I’m a happy ChatGPT Plus user and use the mobile app daily – it’s an amazing tool, so first of all, thank you to the team behind it!
However, since one of the recent updates, a very important feature seems to be missing or changed: the speech-to-text input via the grey microphone icon in the moblie Version. Previously, when using that button, the spoken text would appear in the text input field, allowing me to review, correct, or edit it before sending it as a prompt.
Now, the behavior has changed – once I speak, the message is immediately sent, without showing me the transcribed text first. This makes it much harder to work with, especially if speech recognition makes a mistake.
Also, this is not the “Voice Mode”, but the speech-to-text feature built into the app interface.
One thing I especially loved about the old version was that I could tap into any part of the text field, place the cursor where I wanted, and use speech-to-text right there to add or change something. That’s no longer possible either.
Does anyone know:
If there’s a way to re-enable the old speech-to-text behavior?
Or if there’s any workaround or setting I may have missed?
Or perhaps even a way to roll back to an earlier version of the app where this still worked?
I’d really appreciate any tips or suggestions. This feature was a key part of how I use ChatGPT on the go, and I’d love to see it return!
I just want to chime in and say I strongly prefer the old version of voice to txt. This new setup is a major downgrade. It feels less intuitive, more intrusive, and honestly, it disrupts the creative process. The previous version let me speak my thoughts, see them transcribed, tweak things if needed, and then send…that workflow made sense.
Now, it just sends everything the moment I stop talking, which is clunky and makes me feel rushed. It removes that crucial review/edit step and splits up the flow in a way that kills spontaneity and natural expression. I really dislike this change, and I hope there’s a way to revert to or restore the older experience.
Yep, I preferred it before when I could edit the text after the speech is converted. Now I will just use google speech unless they change it back.
Also they did another update where gpt says “great question” “preceptive question” etc which is insincere. it is annoying. I understand its intended to make people feel better and it could be a strategy to get people to use it more but preferred it purely factual. it made me consider downloading another back up ai app.
I agree this is a very annoying change. To make it worse, if you for some reason get an error during the upload of your voice clip, it is then lost and you can’t hit retry like we used to be able to. Not sure how we thought this was a good change.
I completely agree — this change has been a huge step in the wrong direction. The ability to review and edit before sending was essential, and losing that really disrupts the flow. I really hope they bring the old behavior back soon.
Edit: FYI, I would normally never make an account or comment on things like this, but this change is so bad I made an account specifically to make my voice heard. Please revert this change!
This change has made the speech-to-text feature completely unusable for me.
Previously, I could dictate multiple short messages, edit and refine them, and then send everything as one well-thought-out response. It was perfect for deeper reflections, structured writing, and complex questions. Now, as soon as I stop speaking, the message sends automatically—which completely breaks that flow.
But it’s worse than that. If a network error occurs during transcription, the message is just dropped entirely, with no option to retry. This happens so frequently on my device that I’ve been forced to switch to another app just to use speech-to-text at all.
Why this matters:
I can’t fix mistakes or add anything before sending.
I now have to stop generation, copy the output into a notes app, edit, and resend—over and over—just to recreate the functionality we had before.
There’s no way to see how long I’ve been recording, which matters because speech-to-text still fails if I go beyond ~4 minutes.
When an error does occur, there’s no retry button—just lost work and frustration.
It completely removes the ability to stack multiple inputs into one clean message, which was essential to my workflow.
Bringing back the ability to edit before sending would massively improve usability. But most importantly, bringing back the ability to retry when an error occurs so users can reliable use this feature again. Thanks
EDIT: Just tried again—network error dropped my entire message with no option to recover it. This is happening constantly now.
EDIT 2: I see I am part of beta testers. I left this group and have access to the older version of the feature. Everything works fine again. Be careful with rolling out this update of the feature!
I experienced the same difficulties. I will drop my premium subscription if the regular STT function isn’t available anymore. I might just use Gemini instead.
