I'm Obsessed With New ChatGPT Voice

Okay, look at this. I haven't typed a single word. Matter of fact, my hands are here and ChatGPT is turning a real prompt into a page I can actually use. It's like, heck, using this stuff makes me feel like Tony Stark, like Iron Man sometimes. If you want, you could do a little— I'll be honest with you. Throughout the last week, there were some announcements, but nothing came close to my obsession with voice. It's just these voice tools have become so good in the past 2 weeks that I decided to make today's video A follow-up to last week's video, which you guys seem to really like, that covered the new ChatGPT voice that just sits there and does things with you. Now, for you to understand the depth of my obsession here, check out my ChatGPT account. Boom, 0%, because I used it up mostly through voice. I finally gave in and I bought a pedal that I've been playing around with that I map different voice functions on so I can be completely hands-free like this and prompts just appear. Look, no hands, mom. No, on a serious note though, the voice features in ChatGPT are unbelievable. I have not followed up on a new AI feature in like 6 months. And what I want to show you today is more use cases and more ways that I found to go deeper, like combining it with projects or skills. And I think this is the most relevant thing this week. You know that in this Friday video, I always look at all the things that come out and then I select the one that I want to focus on. And this week it is ChatGPT voice and other voice tools. Let's begin. So as mentioned last week, we looked at the new ChatGPT Voice. A bunch of you asked whether it could do real desktop work, like use projects, create reusable skills. So I tested those exact things. And I want to show you that today. But because OpenAI's naming can be a bit confusing, I want to clear things up first. So I'm using Voice inside ChatGPT Work on the new ChatGPT desktop app. That's this one right here. The regular voice mode is built for conversations back and forth. This version can coordinate the tools and permissions available to work. That's the key to this. And that's what's going to make today's demos possible. And then additionally, my shenanigans with this paddle here are just hooked up to Whisperflow. And I just use it across everything wherever I work. Honestly, I just want to note that I've been looking for like a solution for email with AI. And you know, there's things that help, but my email inbox is so diverse that I can't really let go of answering emails. I still have to do that and that's fine. This, it's not even ChatGPT Voice, we'll look at that today, but this pedal has made it so simple. I just want to show this. Imagine you have like some emails open. Let's just pretend these 3 Word docs are free emails and then like, oh, I gotta reply now and you would start typing. No, look, no hands. I'm using the foot pedal. Hey, thank you so much for reaching out, but after considering some options, I decided against this opportunity. Talk soon. I had a look at your proposal and I actually really like it. I think we could implement it as is. One note would be to add some context on our previous project to make this stick. Nope, not interested. That's how quickly you get through these. Honestly, it just feels smooth. I have a hard time describing it more precisely. Except that the way you kind of move through tasks with a foot pedal is just better. Look, people have been telling me about this. I've been ignoring it, but I'm fully foot pedal pilled. You should try to. I got the fancy Elgato one. It's like $70 or $80. And then I use it with WhisperFlow right here. If you don't want to be paying $20 a month, there's all these open source alternatives. One particularly caught my attention. It's called Quill. If you want to tinker around with it, I still use WhisperFlow because I like the convenience on my phone, on my laptop, on my desktop. But yeah, there you go. You could probably get a $30 foot pedal and use open source WhisperFlow and get the same thing at a discount. But now enough foot pedal talk. Let's actually look at these ChatGPT use cases. And there's some good ones. Okay, let's start with building things because it's just so fun. Generally speaking, like most people, you probably have prompts and skills laying around and they're all over the place, not in one neat place. So I pulled 5 real prompts into one file and I want one page that anybody can actually use that I could easily share. This is just an example that I recently faced where I just wanted to share a few skills with a friend. And my default sharing format oftentimes has become a custom site instead of doing a Word document or a Google Drive folder. It's just sexier, to be honest. So let's do this. I'll just start a new voice chat and I'll say, build me a website with sites using the file attached to this chat. It's a prompt library for non-technical ChatGPT users. Put the context sandwich first, group the other prompts by what they help with, add one sentence explaining when to use each prompt, and give every prompt a copy button. Keep it simple, readable, and mobile-friendly. Don't invent or rewrite any prompt. Show a preview before anything is published.

