ChatGPT vs Claude vs Gemini: A Marketer's Guide to AI Assistants

Hey, folks, Jacob here flying solo on this episode while Colin's off at an event. I think he's actually in the middle of a talk right now as I record this. So today I want to talk about something that I use a lot. Talk about a lot. AI tools. Specifically AI assistants. I do talk about AI tools an awful lot on this podcast and you've probably heard me mention Claude quite a lot even on the blog, written some posts about it. There's. That's because it's the best for writing.

And as a marketer, writing is a really big part of my work. I use it almost every day. But there are other tools that I use for different tasks and they each have their own strengths. So today I'm going to share how I actually use Claude, ChatGPT and Gemini. Not based on theory, but from my actual experience. And before we get into it, if you are enjoying the show, please hit subscribe. We share weekly insights to help you build your creator business. All right, first up is OpenAI.

So let's start with OpenAI's O1 model. This is new. It's quite different from regular ChatGPT GPT4. It's what's called a reasoning model. Instead of just answering questions, it breaks down problems and creates structured plans. I use it a lot when I'm tackling big projects or when I need to

get past that blank page anxiety. Now, for example, if I'm planning out content Strategy, I'll ask A1 to help me identify the key components that I might need to consider, break it into logical steps, highlight any potential gaps or risks. I want to just get a start. I want to get the wheels turning. And sometimes when you are starting something afresh, you're just not sure where to start. And I think that's what O1 is really, really good for. It's those strategy problem,

it's those deep thinking problems. I even, I mean, I use O1 to structure plans, feeding them into other AI models. And I actually think that's really, really useful too. I will expand on that, but it's particularly useful when you need structured thinking, not just fast answers. I wouldn't use it when you need a quick response or when you want to have a sort of, I don't know, something more similar to a conversation. It's not great at conversation.

It's. It's all very structured. It's all very kind of cold. Don't use it to write. It's not very good at writing. It will never, or certainly I've never managed to get it to replicate my tone of voice very well. Now, there's some other gotchas. 01 doesn't have access to attachments, so you can't upload files. You can paste in content and sort of chat with it about that, but you can't add files. You can't browse the Internet in the same way that GPT4O can.

But it's a really, really good thinking, deeply breaking things down and making good, concrete plans. And I use it fairly often.

The next tool is Claude, my favorite, the one that I always, always talk about. And I talk about it because it's excellent for writing, and I've set it up to help me with all sorts of content. I use it to draft blog posts. That first draft, when I've got a podcast transcript, we've talked about something on the show, and I want to put an article out on the blog about it. And not the fastest writer in the world, and I don't have days to write an article. So I get a really

good first draft off that. I just feed it the whole transcript for that episode. It fits in just fine. And I get that first draft and I do my bit of work on it, and within a few hours it's ready to go. Which means that the same day that we record, I can get that transcript and I can actually get that post out often before we publish the podcast episode. That's really, really handy. Emails as well, really good for that is really good

at anything that you want to. You want to be able to coach it into a certain style of writing, a certain tone of voice. I find that in ChatGPT4.0, not.01, but 4o GPT4o. I find that to be quite robotic. I find it to be. I don't mean this as a. As an insult. I find it to be really American. And what I mean by that is sometimes a little bit too emphatic. And I find that Claude has the cap. The capability that has the capacity to be a little bit more

down to earth, a little bit more the way that I write. And it does take a little bit coaching, a little bit prompting. That's probably a topic for another time. In fact, I've got a post on the blog all about that. If you search Google for AI writer Claude, it should come up on the first page and that will show you, actually some of the prompts that I've used. Those are now wildly out of date. I've got better prompts now that I use, but it should give you

good direction. Yeah, really really good for writing. It's quite good as an all purpose tool if you don't want to have multiple subscriptions. I would probably go with Claude because it's decent for planning, it's decent for. It's. But it's the best for writing and it's decent for analyzing transcripts and longer context stuff. But it is limited.

