Since about a week before GPT-5 voice mode launched, I’ve been experiencing serious problems that make voice mode hard to use, despite GPT-5 itself being an excellent upgrade in terms of memory, contextual understanding, and overall conversation quality. I’m a ChatGPT Plus subscriber and use voice mode very frequently in French on both iPhone and iPad.
-
Excessive self-referencing – At the start of conversations, the assistant now reads aloud my persistent instructions instead of applying them silently. For example, it will begin by explaining how it will speak to me or what it has been told to do, even if I never asked for that information. This didn’t happen before and breaks the natural flow of a voice conversation.
-
Paraphrasing loops – When I ask a question, instead of answering, the assistant often just rephrases or summarizes my own question. Even if I insist or repeat myself, it stays in this loop indefinitely until I restart the session. This is especially common if my question is short, if I pause briefly, or if the system misjudges the end of my sentence.
Both issues started before GPT-5, likely due to a backend change, but they have become much more frequent and disruptive since the new voice pipeline was introduced.
Thank you for the outstanding improvements in GPT-5 overall — its memory, contextualization, and conversational abilities are truly impressive. Fixing these voice mode issues would make the experience near-perfect.