Dictation Loss & Negative Timer Bug – Voice Input Failing Across macOS, iOS, and Web
⸻
Issue Summary:
Dictation in ChatGPT (across iOS, macOS, and web) is unstable and prone to completely deleting recorded speech without recovery. On iOS, this issue often comes with a negative timer glitch (e.g., “–5000s”) before failing with the message:
“Your dictation must be longer than 1 second.”
This issue causes complete data loss for any spoken input and occurs across all platforms where voice input is enabled.
⸻
Platforms Affected:
• iOS app (latest, iOS 17+)
• macOS Safari + Chrome (macOS Ventura & Sonoma)
• Web app (Chrome, Safari, Firefox)
⸻
Steps to Reproduce (iOS version):
1. Tap the microphone button inside ChatGPT app
2. Begin speaking
3. Glitch occurs:
• Timer starts at a negative value (e.g., –5000 seconds)
• Input fails to process or send
• Message appears: “Your dictation must be longer than 1 second.”
• Entire speech is lost without any save or retry option
⸻
Steps to Reproduce (Web/macOS version):
1. Use built-in microphone with voice input
2. Dictate a longer sentence (30+ seconds)
3. Dictation fails:
• No visual error, but input disappears
• Mic icon either stays “stuck” or resets
• No way to re-access or save the recorded speech
⸻
Expected Behavior:
• Dictation should save a temporary audio file (e.g., MP3) during recording
• If transcription fails, user should be able to retry, reupload, or save the audio manually
• There should be no speech deletion without confirmation or recovery
⸻
Impact:
• Loss of original spoken content
• Loss of time, effort, and emotional labor
• Breakdown of trust in platform reliability
• For long-form thinkers and creators, this disrupts workflow and erases irreplaceable thought
⸻
Suggested Solution:
Implement a temporary voice recording buffer:
• Automatically save speech as an MP3 or voice memo file
• Give users the option to:
• Retry transcription
• Save/share the file
• Manually delete it if unnecessary
• Ensure no recorded input is deleted without user consent
⸻
Priority Level: HIGH
This is not a cosmetic issue — it causes permanent data loss. For users who dictate emotionally sensitive, philosophical, creative, or spiritual content, the inability to recover voice input is a critical barrier to trust and usability.