And that kind of takes me on to the third tool which is Google Gemini. Now I was a little bit skeptical, skeptical, skeptical about Gemini because I didn't really see use case for it. There was nothing really that it offered in my testing which I couldn't do with Claude, I couldn't do with ChatGPT and yeah, I didn't use it for a long time but actually I found myself using it more recently. And it's because of one kind of crucial thing that it does better than anyone else and that is it can take

a lot of text. It has a massive, what we call context window which is essentially the number of. The number of words that you can give it without it forgetting the first words. You know, it has a. There's a sort of, it's called a context window because let's say you start a conversation in Claude and you give it a really, really big book, full massive PDF and the truth is that it will answer questions but it can only access up

to a certain point. Gemini has something like four times as much capacity for that. And I was trying to repurpose some things from our finally start your podcast book, our book on how to start and launch your podcast and tried to feed it into Claude and it just was too big, it was too long. But Gemini could handle it no problem. It all fit in there. Now it's not great at writing. I wouldn't use

it for writing. It's not great at keeping a consistent tone and it's not certainly not my go to for any sort of structured planning. O1 is better for that but it is really good for data heavy tasks and it's probably not why you're here, but it's decent at writing code as well. Incidentally, probably the best of these three at writing code is Claude 3.5 sonnet specifically. I actually, I use that often to write code and it's. And it's really, really good.

Yeah, those are the three tools that I use the most assistance anyway. There's all sorts of tools out there that use the same models under the hood and they've built some software

around them. They're sort of playfully called GPT wrappers and there's a lot of useful tools out there that are sort of purpose built, that are prompted in a certain way to work and you know, they do useful things and actually I find myself building some of these tools and there's, there's value to them. But I think if you're trying to get, get a grip of where to start with AI or get a sense of how it works so that you know what the limitations are and what you can and can't do with it.

I would start with one of these assistants, one of these three assistants and it depends on, on your use case. If you have to pick one, go for Claude. That would be my recommendation. But if you have the budget, subscribe to all three. Now, all three offer free plans. You won't always get access to the latest and the best and you might get a certain number of messages that you can send a day, but it'll certainly be plenty to give you an idea.

One thing to Note is the O1 model is only available on the paid plan for ChatGPT, so you won't be able to try it. But they're newer, but it's confusing. It's called O3 Mini, which makes you think, eh, it's got a bigger number but it's mini. So it's probably about the same actually for a lot of things. In my experience it seems to be better than 01 and you can actually use O3 mini for free on their free plan.

So if you want to get a sense of what a reasoning model can do to help you plan and kind of dig into that Meteor strategy stuff where you just need something a bit smarter that's not just going to give you the first answer that comes to its head. Try O3 Mini on a free ChatGPT account. It's. Yeah, it's really cool. If you've been paying any attention to the news in the past few weeks, you've probably heard of Deepseek R1 Deep seq R1 is an open source, which means that anyone

can sort of download the code for it. An open source version of 01. We talked about it on the podcast a while ago. There's a bit of controversy around it because it comes out of China and people are concerned. Probably don't need to get into this right now. I want you to make this a quick and useful actionable episode. But I would say that the fears are largely overblown and that they're. One of the big things is censorship.

There is censorship if you use Deepseek's own app but if you use something, some third party app that can run the model too, you won't see that censorship. The censorship is in place because it's a Chinese company and the government in China mandates that it censors certain things. So, I mean, make of that what you will. I've used it, I've found it useful, to be honest. Probably still prefer OpenAI's offerings over it, but it's very, very impressive given the fact that they've essentially

given it away for free. Yeah. So that's kind of cool. But yeah. So those are the three tools and different ways that you can try them, I think. One other thing that I want to talk about, there's a few other things actually, but they all work together quite nicely. So here's how I would actually use an example of how I'd actually use. I would use O1 for planning. If I'm starting a new project, I would start with O1. I would use O1 to break it down into logical steps, create a structured plan.

And then once I've got that, obviously, you know, you want to spend some time going over that yourself. You don't want to just blindly take what it gives you. You could then go to Claude to do some writing. Now depends. A lot of the projects that I find myself using it for are either writing content or writing code. Either way, I start with 01 and once I've got an outline, I've got a good. Yeah, a good plan for it. I will go to Claude

and that's kind of where I make it happen. That's where I do the writing for content. That's where I write the code. It's good for refining as well. And you know, you're not just stuck to the, to the first draft that you give. You can give it, you can give it feedback. But often I find that from my experience, taking the first draft and then just doing, spending a couple hours on it myself is probably best. Because otherwise you'll end up going back and forwards, back and forwards, saying, no,

change this, change that. And actually it's just, you know, how it needs to be and it's much quicker, like it's much quicker to make those changes yourself. And rather than going back and forwards, unless there's some things that are just like really, really wrong and you know, you would wholesale need to kind of rewrite

massive, massive parts of it. In that case, you're probably better off just doing a couple of drafts and then taking it out and put it into your favorite writing program and doing that Human Pass on it. And I think that human pass is really, really important, whether it's code, whether it's writing content, whatever it be. Gemini is a little bit trickier because it's such a,

for me, it's such a specialist tool. So like I use it when I need to, like I said, when I need to pick apart really, really big document like a book, like our. Finally start your podcast book. It's 12 chapters, it's over 140 pages. It's really, really, it's really long. And it's just too much to put into Claude or ChatGPT and have it work. Right.