Okay, let me take a look.

And just like that, it's working already.

Chat context for that attachment.

So what I need to do now is attach this file that I prepped. There you go, a little prompt library source pack. I will just send it. And honestly, it's just this, it's the way it kind of starts working as you're talking and then you add things.

I love it.

And again, you can combine it with answering emails, right? You could create sites to answer emails instead of just writing text. That's more exciting. And by the way, for anybody wondering, Igor, how are you using ChatGPT if you ran out of credits as you showed? Well, I added credits.

I'll build the preview.

Paying per prompt. I just can't stop.

The preview is ready with the context sandwich first and each original prompt grouped with a short when to use it note and a copy button. I'm just restoring a missing local component so you can review it before it publish anything. The unpublished preview is open now. It keeps all 5 prompts unchanged, puts the context sandwich first, groups the rest by purpose, and adds mobile-friendly copy buttons.

Alright, not bad. Now make the context sandwich the obvious starting point, remove any decorative sections that don't help somebody use a prompt, and make the copy buttons easier to spot, like really easy.

Got it. Let me tighten that up. The preview now starts directly with a bold, clearly marked context sandwich section. Decorative framing is removed and every copy button is big, solid green, and high contrast.

One more thing, check every prompt against attached source file. Flag anything you added or changed that wasn't in the source.

Sure, I'll check that.

This is a good little tip that I found helpful throughout the past weeks. Actually, I've been doing this for a while. Trust but verify. Especially when working with Opus 5 these days. I don't know. I'm not the biggest fan of that model. Sometimes it just ignores certain instructions. I feel like I can't fully trust it. Where like 5.6 SOL, I just trust. But Fable is just unbelievable. So I oftentimes use Opus 4.8. Anyone else? Anyway, back to this. Okay, publish it.

All right, publishing now. The reviewed version is ready to go live with the source prompts preserved exactly. It's processing for private publish.

Done.

It's published. Practical AI prompt library. Open it and let me know what you want refined.

And there it is, just through speaking. It's just something about the smoothness of that. Well, taking a while feels right. It feels good. If you haven't tried it yet, I recommend it. And especially building little sites like this on the side while doing other tasks and just muting this thing. You could also map your foot pedal to mute or unmute yourself. And then you can just have this sitting here all day until you run out of credits. It's a new way of life. Okay, so that's the flashy one that produces an artifact and people can try. But this next one is less flashy, but probably more useful every single week. And it uses a skill in the middle, which is something that was recommended by a commenter last week. Shout out to this comment. Great idea. I was trying to add some context to some of these repetitive tasks. And yeah, it turns out skills are the way to do it. If you didn't know, skills are now present inside of chat. And yeah, the same ones that work in Claude can also be imported into here. So it's just great. So let's do this quick little use case that I personally face all the time, which is somebody requesting time to meet. And I'm like, I don't know, I'm in the flow right now. I don't want to be dealing with scheduling and planning things like that. So let's say Daniel, who helps research some of these YouTube videos, sent me an email or a Slack asking for a 30-minute strategy call next week. Normally I would have to open up inbox, calendar, replies, events, then dig through the context before the meeting. Let's see if one voice conversation can solve all of this. Okay, find the newest email from Daniel with the subject YouTube strategy sync next week. Tell me exactly what he's asking for.

Let me pull that up.

Okay, so you won't actually see the email, just pretend as if you saw the email for the case of this demo.

Let me see what I can do here. From the demo, Daniel is asking you to come prepared to align on next week's YouTube strategy. Priorities, audience, content angles.

Then check my calendar from Monday, August 10th. Through Friday, August 10th. Okay, and then stop before you suggest a time, draft a reply, or create anything. I just want to know first.

Let me check your calendar. I'm checking the Monday to Friday window now. I checked Monday, August 10th through Friday, August 14th. Monday and Tuesday are busy. Wednesday is mostly open except 7:15 to 8:15 PM. Thursday has several late morning and evening blocks. Friday has no time.

