My ChatGPT Workflow: When I Use Phone, Voice, Web, and Mac Apps

ChatGPT stopped feeling like a single app a while ago. I now think of it as a small suite of tools: a phone app for thinking on the move, a web app for everyday work, and a Mac app with Codex for heavier projects.

That distinction matters because I don’t want to use the most expensive or powerful mode for every small question. I want to get the most from my subscription without spending unnecessary credits—or making simple work harder than it needs to be.

My basic rule is simple: use ChatGPT for conversation, planning, and lightweight work; use Codex when I need the system to take on a substantial project.

My ChatGPT workflow at a glance

• iPhone app: quick questions, curiosity, dictation, and voice conversations while walking or driving.
• Web app: brainstorming, lightweight research, project context, file management, and everyday tasks.
• Work or deeper research modes: larger research assignments when I’m away from my computer.
• Mac app and Codex: repeatable processes, files, building, coding, and serious multi-step work.

The goal isn’t to find one “best” ChatGPT feature. It’s to match the tool to the kind of thinking or work I’m doing.

I use the iPhone app as a thinking partner

The iPhone app is probably the ChatGPT tool I use most often. Not because it’s the most powerful, but because it’s always available.

I’m increasingly scrolling less when I have a spare minute. If I’m curious about something, I’ll ask ChatGPT. I might want to understand an idea from a podcast, investigate how something works, or explore a question that would otherwise disappear before I get home.

That makes the phone app useful for more than productivity in the traditional sense. It gives me a place to follow curiosity while it’s still alive.

Voice is especially useful when my hands are occupied

The voice feature is where the phone app becomes much more useful than a regular text box. While driving, walking, or running, I can talk through an idea without trying to type on a tiny keyboard.

Sometimes I’ll ask Siri to open ChatGPT voice mode, then start with something simple: “I’ve been thinking about this idea. Help me work through it.” From there, ChatGPT can ask questions, reflect the idea back to me, point out gaps, or help me take it one step further.

I’m not naturally a verbal processor. I usually think best with pen and paper or by sitting down and working through an idea in text. But voice has helped me capture thoughts that would otherwise vanish because I didn’t have my hands free.

The useful part is that the conversation is saved. Later, I can return to it, ask for a summary, or pull out the strongest ideas. A commute or a walk can become a surprisingly productive thinking session without turning into a frantic note-taking exercise.

If you tend to process ideas out loud, this may be even more valuable. But even if you don’t, it’s worth testing. You may discover that speaking an unfinished thought is enough to give it shape.

The web app is where I do lightweight work

The web app is my general-purpose workspace. I use it for brainstorming, outlining, asking questions, and doing lighter research before deciding whether a task deserves more serious processing.

This is also where I try to conserve my higher-cost work. I don’t need an intensive agent to help me brainstorm five possible angles for an article. I don’t need it to spend a long time reasoning through a simple planning question. I can use regular chat to get the thinking started.

That’s especially useful because the different ChatGPT modes can draw from different usage allowances. More autonomous work can consume substantially more credits when it runs for a long time, uses tools, or delegates work to other agents.

So I use a two-stage process:

1. Frame the problem in regular chat. Clarify the goal, explore options, and create a rough plan.
2. Escalate only when necessary. Move the defined task into a deeper research or Codex workflow when it requires sustained effort or action.

This approach also improves the work itself. I’m not asking an agent to wander around trying to understand a vague request. I’ve already done enough thinking to give it a clear outcome.

For more on organizing ongoing work, I’ve written about using ChatGPT Projects to organize AI work. Projects are becoming a more useful replacement for keeping every task in an isolated conversation.

Projects are replacing my old custom GPT workflow

I used to rely much more heavily on custom GPTs. I still have some old ones, but I’ve moved most of my repeatable workflows into projects and Codex.

Projects make more sense for the way I work now because they keep related conversations, instructions, and files together. I have different projects for businesses, side projects, hobbies, and different areas of my life.

That separation matters. When I open a project, ChatGPT has more relevant context and less irrelevant context. I don’t want everything I’ve ever discussed mixed together. I want the system to know which body of information applies to the task in front of me.

Files make projects much more useful

Projects have also become more valuable because I can save sources and files inside them. Some files are documents I uploaded. Others are documents ChatGPT created during earlier work.

For example, I might ask ChatGPT to research a subject and save the result as a document. Later, that document can become a source for a different conversation or project. I can move files, organize them, and reuse them instead of starting from scratch every time.

This has gradually changed how I think about ChatGPT. It’s not just a place to generate a response and move on. It’s becoming a lightweight information system for my work.

