
Join the Target Internet team for a lively debate dissecting three of the most talked-about AI models in marketing today. In this episode, the trio evaluates the capabilities of ChatGPT, DeepSeek, and Google’s Gemini, sharing firsthand experiences...
Loading summary
Kieran Rogers
Welcome back to the Digital Marketing Podcast brought to you by targetinternet.com My name is Kieran Rogers.
Louise Crossley
I'm Louise Crossley.
Daniel Rolls
And I'm Daniel Rolls.
Kieran Rogers
And today we are discussing ChatGPT versus Deep Seek versus Gemini.
Daniel Rolls
Okay, so if you've been following the news, the new AI, Deepseek, which is a Chinese based AI, was launched. Generative AI tool. And it caused a huge shock to the whole AI kind of all the other companies that existed because they managed to create this model at much, much lower cost and using much less processing power than ChatGPT and Gemini and those kind of things. And it really caught people off guard and it caused a huge crash in, in the stock prices of all those kind of technology companies as well. Now, interestingly, a professor that I work with, Imperial College, who I won't name, said to me, this is nonsense, this can't possibly be true. This. There must be some sort of kind of dubious baby going on here. I'm going to buy loads of Nvidia stock. And that immediate moment he went and spent £10,000 on Nvidia stock. And then the story came out the next day that actually maybe they cheated creating this AI by using ChatGPT and a slightly, you know, a legal way, which hasn't been proven yet, I should point out. But immediately next day everyone went, oh, yeah, I see this tr. And the stock price rocketed back up and he doubled his money overnight as well. So that was a brave move, but we've been quite interested to test it out. And I got a message from Kieran saying, I've been using it, it's amazing. And I went and had to play with it. So we've done a bit of investigation. But before we get going, I have a couple of little things to talk about. Now, you may be aware of the mug story, and the mug story is we had all these mugs printed and we had all these Hostess printed and they've all got Kieran's face on them. And he wasn't even aware we had these things and he certainly didn't have one of those things. And he was a little bit put out. Then he found out that he'd been sent a hoodie, but it had been sent to me pretty much from Swido. Thank you very much. It's very comfortable. So, you know, fairly, fairly, you know, a little bit put out about these things.
Kieran Rogers
Miffed, Miffed.
Daniel Rolls
From the island now, got his own back a little bit because. What have you got, Kieran?
Kieran Rogers
Yeah, so I have hot off the press Copy of Daniel Rouse's new 4th edition digital branding.
Daniel Rolls
I haven't got. I haven't got a copy of yet. I haven't even seen it. He's got a copy of my book before I've even seen it.
Kieran Rogers
So, made my day. I casually throw it, oh, thanks so much for the book. And you're like, what? I haven't even had that yet.
Daniel Rolls
I'm like, yes. And then another little thing said that we do a bit of work with the Arts Marketing Association. Lovely, lovely group of people. And the very lovely Carol Jones, who works at ama, during a meeting. Just dropped into the end of the meeting. Can we have more Louise, please? This was an episode that Louise and I did on our own together as well.
Kieran Rogers
I like the sound of that.
Daniel Rolls
So you just want a day off, don't you? That's all it is. Yeah, I'm waiting for it now. She finishes her emails with hashtag more Louise, please.
Louise Crossley
It helps that it rhymes.
Daniel Rolls
It does. I mean, she started a thing, so there you go. So on that basis, Louise, we're gonna have to book in some more. Some more dates for you to have some rants on your own and things like that as well.
Kieran Rogers
We could have rival hashtags. We could have more. Lou, please hashtag more Louisiana. Can.
Daniel Rolls
Kieran Daniel. Yeah, exactly. So, okay, let's move on. So just this thing about Deep Seek, I wasn't kind of that sure about it. First of all, having used it, I am very impressed. But we've run a test, so just to introduce. The test that I've run is to go through and I've given the same prompt to ChatGPT to Deep SEQ and to Gemini and to see what the results were and try and compare them. But before we do that, before you go running out there to install deepseek, Kieran, just tell us you did a little bit of research and you found some interesting things in their privacy policy.
