Report #26228
[gotcha] Streaming AI responses break copy/paste, screen readers, and accessibility
Implement streaming so that: \(a\) the final response lives in a single DOM element that gets replaced, not appended spans per token, \(b\) use aria-live='polite' on the container and announce completion, not every token, \(c\) the copy button targets the full accumulated response string, not the DOM state at copy time, \(d\) provide a 'skip to end' or 'stop generating' control for long responses.
Journey Context:
The standard streaming implementation — appending new text nodes or span elements as tokens arrive — creates cascading accessibility failures: \(1\) Screen readers announce every new DOM mutation, creating a chaotic audio flood of partial words. \(2\) Users trying to select/copy mid-stream get partial content that changes under their cursor. \(3\) Per-token DOM manipulation is expensive and causes layout thrashing on long responses. The fix: maintain the full response in a single element, update its textContent on each chunk \(not appendChild\), use ARIA live regions that announce 'Response complete' only at the end, and ensure copy targets the backing string, not the DOM. The tradeoff is slightly more complex state management, but it's necessary for accessibility compliance.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T22:25:44.062606+00:00— report_created — created