The Library feature helps with this by giving me a central view of uploaded files, images, folders, and connected sources such as Google Drive. I can find documents that would otherwise be buried in old conversations and bring them into a new task.

That sounds like a small interface improvement, but it changes the workflow. The value of AI increases when it can work with the information I’ve already created instead of forcing me to explain everything again.

Chat is becoming more proactive—but I still choose when to escalate

As ChatGPT’s context window and memory improve, regular chat can handle more than it used to. It can remember relevant past conversations, access project files, and sometimes identify useful context without me spelling out every detail.

That makes ordinary chat more capable and more “agentic.” It can edit documents, create files, and help manage project materials without requiring me to switch into a specialized work mode for every action.

Still, there’s a difference between helpful context and full autonomy. Regular chat is usually enough for ideation, simple planning, and lightweight tasks. I move to a deeper work mode when I want the system to pursue a larger goal, inspect many sources, or work through a multi-step process with less supervision.

If I’m on my phone and need serious research, I may switch into the deeper work area and ask for a thorough investigation. But I’m using that selectively. For many research questions, I can now simply describe the depth I need: “Do a deep dive. Find the strongest sources, compare the options, and compile the findings into a document.”

That natural-language instruction often gets me closer to the result I want than choosing a feature because its name sounds impressive.

Of course, there are still times when the dedicated deep research workflow is the right choice. The point is not to avoid it. The point is to understand when the extra depth is justified.

Codex is where I do substantial work

On my Mac, I almost never use the regular ChatGPT side of the app. I go straight to Codex.

Despite the name, I don’t think of Codex as a tool only for writing code. I think of it as my work mode. It can work with files and folders on my computer, follow documented processes, create things, edit things, and complete a much larger sequence of tasks than a normal chat conversation.

That distinction helped me understand its value. I had initially associated Codex with programming. Now I use it for many kinds of work, including content production, research, image creation, and repeatable operations.

Give Codex a process, not just a prompt

The most powerful setup I’ve built is a folder for my personal brand with documents that explain my recurring processes. For example, I have a process for starting a new piece of content.

When I say, “Let’s start a new episode,” Codex already knows the sequence. It can help develop angles, generate title options, explore image concepts, and create the related assets. I don’t have to explain the entire process every time because the process exists in the project files.

That’s a major difference from asking ChatGPT to perform a one-off task. Instead of only giving it instructions in the moment, I’m building a reusable operating system around the work.

The workflow is still collaborative. Codex moves through the steps, and I decide when to continue, adjust the direction, or reject an option. I’m not handing over judgment. I’m reducing the amount of repetitive coordination I have to do.

For a related example, see how I built a custom GPT to handle repeatable work. The tools have changed, but the underlying principle remains: document the process so the AI can help execute it consistently.

Keep project folders contained

Codex projects typically correspond to folders on my computer. Those folders hold the files, instructions, and assets associated with the work.

I like keeping projects contained because it gives the system a clear boundary. It can work with the relevant files without automatically having access to my entire computer.

I don’t recommend giving an AI unrestricted access to everything unless you understand the risks and have a dedicated environment for it. A well-organized folder is safer and usually more useful than a giant pile of unrelated files.

Inside a project folder, Codex can create files, update documents, and build on previous work. I don’t need to understand every line of code it changes to benefit from it—but I do need to review the results and remain responsible for what gets shipped.

How I decide which tool to use

When I’m unsure where to begin, I use this simple decision tree:

1. Do I need to capture or explore a thought? Use the iPhone app, especially voice mode.
2. Is this a small question, brainstorm, outline, or simple task? Use regular ChatGPT in the web app.
3. Do I need substantial research while I’m away from my computer? Use the deeper work mode selectively and specify the outcome clearly.
4. Does this involve files, a repeatable process, building, or multiple steps? Move to Codex on the Mac or Windows app.

I also try to separate thinking from execution. ChatGPT is excellent for helping me understand what I want to do. Codex is better when I know the goal and want the system to help carry out the work.

The practical takeaway

Don’t treat ChatGPT as one giant text box. Use the phone app when you need a thinking partner in the real world. Use the web app for lightweight work and project-based context. Save deeper research for tasks that genuinely need it. Then use Codex for substantial, repeatable work with files and processes.

The biggest productivity gain hasn’t come from finding a magical prompt. It has come from giving each interface a job.

Start with the least expensive tool that can handle the task. Build context in projects. Document the processes you repeat. Escalate to Codex when the work becomes multi-step or operational. That workflow gives AI room to create leverage while keeping my judgment, taste, and responsibility in the loop.