I just tested the app again, and it looks like everything is back to the way it was. The speech-to-text feature is working smoothly again – which, in my opinion, is definitely the better setup.
It really seems like our concerns were heard. So a big thanks to everyone who spoke up and helped draw attention to the issue – and a special thank-you to the OpenAI team for addressing it so quickly and effectively.
Fingers crossed it stays this way! If anything changes, we’ll be sure to follow up here.
great job to them they listened, now I no longer need to use google text to speech.
next nice update could be a setting where they allow a toggle option to have gpt talk more formal to less formal depending on the user settings. some of us dont need constant words of affirmation that we are asking a “great question” or “sharp observation”
It still is a bit different for longer inputs. There’s no longer a “retry transcription” if it fails. So longer messages are discarded/lost if it fails. Very frustrating
I hadn’t really noticed this issue yet — probably just because I haven’t recorded any longer messages recently. But reading through your comments, I realize that this new behavior would definitely disrupt the flow I used to rely on.
Also, I’ve had the impression that another useful feature — jumping back into a specific point in a transcribed message to continue from there — isn’t working properly anymore either. That used to be super helpful, and now it feels a bit broken.
Would really be great if OpenAI could fix these things or bring back the previous behavior. It made the whole process much smoother and more flexible.
The only way to not notice it is if your app hasn’t been updated yet or you just haven’t been using the feature.
When using speech to text now, the text bar will turn blue and upon pressing the button to stop recording, it will immediately send what it heard or thinks you said without you being able to verify and modify first.
Still a problem as of today, and today is the first day I’ve noticed it, so it seems to be rolling out differently for users. This is a massive downgrade that I would assume is just an April Fools joke that hasn’t been removed yet lol.
It’s a major step backward and needs to be reversed immediately. Previously, after speaking, users could review the transcribed text in the input box before sending it—allowing time to confirm accuracy, edit for clarity, remove unintended content, or trim fluff/filler for token efficiency/preserving conversation lifespan.
Now, the system sends the transcription to the model instantly without user confirmation, stripping away critical control over how we communicate. This not only risks misrepresentation of our intent but can lead to accidental disclosure, confusion that we have to waste time clearing up and insuring doesn’t persist with the model, or wasted context space—which is especially concerning given the token limits on conversations as a whole that will leave the entire instance dead in the water once reached.
It undermines both usability and trust. Please bring back the ability to review and edit speech input before submission.
Same here — I also had the exact same issue with the blue input bar and the instant auto-send after speech input. It completely bypassed the editable text field, and I found it really disruptive, just like w1nner, Nicodemus9204, and the others described.
However, I’m happy to report that for me, it’s now back to how it used to be. I can see and edit the transcribed text again before sending, and there’s no instant submission anymore. So I’m really hoping this isn’t just a fluke on my device — maybe a quiet follow-up update fixed it?
Fingers crossed it rolls out soon for w1nner and anyone else still stuck with the auto-send version. Would be great if we could get some clarity from the devs whether this is being rolled back officially.
Fingers crossed it just reverts automatically. The new “update” essentially broke one of my favourite ChatGPT features - the one that kept me away from the competition. While the voice-to-text model is mind-blowingly accurate, depriving users of the ability to vet their message before it’s sent is a wild idea. It’s surprising this even made it into a public release.
I just checked my Play Store to see what version of the chatGPT app I’m running - 1.2025.091. The new primitive replacement for the past voice to text is still in place unfortunately. Waiting for the return of the original.
Just wanted to add: I’m currently using a not-quite-latest version of the app (1.2025.084) and luckily had switched to manual updates — which saved me from the worst of the auto-send change.
I still fully agree with everyone here: OpenAI really should bring back the old speech-to-text behavior where we could review and edit before sending.
And if that change ever gets rolled back, I’d really appreciate it if people could post here — would be super helpful to know!