Amazing. It's just the fact that it knows It still gets me. Hey, okay. So look, every time we do this, I have to reiterate kind of the meeting rules and how I like scheduling. I like for meetings to stick together rather than to be scattered throughout the day. So create a personal Igor skill called Igor's Meeting Rules. For this demo, my rules are 30-minute meetings on weekday afternoons in Lisbon time, a 15-minute buffer before and after, never double booked, choose the earliest valid option, and always show me the reply and calendar event before taking action. 'Ask if anything is unclear, then let me review and install the skill. Also, I like meetings to stick together, as mentioned. So package all of that into an Igor Meeting Booking skill.' Got it.

'The skill is ready for review. Open Igor Meeting Booking to check the details, then tell me if you approve installing it.' That's so good.

All right, so now let's handle Daniel's request with this skill.

Oh, let me take care of that.

Give me a proposed time, draft a brief and friendly reply, and show me the calendar event with the title, attendee, date, time zone, agenda, all this stuff.

Right. I'll pull that together. Skill is installed. I'm refreshing that week now. Then I'll show the reply and event draft without taking action. Proposed for Daniel, Monday, August 10th, 12 to 12:30 PM Lisbon time. Reply draft asks, Would Monday, August 10th at 12 PM Lisbon time work for a 30-minute YouTube strategy sync? Event draft title is YouTube strategy sync. Agenda is to align on priorities, audience, content angles, and next steps. Attendee is Daniel, but I'll need his email before sending. 15-minute buffers are reserved on both sides. Nothing sent or created yet.

Unbelievable. Put that in my calendar and send the email.

Okay, I'll set that up. Quick question, what email address should I use for Daniel? I'll need it for both the calendar invite and the email.

Just do daniel@gmail.com.

Alright, using that. The calendar event is created for Monday, August 10th at noon Lisbon time with Daniel invited and the agenda included. I couldn't send the separate email because this chat still has no connected mail account.

Sure enough, there it is.

Great. If you'd like, we can connect mail so I can send that confirmation email too.

All right, so now that we have this, use the email thread, the calendar event, and the YouTube strategy files that you have in this project to make a one-page briefing. Use the Igor Pogany demo clone project for context.

Let's sort that out.

So instructional moment, this is the part that I didn't know beyond creating and using skills with voice. You can just tap into projects. And what he does is he spins up a new chat and that chat is inside of a project. So last week I did that. I was like, oh, it needs a new chat. So we can't do project, but the chat can access the project in which it's created, which means that it comes into here. So you could set up projects as different context containers and just reference them from here.

Stronger own channel focus, timely educational angles, human presenters, and model agnostic teaching. I'm assembling the one-pager around those points and the scheduled—

All right, I think you see the point. That's really how you use skills and projects with this. And yeah, I just wasn't aware that it's this easy. And those skills are a real unlock. You'll just have to build a few skills for some standard operating procedures or things you do regularly. But yeah, that's just good old prompt engineering, which AI by now is really good at. So you can kind of just speak things into existence. If this isn't AGI on like a personal level, heck, I don't know what is. So the important concept here is that the skill makes this reusable and the skill makes it useful over and over again. Whereas like doing a one-off task like that is useful once, but then restating the rules of booking stuff every time, probably just not worth it. Hey, if you enjoy this type of real testing, make sure to subscribe. I do this all the time on the channel. We take new AI features, turn them into workflows you can actually use and show you where they break or follow up on things that we figure out over time. Okay, one more. And this is a really fun one that I found. And I wanted this for my own job, what I do. But first, a quick correction. Last week I said projects don't work. And as I established, like you can access them like this by opening this chat that's then inside of project, as you can see right here. But also you can go into just a project and start the voice conversation. I feel like this one didn't work last week. Am I crazy? Maybe they enabled it. There's no workaround needed. It just works. Okay, so here's the last use case. So let's say a late AI story just dropped after me recording something. And I don't want ChatGPT just researching it blindly. I want to hold on to the useful parts while I decide what matters. And the useful parts are usually on my screen, you know, like I open different tabs, go through things. So let's say we're researching the new Hermès update, use these project's files and instructions, create a running note called late story addendum. For every source I approve, capture the source link, the exact claim it supports, why it matters to this audience, what still needs verification, and one visual we could show to support the video. Never add anything without a source, keep confirming facts, and infer separate sections.

