Android: Issues with new voice-to-text feature and uncontrollable cross-posting of messages across new and existing chats

1. Chat deletion bug causing message repost and automatic replies

When I switch between different chats or delete a chat on the Android app, whether it’s a temporary one or a long-term thread I’ve been using for weeks, the last message I sent in that chat is automatically reposted into the newly opened chat. The app does this instantly, without any input from me. ChatGPT then immediately begins responding. I don’t get to start a fresh chat manually. I get a recycled, out-of-context message I didn’t ask to send, and a response I didn’t ask for.

This happens every single time I delete a chat, and often when switching between existing chats. I have to either wait while it replies to a pointless prompt, or stop the response and delete that chat too. Sometimes the loop happens again before I can finally start a blank chat.

This has caused me to accidentally delete entire projects containing 20 or more chats with critical, ongoing work because I don’t realize it’s taken me to the project page while I’m in this repeat loop of starting and deleting chats that are out of my control. I have screen recordings showing how this plays out.

I lost a project with 20+ chats that I’ve been managing for weeks, and it’s the SECOND TIME that I’ve lost WEEKS of work the same way. I pay for this tool.

2. The new voice-to-text feature on Android is broken and disruptive

This new voice-to-text experience, rolled out in the past month, immediately caused problems. Instead of letting me review what I said before sending, the app now auto-sends the transcribed message the moment I stop talking. There is no way to review, edit, or build the message incrementally like before.

It was briefly removed after rollout, and I was relieved. I assumed someone realized how dysfunctional it was. But now it’s back… and it’s the only option.

Here’s what’s happening:

I will voice a message for one or two minutes, full of important information, and when I click send, it says “sending” but then nothing happens. The message evaporates. The entire voice note is just gone. I have to start all over and repeat everything I just said. This sometimes happens more than once in a row. It’s a huge waste of time.

I can’t build a layered message anymore. With the previous voice-to-text, I could dictate part of a message, then continue dictating more, adding layers and context, and only send it once I was ready. Now I can’t do that. I have to cram everything into a single dictation and if something goes wrong, it’s just gone.

And the longer I speak, the more likely I am to lose the whole thing. This forces me to break up my thoughts unnaturally, because I can’t trust the feature to capture anything longer than a few sentences, and it responds to each individual message instead of responding to an entire, well-constructed request.

Furthermore, if the voice-to-text feature misinterprets a word I’ve said, I cannot review the text of my request before sending. This can alter the app’s interpretation of my request, creating a nonsense response that is totally irrelevant and based on its misunderstanding of a word I’ve dictated. The other day while leaving a voice-to-text inquiry, the app interpreted a word that I said as “ventriloquism” and it gave me a comprehensive response about ventriloquist dummies. I was not asking about ventriloquist dummies and had not made any previous mention of ventriloquism as it had nothing to do with the topic at hand. I then had to spend time explaining the error and re-guiding the chat to the topic I was requesting information about. Waste of time, all because I could not review my voice-to-text before sending.

I also used to be able to add attachments like files or images before sending a voice message. Now I can still add them, but if I use voice-to-text afterward, only the voice message sends. The attachments don’t get sent simultaneously. ChatGPT then responds to the voice message alone, which lacks the intended context, and the result is a completely unhelpful response. I then have to explain the error that has occurred, reattach the image, and repeat myself.

This one update has broken what used to be a vital part of my workflow. It’s not just frustrating. It’s actively sabotaging how I use this app.

1 Like

Im also experiencing Voice to Text issues. Speak message, converting to text message appears, and then nothing. Message lost and Listening message appears again. This has been working fine up until a week ago.

2 Likes

I’m having the same issue since today.
Speech to text seems to work fine outside of projects for some reason but since today, it’s converting voice to text and suddenly reverts back to “listening” which never completes.
I even switched to the beta version but this didn’t help at all.

Tested further and same as reported by eb8880, this works ok outside of projects.

(used ChatGPt to improve my non native English)

Sounds partly like my issue, too.

I’m experiencing this on two devices: a Samsung S23 Ultra smartphone and a Samsung Tab F7SE tablet.

  1. For about a week now, the last sent message automatically reappears when I switch between chats or create a new one. It gets instantly re-posted, and ChatGPT starts replying without my input. There was an update yesterday that seemed to fix the issue — but then, just a few hours later, another update brought the bug back.
  • Workaround: fully closing and reopening the app seems to help for a while.
  1. For the past few weeks, voice-to-text fails completely if the message is longer than ~30 seconds. It says something like “check your internet connection” and the entire message disappears.
  • I’ve started using Google’s speech-to-text instead because the OpenAI version is currently too unreliable. I simply can’t trust it with longer inputs.
  1. Since yesterday, voice-to-text stops working entirely if the chat is part of a project. It says “sending,” then immediately switches back to “listening,” but the message is never sent.
  • In non-project chats, it still works.
  1. Also since yesterday, in advanced voice mode, the app uses the wrong voice if I’m in a project chat, and the feature breaks after just one try.

EDIT: I searched the web last week for my reposting problem and found 1 person posting this here and after 3 days there was no response to it. (I did not respond because I had no account for this board at that time).

2 Likes

Yes same here. Im getting sending… then back to listening, And nothing. I used the original feature alot as it was the best way for me. Now its frustrating

Same for me here, now I’m back to using Claude for voice-to-text feature… What are ChatGPT’s teams doing??

I’m experiencing what seems to be the same issue others have mentioned here, specifically when using the voice input (Whisper) feature in the ChatGPT Android app. The little microphone button appears normally, and I can start recording—everything looks fine while it’s listening, and the timer counts up as expected. But as soon as I tap the checkmark to submit, it shows “sending” for a moment, then the bottom part of the screen—the section with the timer, X button, and checkmark—just disappears. There’s a bit of empty space where those controls were, and then both the voice input button and the regular text input box are gone. I’m left unable to type or say anything at all to continue the conversation. What usually fixes it is switching to a different chat and then going back to the project chat. But all that fixes is the missing buttons. If I accidentally try to use the speech-to-text button again, it breaks again.

Just adding my voice here in the hope that this gets on OpenAI’s radar soon. This bug significantly limits how I use the app.

I had a chat with their support AI and it seems they already know about the issue.
Why it takes them so long to fix something like this, I don’t know.
The AI suggested I can use the voice to text feature in a chat outside a project and then copy the text. This, of course, isn’t a viable solution to the issue or even a temporary fix.