Your phone's voice AI is better than its defaults

Heads up before you scroll: there's a small bonus at the bottom for readers who finish this issue.

OpenAI GPT-Live now translates videos as they play,
fixes grammar while you speak and can finally count to 100 👏

xAI added 21 Grok voices while their users are still frustrated with usage limit. Each voice is built for a job: customer support, education, advertising, commentary, and fictional characters.

OpenAI rushed out ChatGPT Work as Claude Cowork (My Own Workstation) has been absolutely insane since January. I'm running ChatGPT Work and Claude Cowork on the same tasks. Full comparison in an upcoming issue, once I've used both properly.

You pause for half a second to think. Your AI decides you're done and starts answering.

Every voice assistant I use does a version of this. So this year I decided to find out how much of the problem is the tools, and how much is my settings.

Quick background. I work on a MacBook and a Pixel 9 Pro XL. Weird combo, but it's the most productive one I've found.

I left iPhone last year, right after Apple announced a smarter Siri. I still don't know where she is.

That was my Pixel too. Gemini answered questions all day, then asked me to unlock my phone to turn on the flashlight. For a flashlight 🫩

Two settings changed most of this. Both took me under a minute.

Setting 1: let Gemini work on the lock screen.

Open the Gemini app, tap the two lines at the top left, tap your profile icon, and scroll down to Settings. Tap "Gemini on lock screen," then turn on Use Gemini without unlocking.

Newer Google updates may have this on already. Check it anyway. The flashlight worked from the lock screen after this.

Setting 2: connect it to your actual work.

Same menu. Tap "Personal intelligence," then "Connected apps," and turn on the Google Workspace toggle.

Pause before you flip this one: it gives Google's AI access to your Gmail, Calendar, Docs, Drive, Keep, and Tasks. I decided that trade was fine for my work account. Plenty of readers in those comments disagree, and one put it plainly: "I am not giving my inbox access to any agent." Fair. If that's you, skip this setting and keep the first one.

Here's what I now do by voice, tested over the last few weeks:

Reply to this email with “Steps” if you want me to record a video doing any of above 4 tasks so you can follow along and set it up for you.

Anyway for calendar, email, and file tasks, I can't imagine going back to typing. And for anything factual, I add one phrase: "answer from Google Search results." I started doing this so it stops inventing answers, and it works.

Driving is where it's still 50/50. "Play [song]" worked maybe 40% of the time for me. "Play [song] on YouTube Music" worked almost every time. Naming the app is the difference.

And one real win from the road: I asked it to find a restaurant in the direction I was already going, based on my location. It did.

For open conversation, ChatGPT voice is still my number 1. I'll talk to it for 30 minutes straight on a drive: work problems, half-formed ideas I'd never sit down and type. But it still cuts me off mid-sentence. I gave it a standing instruction to wait until I say I'm done. It agreed. It still interrupts me.

A comment under one video: "If you pause in your sentence for so much as half a second, it responds to you." And my favorite reply to it: "ok so im not crazy!"

Half the comment section is that reply.

Know someone still waiting for the smarter Siri? Forward them this issue.

A pattern from those 68 comment sections: voice AI does its best work when you give it one narrow job.

The dictation crowd holds a hotkey, talks, and text lands where the cursor is. The call-screening crowd lets their phone ask "who's calling?" before they pick up. And small business owners are putting voice agents on their phone lines, because by one 2016 survey of small businesses, 62% of inbound calls go unanswered (411 Locals).

That last one is exactly what this issue's sponsor does.

I turned this story (how I use voice AI, plus what you've shared in past polls and replies) into a short cinematic video. One document, one plain sentence, and NotebookLM built the rest: the script, the scenes, the narration.

Two things before you try it yourself. The Cinematic style needs a paid Google AI plan (Google changes the plan limits often, so check what yours includes). And mine took a few tries; you can't edit the finished video, you regenerate it with a clearer sentence.

And the bonus, since you finished: I wrote up the full walkthrough, with my starter prompt and the fixes for when the video comes out wrong. Reply with "Video" and I'll send it to you personally.

Last week I asked Gemini to summarize a client PDF from my Drive. It worked. Then it started reading the entire summary out loud while I sat in the car, because what I wanted was that summary in an email, and I didn't know how to tell it that. I listened to the whole thing anyway.

What's the job you still don't trust voice AI with? Hit reply. I read every answer.

Dan Rice · AI Signal · Every Tuesday Read once. Use AI better all week.