Sure. Which saved project should I put the late story addendum note in?

Take a look at this. Tell me the main new claim, who published it, and the publication date.

Let me take a look. It says that voice now works inside projects in the current ChatGPT desktop app published by AI News You Can Use on August 6th, 2026.

Now this is what I meant. Look at my screen now. Take a screenshot here.

Okay, let me check that.

Ha, see how they too now added streaming and conversational voice with Barjin. This is the Open Claw update I'm looking for. All right, then. And then also consider these Reddit comments that I have on screen now.

Let me see those.

A few more down here.

I'm seeing the Hermes Agent release post and its comments. I'll treat the release posts as the source for claims and the comments as audience reaction or leads, not proof. Checking those. Got the extra comments. People are asking if the talk feature replaces transcription, expressing strong support.

Okay, turn all of this into a one-page recording brief. Start with the verified headline and why it matters, then give me 3 sourced claims. What to show on screen and the strongest counterpoint and a 20-second hook.

Let me put that together. I've drafted the recording brief. Headline first, why it matters, then 3 sourced claims plus screen suggestions, a counterpoint, and the one thing not to say. The 20-second hook is framed around voice plus approvals with a quick caveat.

And boom, you just get a thing right here. Usually I would go for like a dozen sources and this way it can get everything, the content, the comment sections, the headers, and it could just pull things together really easily. It's like having kind of a scribe. I don't know, it's almost like commanding an assistant that is on a laptop and saying, hey, do this, now look at this side, now pull it all together and now prep it for me. I don't know, it's not night and day to how we're used to using ChatGPT, but then also it is. It feels completely different. And after using this for a while, my takeaway is pretty simple. It's not just better because speaking is faster than typing. It's better when typing would pull you out of your workflow state, or when tasks need various tools that you'd otherwise keep switching in between. The site was most impressive in terms of like, hey, just do this, boom, boom, boom, site. Meeting workflow is arguably the most practical. It's like everybody essentially gets an assistant that can manage their calendar like this and do other things. The Research Scribe is the one I'll probably keep using. That's one for me personally. But overall, it's just a shift in how I use this tool. And I just really, really wanted to let you know, because between this thing building stuff over here and then me throwing up a sticky note and with the pedal on the floor kind of just dictating things into here, I'm getting things done left and right. Heck, I'm getting things done at a speed that I'm not even used to. And I use these tools all the time. It feels more fluid. I like it. And I think you should try this too if you haven't yet. As of other news this week, well, it's actually, I think, one thing. And it's probably this OpenAI release from a few hours ago. Well, I guess on Thursday. And they're basically coming down with pricing on ChatGPT 5.6 Sol, which currently I do prefer over Opus 5. Please leave your thoughts below. And now they power both the instant and the deep reasoning for Plus and Pro users, delivering more factual-focused responses.. And free users now get the Luna starting tomorrow, which is also an upgrade. And there you go. That's what I really wanted to share this week. If you have any further notes on how I could improve the way I work with it, or if you found anything else, please leave them below. I want to see them. I think both projects as context containers and skills to empower whatever you're doing are just so powerful. And honestly, this use case of just getting a daily brief of your different communication platforms and your calendar every day, and then using this voice feature to kind of interact with it and get through any action items is just— it's amazing. It makes me feel like Tony Stark. And I hope this slightly more enthusiastic video than usual will make you feel that way too. And that's pretty much everything I have to share for this week. My name is Igor Pogany, and I hope you have a wonderful rest of the week.

Yeah, this is AI news you can use. Hey, hit play, stay smart, never lose. From the trends to the tricks you could choose, it's AI news. You can use AI News. You can use one stop, one spot for the clues. Every week, new kids, new views. It's AI News. You can use—

let's go!