Kieran Rogers
Yes. Do you know what? It actually wasn't me. I've got to give a shout out to the amazing Matt, who I work with on lots of. Lots of projects. And it was Matt that picked up on this map is very much privacy like Guardian, and he very, very good on, like, security issues and stuff. So, Matt, Stacy, if you're listening, which I know you're not, thank you for being brilliant at what you do. But, yeah, he spotted that. Deeksee seeks privacy policy notes that the company collects and will use many types of data to train new models, such as text, audio, prompts, feedback and chat history, shared with the chatbot, user account information, personal details, data about users, devices, operating systems, crash reports, key straight. That's the terr bit, cookies and IP addresses. Yeah, it's like whoa. Actually the full nine yards. So they're not saying they definitely do, but they're. They collect and may use that.
Daniel Rolls
This was similar to the TikTok policy that's created all the challenges of TikTok about privacy.
Kieran Rogers
Yeah.
Daniel Rolls
The idea that it can track your keystroke patterns across other apps is that that could you be logging in, that could be your password potentially. I mean it's not saying that specific.
Kieran Rogers
Well, it does say keystroke patterns, it just actually referenced that. But it also adds that users operating it on a local walled version. Deepseek sends collected data to servers in the prc. But if you're operating a local walled version of Deep seq, it doesn't do anything.
Louise Crossley
And what does that mean, a local walled version?
Kieran Rogers
Deepseek is open source, so you can actually run it on your own servers if you want to. And they need some technical expertise to do that. Sure. It's not a touch of the button kind of thing for us mere marketing mortals, but.
Daniel Rolls
Well, I was just going to say, you know, I was thinking even if you run a local version, is it not, you know, there might be some backdoor stuff in there. It's a bit dangerous. But actually agent AI, which we've mentioned before, which is something that HubSpot have created or the, one of the co founders of HubSpot created where you can play around with AI agents, you can get them to request stuff from each other and there's been loads of changes. You haven't taken a look. Go and take a look at it. You can register for free, but they've actually now set it up so that you can create these agents that do various different tasks, but you can choose which AI they use. So they use ChatGPT. But actually now what they've got is a local version of Deep Seek that you can use on there as well. And they say it's completely secured, it's hosted in the US and it's, it's, it's good. They would have gone through and checked all the security stuff on this as well. So I think that's fairly reassuring that that local version is a, a secure thing to run as well. So interesting.
Kieran Rogers
It's the fact that they're using all these different identifiers to profile users activity outside of the AI model that raised my eyebrows. It's like, oh yeah.
Daniel Rolls
And that's Why I would probably use the web version as opposed to using the app version. She's not actually installing it onto your device from that point of view as well. The other thing I would say with that is that whenever you're a little bit unsure of the particular app or the particular website with your passwords, just don't reuse passwords. I've. I got into a terrible habit of reusing passwords to the point Louise would ask me the password for something because you kind of already know because I'd use the same password so often. So I'm. Now we use 1Password and this, you know, it will generate passwords for you ridiculously long, ridiculously strong passwords and save all those across all your devices. So a little bit of a Shout out to 1Password because it is a great solution. But there is something built into Max now that will do a similar thing as well. So I went to each of these AIs and I gave it this prompt and I said, please write an article titled the best Online Digital Marketing Courses, which is a review of English language online digital marketing learning platforms. Make sure you include Targetinternet.com and Google Garage. Include a comparison table of all courses with a comparison of all features and functionality. So that was run. Yeah. So we ran it across the three different platforms and kind of fascinating results. So when I went into Gemini, they've got this new deep research mode. So I thought, oh, we'll give that a go. So I went into Gemini and it went off and said, I'm going to look data up. And it listed off 58 different websites it was looking at. I said, wow, this is incredible. And it said, I'm going to create a Google Doc. So suddenly it created this external doc and I was getting quite excited about it. And then I opened the doc up and it had a headline and half a paragraph and then it had broken. So, so then told it it was broken. It went, yeah, oh you're, you're right, I'm sorry, I'll, I'll fix it. And then it did nothing. And then I ran the prompt again and it did exactly the same thing. So then I went in and changed the model on Gemini to an older version that's been around for a while, which I thought would be more reliable. And it basically wrote a four paragraph article of no depth and really didn't give me it. So it was a bit of an abject failure with Gemini. Now, not to be completely critical of Gemini, but I am finding generally it's less capable and I've got the Pro version as well than some of the other models and it tends to give you a lot of fake links, so you ask it for references. And I've, I've had some real, real problems with it doing that in particular as well. So keep an eye out for that one.
Louise Crossley
I keep getting errors in my Google AI overviews as well, facts that are incorrect.
