While text is in the text box? As many have mentioned, using speech to text is already unusable if you speak for more than like 3 to 4 consecutive minutes (wasn’t the case a year ago), but that was at least somewhat mitigated by an update oushed out a few months ago that allowed the speech to text button to remain next to the send button even with text in the text box, allowing you to press it again and record in pieces, have it transcribed, and then record the next bits so that you can actually have your full message/code/prompt/context/whatever ready to send to the model at once in the proper format.
For some inexplicable reason though, that has recently been removed, leaving us to have to deal with the wonky speech to text function that will often delete the message altogether or bug out, especially over a certain length, which we now have no method of mitigating other than recording a piece, cut and pasting it to a notepad app or something, and then recording again, just to have to piece it all back together like a scrapbook and finally send to the model.
ChatGPT getting constant updates is good, but it seems to be evolving backwards sometimes. It comes across like experimentation, but with final consumer published products. If we could opt out of this so our productivity doesn’t randomly take massive hits for months on end, that would be much appreciated.
Still an issue — and unfortunately, the latest update has made things even worse.
The timer showing how long you’ve been recording has now been removed, which means you’re now stuck manually tracking your recording duration to avoid going over the 2–3 minute cutoff, which breaks tje flow of thought and defeats the purpose of a natural voice interaction; instead of focusing on what you’re saying, you’re busy watching the clock, remembering to keep track of what time you started as opposed to glancing at the timer that helps you micromanage it down to the second.
Additionally, they’ve reinstated the behavior where stopping a recording immediately sends the message without review. While yes, there’s now an option to preview the transcribed text before sending (which is appreciated), the default behavior of the big button being to send still feels unintuitive. It would make more sense upon hitting the big done arrow for recording to end, then show the text for review, with the option to send manually — especially since mistranslations and truncations are common.
Just from a usability standpoint, it would make much more sense to default to “stop recording = transcribe and preview” rather than “stop = send.” That small added review step ensures clarity and if someone really doesn’t want a review step, it’s a split-second action to hit send again.
These changes, paired with the earlier removal of the ability to re-record in voice mode while there’s already text in the box, have seriously impacted the usefulness of the speech-to-text system, particularly given the still ongoing issue of recordings failing entirely, especially over 3 minutes, still plaguing the feature. And instead of enabling fluid, multi-part recordings, which at least helped mitigate that issue in the meantime, we’re left stitching things together manually using copy and pastes in and oit of the app from different text editors— which shatters workflow and adds even greater friction to what used to be a powerful feature.
Yes, this change feels abrupt, inconsiderate. It is a major downgrade in usability.
It actively leads to: - Incomplete or faulty prompts, - Messages being sent that weren’t meant to be, - AI responses that are irrelevant — simply because we had no way to review what was transcribed.
I realized it yesterday with horror. The entire workflow is ruined.
I urge OpenAI to bring back the classic voice-to-text input mode — with visible transcription and correction before sending.
