Copilot Vision** supports screen readers like JAWS, Narrator, and NVDA. You can use it to explore what's on a screen, ask questions about it, and hear voice responses.([support.microsoft.com](https://support.microsoft.com/en-US/accessibility/copilot/basic-tasks-using-a-screen-reader-with-copilot-vision?utm_source=openai))  \n- You can have Copilot *read aloud* its responses. This is useful for visual impairments or simply listening rather than reading.([support.microsoft.com](https://support.microsoft.com/en-us/Microsoft-365-Copilot/have-microsoft-365-copilot-read-aloud-to-you?utm_source=openai))  \n\nSo, those are positive signs: Microsoft is building in some support for screen reader users.\n\n---\n\n## What users and institutions report: What\u2019s still broken\n\nHowever, there are a number of known issues (some acknowledged by Microsoft or external institutions) where Copilot doesn\u2019t play nice with JAWS or other screen readers. These problems might be what you\u2019re running into.\n\n- **UCL\u2019s accessibility statement** cites problems like: when revisiting a chat input field (that already has some text in it), JAWS reads the existing text but doesn\u2019t read identifying information (field name, description, etc.). This makes navigation confusing. Buttons may not have clear labels, and some controls fail basic accessibility criteria (for example WCAG guidelines) for name/role/value definitions.([ucl.ac.uk](https://www.ucl.ac.uk/accessibility/digital-accessibility-statements/microsoft-365-copilot-accessibility-statement?utm_source=openai))  \n- In **VS Code\u2019s implementation of Copilot chat / file reference features**, there\u2019s a bug: when Copilot returns a list of files, JAWS reads position (e.g. \u201citem 1 of 5\u201d) but *not the file names or content*. NVDA also has issues (reads \u201clist item\u201d without detailed info).([github.com](https://github.com/microsoft/vscode/issues/202616?utm_source=openai))  \n- In community forums, some users say the Copilot PC app has *\u201cgone to shit\u201d* in terms of accessibility with JAWS/NVDA\u2014citing that it works much less smoothly or not at all compared to ChatGPT.([reddit.com](https://www.reddit.com/r/Blind/comments/1joxtpk?utm_source=openai))  \n\nSo, it\u2019s fair to say that Copilot has several bugs or gaps in its screen reader support.\n\n---\n\n## Why ChatGPT may *feel* better with JAWS\n\nWhile I haven\u2019t found a Microsoft-authored list of \u201cworks perfectly\u201d for ChatGPT with JAWS, users commonly report that:\n\n- ChatGPT tends to have better **consistency** in how chat messages are announced (sender name, message content) when using screen readers.\n- The web app for ChatGPT is often updated with accessibility improvements visible in changelogs/forums, and since it's a simpler layout (chat + optional tools), there are fewer interactive complications.  \n- Community feedback suggests less breakage in navigation\u2014heads-up: this could be in part because fewer integration points (less \u201cembedded\u201d features like file references, sidebar context, etc.) than in Microsoft 365 apps.  \n\nSo overall, for many people, ChatGPT is *more usable* with JAWS right now.\n\n---\n\n## What you might be doing that could help\n\nHere are some ideas to test or try that may improve your experience with Copilot + JAWS:\n\n1. **Check version updates.**  \n   Sometimes fixes are out but not yet in your installed app. If your Copilot (or Copilot in Word/Excel/Outlook) is old, updating might fix a known bug.\n\n2. **Use the web-based version (if available).**  \n   Sometimes the browser-based UI is more accessible\u2014though not always perfect.\n\n3. **Enable any \u201cread aloud\u201d or speech features**  \n   If Copilot has the speaker/read-aloud icon, try using it. It gives audio responses that might be easier to hear through JAWS.([support.microsoft.com](https://support.microsoft.com/en-us/Microsoft-365-Copilot/have-microsoft-365-copilot-read-aloud-to-you?utm_source=openai))\n\n4. **Simplify context.**  \n   If a particular document, field, or file reference list is causing trouble, try isolating just the chat panel. Sometimes embedded elements lose context for JAWS.\n\n5. **Use feedback channels.**  \n   Report bugs to Microsoft Disability Answer Desk or through Copilot\u2019s feedback tools. Known reports are what drives fixes.([support.microsoft.com](https://support.microsoft.com/en-US/accessibility/copilot/basic-tasks-using-a-screen-reader-with-copilot-vision?utm_source=openai))\n\n---\n\n## Bottom line\n\nYou're *not* just imagining this: Copilot currently has documented, real accessibility issues with JAWS that affect field labeling, navigation, content announcements, etc. ChatGPT generally seems more usable in these scenarios, but it too is not perfect (accessibility bugs happen everywhere).\n\nIf you tell me **which version of Copilot** you\u2019re using (PC app, Word, browser, etc.) and what kinds of failures you're seeing (fields not named, file lists unreadable, etc.), I can help more specifically\u2014figure out what\u2019s broken and if there\u2019s a known fix. Do you want me to check that for *your setup*?"