And even worse is that the longer that conversation goes, it'll start to cut off the beginning of the conversation, which means that it'll start to cut off bits of, you know, after so long it's just useless. But Gemini is really, really good for that. And you know, I'm talking about. Gemini is the third thing in this, in this kind of process. But actually it might make sense if you were

depending on what it is that you're working on. Like if you were wanting to, I don't know, make a content strategy for your podcast. You might get transcripts for 10, 15 of your latest episodes and you might feed them into Gemini and you might ask it to start mapping out some kind of content clusters and pillar content suggestions. And I think it would be really, really good for that. And then you can take that over to O1 to do the

actual planning and think about it in a little bit more detail. And then you could take that to Claude and start doing some of the writing, getting those first drafts on. Like there's lots and lots of different ways that you can use it and you're only ever going to figure out the best way for you when you give it an actual try. But that's how I use them anyway. That's my experience.

So yeah, I would recommend trying them out and just try and play them to their strengths. You know, it's a toolbox creator toolbox. Haha. Yeah, they've all got their strengths. So have a go and see what you think, see what you feel. Right. There are some things to be aware of. Do try not to use one tool for everything. So don't. I mean, I'm kind of going against myself here, but if you were going to use one tool for everything, I would say Claude. I still think though that it is worth having.

It's. If you can afford it, it's worth having at least Claud and chatgpt subscription paying for both of them, you get much more usage, which means that you're not going to get halfway through a project and suddenly get told, oh, you can't use it for 24 hours. You know, that does happen and it's very, very, very annoying, especially when you're in the middle of something. Even just for that

fact alone. Even on the paid accounts, you can get limited. So if you're really going at it, especially when you're kind of experimenting and trying to figure out how it works, certainly in the early days, my goodness, I got limited all the time. But being able to switch between the two, it can be quite helpful. Now the other thing to be aware of is the quality of your output is always dictated by the quality of your input. AI isn't a mind reader.

The more detail you give, the better the output. Now practically, that might look like, you know, to go back to this content strategy example, you've got 10, 15 podcast episodes. Try to, try to be careful about what you give it. You know, certainly you could, you could give it the full podcast episodes, but exclude any that you don't want to include in that content strategy. Don't just give it more context, more content, more instructions

for the sake of it. Another little tidbit that I've picked up along the way is you'll have heard of the idea of prompting of the messages that you send and kind of having an impact on the, on the result that you get. That's, that's, that won't be new to you, but the way that you write the prompts makes a difference too. If you.

My prompts tend to be pretty long because I've just got, you know, over the, over the years, now that I've used these tool, I've picked up things that often happen that I want to tell it not to do or things it doesn't do that I want it to do. And you know, there's just, you want to be quite precise in what you want. So actually my prompts often will be 100 lines long, something like that, you know, 500 words,

something like that. But if I have a 500 word prompt, the way that I actually write that prompt makes a difference. So if I write that in my tone of voice and my style of writing, and it's a prompt explaining how I want it to go about writing an article, you will see that voice better reflected in the output. It's sort of a stochastic parrot is a term that I've heard thrown about when describing these AI models. And it's Very, very true. It is just a really smart parrot

in some ways. I think there's arguments to say that we're kind of going beyond that now, especially with these reasoning models. You know, we're able to think outside the box a little bit more, but it's still largely true. It will reflect back at you what you give it. And in that sense, you do have to think carefully about the context that you're

given. If you want sort of anthropomorphic way of thinking about it, imagine that you've just hired a brand new intern who is, you know, not even out of university, has absolutely no idea what you do, no idea about the industry that you work in, no idea how to do their job, essentially, you know, just complete babies. That's kind of what you're dealing with. But it's the smartest intern that you've ever, ever worked with because it will do as you say if

you tell it to do that. Another example I've given before is that if I, if I go to ChatGPT and say, Write me a blog post, just very simply, verbatim, write me a blog post on how to start a podcast, it's going to give me really, really, really generic advice. But if I go to ChatGPT and say, Write me a blog post on how to start a podcast using this framework. So we've got a framework that we have devised over the podcast host for that initial bit of planning that you

need to do when you're starting a podcast. It's P, A T H, purpose, audience, topic and hallmark. If I give it that, that suddenly changes everything. That then cascades on to not just that initial planning bit, but actually it will affect the. It'll affect the output for the rest of the post and everything else that needs to go into launching a podcast. And suddenly that becomes much more mine than any other person that could go and ask it to write me a blog post about starting a podcast.