The conversation covers the workflow of using ChatGPT across different platforms and apps, including the iPhone app, web app, and Mac app. It explores the use of voice features for productivity and ideation, transitioning to Codex for intense work and research, leveraging projects and file management, and maximizing productivity with the Mac app and Codex. Takeaways ChatGPT is a suite of apps and functions Voice feature for productivity and ideation Chapters 00:00 Introduction to ChatGPT Workflow 00:58 Using ChatGPT iPhone App for Productivity and Ideation 03:06 Utilizing ChatGPT Web App for Lightweight Tasks and Ideation 05:27 Transitioning to Codex for Intense Work and Research 06:24 Exploring Projects and File Management in ChatGPT Web App 07:22 Leveraging the Library Feature and File Management in ChatGPT Web App 08:21 Enhanced Context and Proactive Features in ChatGPT Web App 09:15 Deep Research and Intense Research in ChatGPT Web App 10:32 Maximizing Productivity with ChatGPT Mac App and Codex

Dan Sanchez: ChatGPT is more than just one app these days. I'd say ChatGPT is kind of a suite of apps and functions and modalities, and all kinds of use cases are spread throughout the ecosystem called ChatGPT. So today I thought it would be helpful to work through my personal workflow, how I'm actually handling all the different aspects of ChatGPT in order to be more productive, get more done, and actually get the most out of this $20 or $100 or $200 a month subscription that you got going on. So let's dive into it. Welcome back to the AI Driven Marketer. I'm Dan Sanchez. My friends, call me Danchez. And I got to start with my probably the most commonly used one is just the iPhone app when it comes to ChatGPT. Of course, this is very much on the go when you have questions, when you want to get things done and you're looking to kill some time. Right. I find more and more I'm scrolling less and I am working with Chat GPT more or just satisfying some kind of personal curiosity I have and diving deep and trying to find and understand a little bit more about the world we live in and how all these things work. So the iPhone is for that. But if you haven't done this yet, one of the cool things about having it on your phone is the voice feature. More and more as I'm on a drive and I'm listening to podcasts, I'll just ask. Siri to cut over to Chat GPT voice in order for me to start a conversation about something. Because oftentimes we'll have ideas while we're listening to that show, while we're listening to music. And we need to just talk them through. We don't have the ability to take notes down while we drive, but if we can talk through it and then have some things repeated back to us and maybe maybe even poked at it and push our ideas along farther, you'd be surprised how f how productive your commute can become. If you want to be more productive on the road and you have ideas all the time, try spinning up ChatGPT voice where you can actually just talk to it. Hi chat. How you doing? Hey there. I'm doing great. Thanks. What's on your mind this morning? And that's how the conversation usually starts when I launch into it. I'm going to close him up before he starts talking more. But the cool thing is the whole conversation is then loaded into a message that you can go back and summarize later or pull the best ideas as you had them on there. I've been surprised how often I'm just kind of verbally processing, and I'm not even a verbal processor. If you're a verbal processor, I can imagine this is even better. Generally I process best with pen and paper and just kind of sketching out an idea or just sitting down with chat and hammering it out. But with voice interaction. All of a sudden my ideas can actually take shape and I don't lose them. I don't have a great idea while I'm driving or doing something that requires me to focus or not not have my hands available to type. And I can get things done more quickly while I'm running or walking. ⁓ oftentimes if I'm just walking, I might just dictate and just kind of go back and forth that way. But I love having Chat GPT as just a standalone app on your computer. Of course, there's a little trick with chat in order to get more intense work done, but I'll circle back around. To that one. The next app I want to talk about is the Chat GPT app. This is the web app that most people like using on a day to day basis. But they've made it a little bit more confusing lately if you haven't been in it recently. If you come here, you notice they have it separated between chat and work. I don't recommend using work on the web app much. It's actually better to use the Mac app when it comes to getting a lot of good work done. But there are occasions when I do cut over to work versus chat. So let's come back to chat. And chat's a little simpler. You notice when I cut over to work, you can actually select the exact type of model you want. You right? You can go over to Astra and all the way down to ChatGPT 5.5, and of course the three different versions of ChatGPT 5.6 of Luna, Terra, and Soul and their intensity level. You can take, you know, you can pick Astra and go all the way up to like max ultra or what they call ultra, you know, reasoning, you know, if you need to go all the way. So work gives you more flexibility, but there's one thing that you need to understand the difference between chat and work. There are two different, I don't know, allowances or budgets that you have when it comes to credits with Chat GPT. There is one with chat and there is a different one with work that overlaps with the same budget you have using the Mac app codex. And that's where I do most of my work. So I actually use chat as much as possible if I'm doing something that doesn't require it to take action for me or do substantial reasoning for me. So all my more casual things or even my more light research, I'm doing in chat. In fact, I like to frame out ideas and programs and all kinds of things here in the the web app before I even take it into a more