    },
    {
        "role": "user",
        "content": "How%20about%20when%20it%20comes%20to%20vibe%20coding%3F%20Some%20use%20chatgpt%20codex%2C%20some%20use%20claude%2C%20and%20I%20think%20you%20can%20also%20do%20it%20through%20copilot%20gethub.%20Which%20is%20the%20most%20accessible%20and%20usable%20with%20a%20screen%20reader%3F%20What%20are%20the%20key%20differences%20between%20these%20vibe%20coding%20options%3F"
    },
    {
        "role": "assistant",
        "content": "Here\u2019s a breakdown of how some of the popular \u201cvibe\u201d coding tools stack up when it comes to **accessibility with screen readers (especially JAWS)** \u2014 specifically focusing on ChatGPT\u2019s Codex, Claude (and Claude Code), and GitHub Copilot. I\u2019ll compare what each one offers, what their limitations are, and which one might be best depending on what you need.\n\n---\n\n## The Players: What They Do\n\nBefore comparing, a quick recap so we\u2019re all on the same page:\n\n- **Codex** (by OpenAI / ChatGPT): A more developer-oriented mode of ChatGPT; generates code, helps debug, can be embedded in apps or used via prompt workflows.  \n- **Claude / Claude Code** (by Anthropic): Similar largest language-model setups with features to run code, tools, and specialized \u201ccode CLI\u201d workflows.  \n- **GitHub Copilot** (for VS Code, CLI, etc.): Integrates deeply into code editors, offers inline suggestions, chat views, custom agents, and more.\n\n---\n\n## What the Tools Offer for Accessibility\n\nHere\u2019s what I found out recently, with citations:\n\n### Claude Code\n\n- **Screen Reader Mode**: Claude Code has a built-in screen reader mode in the terminal for coding workflows. When enabled (via flag, environment variable, or setting), it outputs text in a plain sequential format with labels (\u201cyou:\u201d, \u201cclaude:\u201d, etc.), removes visual or decorative UI elements, uses numeric menus instead of visual-only navigation. ([support.claude.com](https://support.claude.com/en/articles/15924927-use-claude-code-cli-with-a-screen-reader?utm_source=openai))  \n- This makes terminal-based vibe coding fairly smooth with screen readers\u2014what you see is clear and what\u2019s announced matches what\u2019s there. It supports code CLI well in that context. ([support.claude.com](https://support.claude.com/en/articles/15924927-use-claude-code-cli-with-a-screen-reader?utm_source=openai))\n\n### GitHub Copilot (VS Code + Chat + Agents)\n\n- **VS Code Accessibility Support**: Copilot in VS Code has a published conformance report, tested with JAWS and NVDA, for its chat pane, inline suggestions, etc. It covers many WCAG 2.2 Level A & AA criteria\u2014but not all, meaning some features still have gaps. ([accessibility.github.com](https://accessibility.github.com/conformance/copilot-chat-vs-code/?utm_source=openai))  \n- **Accessibility Agents and Custom Instructions**: Copilot supports \u201cagents\u201d and \u201cworkspace instructions\u201d that let you add accessibility-aware prompts and automations. For example, you can have Copilot always generate accessible HTML (with ARIA, semantic elements, etc.) via path-specific instructions. ([accessibility.github.com](https://accessibility.github.com/documentation/guide/getting-started-with-agents/?utm_source=openai))  \n- **User Interface Considerations**: Copilot provides sound cues, audio signals, accessible diff view, keyboard navigation, etc., especially in VS Code. The conformance report shows many core functions *are* usable with screen readers, though some advanced/dynamic UI areas may still be rough. ([accessibility.github.com](https://accessibility.github.com/documentation/guide/github-copilot-vsc/?utm_source=openai))\n\n### Codex / ChatGPT (Coding with OpenAI Tools)\n\n- **Read-Aloud & Voice Features**: Some users ask for a \"read aloud\" or speech output in Codex / ChatGPT desktop app. Some partial support via OS-level selection or browser read-aloud exists, but it\u2019s not always built in consistently. ([reddit.com](https://www.reddit.com/r/codex/comments/1tty15d/is_there_a_pleasant_way_to_add_read_aloud_to/?utm_source=openai))  \n- **Known Accessibility Issues**: There are reported accessibility bugs in the ChatGPT (Codex) desktop app\u2014e.g., JAWS not reading chat