Jan 8, 2026 · 2 min read
Until a few weeks ago, I really enjoyed going for walks with ChatGPT's Voice Mode, but it's become increasingly difficult to use.
When it works, it's an incredible tool. I can speak my thoughts aloud (whether they're notes about my workday, update emails to the team, or drafting PRDs) and have it spit them back in a coherent order (as I do babble). I get to think out loud, get some fresh air, and write it out fully behind my computer (I do completely revise the text it comes up with, as it sounds like a robot. But it captures the essence really well, and rewriting it in my voice is really not such a big ask to do). It's genuinely useful.
But recently, something changed. ChatGPT's responses now come in fragments. It's like a very deep, robotic stutter. These pauses that sound like it's buffering or lost connection, except it's not. It's there, it's just... fragmented. And it keeps doing this until the sentence finally ends. It's unbearable.
Of course, I understand this technology is genuinely hard! There are probably millions of people using it at once, and the app needs to record your voice, transcribe it, send it to the LLM, have it understand and respond immediately, all while juggling different APIs in the background. That's incredibly impressive!
Claude's Voice Mode handles this differently. They don't give you real-time responsiveness (I can only guess because they want to guarantee quality). You hit "Send" when you're done talking, wait a noticeable 1-2 seconds, then get a crisp, clear response. Yet that waiting moment, and having to consciously hit send (while you're on a walk) is just as jarring as ChatGPT's stutter. Both break your train of thought, just in different ways.
So I did some testing, and I found something that works reasonably well. If you want the best experience with ChatGPT's Voice Mode right now, try this: stay on the same WiFi (avoid 5G tower switches), use wired headphones instead of Bluetooth, and mute yourself as soon as you stop talking so ChatGPT isn't interrupted mid-response. It genuinely works much better.
Whether that's worth the friction is up to you! Personally, I still enjoy these walks. Hopefully OpenAI figures out a way to make this work seamlessly on 5G again.
Starting in 2026, I'll share more focused notes on product design, technology, and business. If you'd like them in your inbox, leave your email below. I'm always happy to connect via email, Bluesky, or LinkedIn (blergh).