intense app like the Mac app or into something like Work. That way you can use your your credit allowance for all the ideation, all the planning and a lot of ⁓ lightweight lifts here before you actually start leveraging your credits, which can be much more intense because agents over in the Mac app or into work can spend through way more credits because they can spin up other agents. They can work for tens or tens of minutes or hours, sometimes days, if you let them go long enough and it's a big project. It could take a long time and a lot of credits. If you want to save money on those credits, To highly recommend. Like use chat for everything that you can and then shift it over to Codex for work when you need to. Now, more and more, they're like phasing out custom GPTs. So you haven't heard me talk about GPTs in a long time. I still have my old library, but I moved all of these to Codex long ago. No one talks about GPTs anymore. I think they're actually phasing them out soon. They're going to be gone, dead. They're still here. So what has happened now is that we're using projects. more and more. And they even have some like more sophisticated projects for finances and health, though you're sharing a lot of personal data with them. So use those at your own risk. I'm experimenting with both and I actually kinda like both. Especially the health one is pretty fun if if you have Apple Health data going into that. I have ⁓ placts for everything. I have tons of different projects for different areas of my life. All my different hobbies, all my different businesses or side businesses I ⁓ or employer like I have all the different projects organized there. And I try to keep those contained so it's easier to find later. And chat has more of course context about those different things. And these projects are very lightweight compared to something like Codex. They only need minimal context. The one thing that's changed with these recently is that you can actually save files here. So if I opened up this project for Trim Healthy Mama and look at sources, some of the sources are things that I've uploaded. Some of the sources are things that chat's created itself. I don't think There's a difference in sources, even from chat to work, which is helpful because then you can again use your chat credits and not your work credits. A new addition that they've added recently that I love is this library feature. Because under library, you now have a central place where you can see all the things you've uploaded to Chat GPT. And you can organize them by folders, which of course are your projects, ⁓ your images, and you can have all kinds of things here. Even your Google Drive gets listed here as a source. This is helpful because then you can open up. Things like I'll open up my Trim Healthy Mama folder and there's all kinds of documents in here. Some that I've uploaded here, but some that it's created itself. This is helpful because what if you want it to do some research and then save it to a document to reference in future projects? But then you want to be able to go back to that document. We used to have this feature called Canvas in ChatGPT, and that's kind of gone away. Why? Well, because we can it can just make you Word docs now. And you can open up these Word docs, put them in Google Drive and do all kinds of work with them. And it can reference it for future work. And it's just really handy because you can take this document and move it around different folders, duplicate it, and actually manage it kind of like you would your desktop folders, right? So chat GPT without even turning work on can edit and manage and move these folders around and actually do work. It's becoming more even just plain vanilla chat GPT is becoming more agentic because it has access to more of these files. One of the big changes, again, that's happened within the last few months is that its context window has gotten so much larger it can handle So much more of the memories and it's pa remembering past conversations. It can go and access these files proactively in order to get more work done. It's not quite as agentic and proactive as moving over to the work tab in ChatGPT. Again, it's at the top of the screen. If you haven't seen it, it's chat and work. Work it is again slightly different in that it's more proactive and trying to find context and getting work done and setting goals and working them through. Chat's a little bit less aggressive, and that's okay for most ideation and simple. work where you're brainstorming and coming up with things one thing at a time. So that's the web app. I sometimes code over to work if I'm on the fly. Generally I only cut over to work if I'm on my phone, but I need to get something more intense done. The biggest example of things that I use work for is when I want to do more intense research and I'm not already sitting in front of my laptop with codecs up in front of me. So I will open up work Switch over to Astra and then depending on how much research I wanted to do, I'll switch it from, you know, light on the light side all the way up to max or ultra if I wanted to spend a long time digging through everything in order to come up with research. And then you can have it compile the research in a doc. So for intense research, I am now switching over to work and using Astra for research. And of course, deep research is still a feature. on chat, you can access it, but I'm actually even using this deep research thing less and less. And I just asked Astro to go and research it for me now. You can still select deep research as a feature to kind of let it know, hey, we need to dive deep on this, but now it understands better and you can just tell it, hey, do a deep dive, do a big research. I want you to find everything. Just use language like that and it'll go and do a deep research project for you. So that's how I'm using the web app. And oftentimes when it comes to the web app, the iPhone app, I'm using the dictation button in order to delegate to it. I'm not