titles in history, issues in recognizing new messages as headings for navigation. ([github.com](https://github.com/openai/codex/issues/34211?utm_source=openai))  \n- **Capability of Producing Accessible Code**: Studies show ChatGPT can generate accessible web code *if prompted specifically*, but by default, it often omits things like semantic HTML, skip links, proper ARIA, etc. So you need to explicitly instruct for accessibility compliance. ([arxiv.org](https://arxiv.org/abs/2501.03572?utm_source=openai))\n\n---\n\n## Strengths & Weaknesses (for Screen Reader Users)\n\nHere\u2019s how they compare for someone using JAWS or similar:\n\n| Tool | Strengths | Weaknesses |\n|------|-----------|------------|\n| **Claude Code** | Screen reader mode in terminal is very accessible; clearly labeled prompts; stripped-down UI helps; good for CLI-based vibe coding without visual distractions. | Less strong with visual-based interactions; full desktop app or file system browsing might lack consistent labels or dynamic content structure. Also, voice/read-aloud features are less mature. |\n| **GitHub Copilot (in VS Code, etc.)** | Strong integration in full editor; accessibility agents + sound cues; pretty solid keyboard navigation; readable inline suggestions; custom agents can enforce accessibility as you write. | Some interactive/dynamic UI parts still have issues; might require pulling in community tooling; needs latest versions; some visual embellishments or decorations might still confuse screen readers. |\n| **Codex / ChatGPT** | Very capable if you explicitly ask for accessible code; flexible prompt-based workflows; accessible via browser which often has built-in screen reader support; good as a companion tool for audits, fixes. | Desktop app has known flaws; automatic accessibility defaults are weak; history and conversation navigation not always labeled well; read-aloud and message hierarchy sometimes broken. |\n\n---\n\n## What \u201cMost Accessible / Usable\u201d Looks Like\n\nIf I were choosing one of these for \u201cbest out-of-the-box screen reader / JAWS experience\u201d for vibe coding, here\u2019s what I\u2019d prioritize:\n\n1. **Strong screen reader mode** \u2014 minimal visuals, well-structured output, clear labels. Claude Code seems best here for terminal workflows.  \n2. **Editor integration with accessible UI** \u2014 Copilot in VS Code is strong if you keep software updated.  \n3. **Prompt or agent support to generate accessible code** \u2014 always ask for accessibility. Copilot\u2019s custom instruction files and accessibility agents can help enforce best practices.  \n\n---\n\n## Key Differences (Summary)\n\n- **Interactivity**: Claude Code in terminal is sequential and simple; Copilot in VS Code has richer interactivity, but also potentially more barriers.  \n- **Default Accessibility**: Claude Code tends to ship more ready-to-use for screen reader users in coding CLI. Copilot requires both updated tools *and* enabling accessible modes/custom agents to get the best experience. Codex / ChatGPT requires explicit constraints/prompts to generate accessible code; navigation in the UI may be hit-or-miss.  \n- **Customizable / Enforceable Audits**: Copilot has richer tooling (custom agents, automated reviews) to enforce accessibility in project codebases. Claude is stronger for personal CLI workflows.  \n\n---\n\n## So What Should You Use?\n\nDepends on your workflow. Here are some recommendations:\n\n- If you work mostly **in the terminal** or want simple, minimal UI, **Claude Code** probably gives you the cleanest experience with JAWS.  \n- If you code in **VS Code or similar IDEs** and want tooling, agents, suggestions in line with editing, **GitHub Copilot** is strong \u2014 make sure screen reader mode is enabled, and/or use the accessibility agents.  \n- Use **ChatGPT / Codex** when you need flexible prompts, style guidance, figuring things out, or reviewing code \u2013 but always prompt for accessibility, and test the output.\n\nIf you tell me your usual setup (OS, editor/IDE, whether you prefer terminal vs GUI, etc.), I can help suggest which of these tools will serve *you best*, and walk you through setting up your ideal environment. Do you want me to customize recommendations for your setup?"
    }
]