You know, actually if you ask it that, if I ask it that and you ask it that, we'll both get really, really similar answers. But that initial prompt, that initial context that you give it really, really changes the output that you get. And you can go to the nth degree with that, you know, you can get to the point where you're almost sort of half right in the entire thing and you just use the AI to pull it all together. And I find that's really, really useful as well, because you could

even make a voice note, you know, and just. I don't know about you But I, I think messily, I talk messily when I've not thought properly ahead of time. But I find it really, really useful to just brain dump verbally into a voice message, get that transcript and chuck that into an AI model and have it, you know, organize those thoughts for anything. You know, it might be an idea for content, it might be an idea for a little bit software, it might be ideas on how I want to manage my finances

this year. You know, it doesn't matter. Yeah, that's a really easy way to get good context. Brain dump into a voice note, get a transcript and then put it into the tool that you're using, the AI assistant that you're using. Yeah, I would really, really recommend that. I've done it myself. There's entire tools that do that for you. But like I said before, I think you're better off starting with one of these assistants and then maybe looking at some of the more specialized tools

that have been built around them. Yeah.

And I think lastly, pitfall to keep an eye out for is alluded to this, but don't accept the first output. Don't be afraid to, sure. Ask for redrafts, but if you're not happy with the output in the first instance. So that first message that you send that first prompt, you can, in all of these assistants, you can edit that prompt even after you've had a response and it'll regenerate a response. And I find

myself doing that quite a lot. You know, if it just, I realized that it didn't pick up on some detail that was in my head that never made it into the prompt, I can go back and change it and expand it and you know, and add a little bit more guidance on this or that and it'll regenerate the whole thing. And the important thing about that is that you're not messing up

the conversation. What I really want, what I'm looking to do when I'm using like say Claude for writing, is I want that first message from me to be the instructions. I want the second message to be a really solid draft because it means that the third message, the one, my second message, can be all about writing a second draft. If we get stuck in just trying to get that first draft right and to try and get maybe get a good outline made or whatever it is, you need to be careful about getting

off track. You know, the more messages you add to a conversation with these tools. Yeah. The worse the response down the stream is going to be. You want to, you want to manage that. You will get a sense for that when you use it as well. I'm probably making it sound more complicated and difficult than it actually is, but these are just things I've learned along the way. Little, little things that just make it really viable tool to. To help me do my job.

And I think it would be really useful for you. But don't worry too much about doing that in the first instance, or if you're already using these tools and you're just wanting to get a better output from them, make them more useful, then, yeah, definitely do implement these things. And I think. I think that's. That's probably about it. God, this is gonna be this one of the shortest episodes we've ever had for creator Toolbox. But, yeah, I had to. I had to try this out. I think the column was out.

I didn't really have any guests off the top of my head, so I thought, I'll do a solo episode and see how it go. And, I mean, I've quite enjoyed it. We'll see. We'll see. I'm sure you guys will tell me if it's rubbish.

So. Right, let's wrap up. Here's what I think you should go and do right now. If you want to take action on this. Start with the problem, not the tool. Think about what it is that you actually need to do. As I said, if it's planning, I recommend a 1. If it's writing, I recommend Claude. If you're working with large files, I recommend Gemini. But start with the problem, not the tool experiment. You're just going to have to try it and build your own workflow. I mean,

my workflow changes all the time. I could tell you today and it'd be different tomorrow, but you will get into a rhythm with it and you'll get a feeling of what works and what doesn't. And you're only ever going to get that by trying it. You know, these. Yeah, there's. It's. It's hard to pin down. It's hard to pin down something that'll work for you, that works for me, because it's all very. I don't know, slightly otherworldly in ways. You know, I like. I find that saying please and

thank you helps. And I don't know if that helps me or if it helps the AI, you know, so, yeah, you're just gonna have to get a feeling for yourself. And I think probably the last thing to say is these tools move really, really fast. Everything's evolving really fast. I mean, a few episodes ago, I spoke about my feeling that this year maybe things will slow down, and that remains to be seen.

But I think it's a good idea to keep on top of the major, major moves, especially ones that concern creators. And if you subscribe to the podcast, then I will make sure that you are always on top of everything that matters when it comes to AI tools for creators. So if you haven't already, subscribe and I'll see you next episode when Colin is back.