really picky about how I format my prompts. I'm just trying to give it the context and the outcome that I'm trying to achieve. And that's about it. ⁓ it's more and more as it gets more and more intelligent, you can be looser with your your specifics. And it's pretty good at intuit intuitively trying to figure out what you're trying to accomplish. So lastly, let's talk about this Mac app. Okay. This thing is where the real work happens. And there's two different sides to it. There's the chat GPT side of the Mac app and the Windows app as well. ⁓ there's the ChatGPT side, and then there's the codex side. In your mind, I want you to get rid of the word codex and just put it with work, because it's practically the same thing, but it's better than work on ChatGPT, which is probably why they still leave it as codex. But do not think of codex as code. Think of codex as switching into work mode, really. And think about it more like that. Now it's a little confusing. So I never, I never, never, never use ChatGPT on the Mac app. I only ever use Codex. And I use it for way more than code. I use it for a lot of code, but I use it for way more than code. Okay. But I I highly recommend doing the same thing. Like use ChatGPT on your iPhone app, use it on your web app. But when you get into the Mac app, like just keep it in codex mode. Because this is where you can get real work done. Again, we have project folders like I've covered in past videos about codecs. Where these project folders are little folders that actually already exist in your documents folder on your computer. And you can have different files in there. You can have AI make files for it and have all kinds of things going on within these folders, but it tends to stick to work within the folder and have access to only the content in the folder unless you give it access to your whole computer, which I don't recommend for most people, unless you have a dedicated computer for it. And the cool thing about this is you can actually do much more substantial work, like the work I did for this episode. I have a whole process and it used to be a custom GPT. Now it's a whole Just a s a set of processes in my Danchez folder, right? For my personal brand, which this podcast is a part of. And I just kicked it off with the general idea for this episode and said, hey, let's find some angles. And it gave me some angles and it's walking. I don't have to tell it to run the process or what the process is. It already knows, right? Because it has a process doc for how to do a new episode. So when I say, Hey, let's start a new episode, it already knows what I want. It knows I want to flesh out angles and then come up with titles. And then find options for the i the the the image, you know, and then make all the images. Like it does all these things. And I just kind of move it along and decide when it's in between the tasks, the steps in the process. ⁓ and I've been using this for years now, but now it's way more sophisticated in codecs. And I don't have to make one image at a time. I can have it make all the images and they all look good because we're dealing with Chat GPT's image two point five generator now, and they're just freaking amazing. They've been amazing for about a year now, and I I That's why more and more I can tell YouTubers are using AI for part or all of their thumbnails because it's just really good. And of course, using codecs you see all the little like code edits. Like it's like, hey, I edited two files, plus five image ⁓ lines changed in this file, plus seventeen. I don't really pay attention to that. I don't really care. If I'm even if I'm coding, because I just like I don't know enough about code to even validate whether that was good or bad. I just let it go at it and build things. And I find more and more I'm able to be more productive, especially with the upgrade of Astra in between, which since the last video I've gotten access to, and I can tell you, is so good that I absolutely upgraded to Pro for at least the next month in order to get some more substantial work done with it because it is a little credit hungry. But it is an absolutely fantastic, I'd say game changing step up from soul. And What's good to know is that using Astra Lite is actually less expensive and faster than using using something like Soul High, extra high, or even just high. So I've pretty much switched between just using Astra or Terra or Luna. Like if I'm doing low level tasks, I'm just using Luna or Terra now and Astra more and more. Astra lights kind of become my go to for most work that I'm doing. If I didn't have the pro right now and I was just on a plus plan, I would probably just be using Terra most of the time and then only leveling up to Astro when I had some pretty heavy lifting I needed to do. But since I'm on a pro plan, I'm using Astro Lite pretty much as my default thing for everything within Codex. Now, of course, with ChatGPT or the iPhone app or the web app, then I'm just using, you know, medium reasoning of whatever model it's giving me. But more and more, I highly recommend. This is the next level. If you're not using the codec side of the Mac app or the Windows app, this is where a lot of the power is in AI. Go and watch tutorials on how to get the most out of codecs, how to code your first project, even if you're not sure if it's going to be the most helpful product or just tool that you're making. It's becoming the go-to tool for doing heavy amounts of work. And I'm hoping to do some more videos on how to organize codecs and actually get more done with it. I'm probably going to be doing a tools video soon on the types of tools that I'm creating. With codecs, how I made them, and how you can build similar tools for yourself. But hopefully this kind of gives you a lay of the land of how I'm using ChatGPT, how I'm organizing it between the phone, the voice, the web app, the Mac app, and what kinds of projects that I'm using for which one. Let me know if this has been helpful to you and drop a comment below.