Daniel Rolls
So, so what are the top, the search generative experience stuff that you get the top of the Google results? Yeah, I thought they were going to be ramping those up, but if anything they seem to ramp them down a little bit recently. So they're obviously experiencing some problems with it or it's not quite having the user experience that they were, they were kind of hoping for as well. So, okay, so Gemini was a bit of a disaster. We're not going to give up on Gemini, but it seems to be struggling at the moment. Then went through and ran it in ChatGPT and in deep Seq. And what was really interesting is Deep Seek and this is something you'd point out to me. Kieran then gave me this extremely long logic of what it was going to do. It told me the thinking process behind what it was doing. Now what's interesting is ChatGPT4 and 4.0 don't do that generally, but when you use 01, which is their newest model, that's very good at reasoning, there's a little arrow that you can ask it to tell you its reasoning. So I thought, oh, that's very similar to that particular version, but you only get that in the paid for version. I then ran it and what I ended up with in ChatGPT was a summary paragraph at the top. Then it went through, I think it was eight different kind of online learning platforms. So the Google garage, the Skillshop, Udemy, LinkedIn Learning, all those kind of big competitors that we got. And it said a bit of a summary of what it was about and then it went through and gave the pros and the cons and then it suggested who it might be useful for. It was really good. I mean it kind of understood that we had breadth and depth and we were for like professional marketers, where the others may be a bit lighter touch and so on. So I was pleased with that. And what was really interesting is when I did it In Deep Seq, DeepSeek gave me a summary paragraph, it went through each of the tools, it did it in exactly the same order, it told you the pros and the cons and then it said who this was. For it was almost the same response. So what I was like, okay, this is remarkably similar. This is living up to this. Did they use ChatGPT to help train Deep Seek? Because that would explain some of the similarities. And I ran it again on ChatGPT and I ran it again on Deep Seq and again the two of them look like a variation of the same answer. So to me, to my external eyes, without, you know, the proof of this, it does seem that they've accessed and apparently the story was they think that they accessed the API of ChatGPT, but they managed to kind of hack it. So it was doing it really, really quickly. Now I'm stating this as a news story I've read, not as a fact, because I don't think there's any proof and it could be completely proved, completely wrong. But what that means fundamentally, if we look at this, Gemini was a bit of a washout. ChatGPT, the paid for version was very good, but the free version of Deep Seek is as good, in my opinion as the paid version of ChatGPT at the time I tested it. So I just think it's interesting that we're seeing this evolution of this and all these models are really kind of out competing each other to try and give you the best results there is.
Kieran Rogers
A lot of. The server is busy. Please try again later on. Deepseek, no doubt.
Daniel Rolls
Yeah, I mean, I think to have the infrastructure that you'd need because the amount of global interest they suddenly got was, was quite incredible as well.
Kieran Rogers
Yeah. Widely reported the number one in all the app stores.
Louise Crossley
Yeah, I think in January it was the most downloaded app in the iOS app store.
Daniel Rolls
Yeah, right.
Kieran Rogers
And I think everybody's testing out because it's like, well, you know, we, we love chat GPT, but if we can save $20 a month and get the same result as. But it's bye bye.
Daniel Rolls
My students will 100% look at it and go, a lot of them won't paid for a pay for account, but if they suddenly got something that's getting better results, they'll go through and use that.
Kieran Rogers
What I really liked about this, explaining its thinking is that it brings back deep thinking into the process of being creative. Now one of the things I think we should all recognize is that, you know, the danger is that we don't think anymore because actually it's easier to just ask the machine to think for you. And it's very, very good thinking things through. And actually we're not exercising those muscles. And as any, you know, sports technician will Tell you, you know, if you don't use those muscles, they waste away. And I think it's, that's, it's a real concern, right? Like what happens if we have a whole generation of marketers who've never actually thought stuff through or don't actually know the theory. They just ask the machine to, to tell them what to do. And that's a very, very real danger. So I think that that gave me a little crumb of comfort that actually, actually maybe these things can encourage us to stretch our brains.
Daniel Rolls
Well, I was just going to say with that, I mean, you can get the O1 model to do it as well. So what's, what's really good and the way the technique I've learned from this is when you get really good results from one of these kind of prompts, take that reasoning bit that it's given you and then use that as a way more detailed future prompt and keep building on it. Because what you fundamentally see is that it's kind of told you the way it's going to approach your problem. And there may be things there you like, there might be things now that you want to repeat and so on as well. So one of the little techniques is that you go through, you do these prompts, you look at the reasoning. It's done using deep sea core chatGPT. You take that and you create a custom GPT with it if it's a repeatable task. Because a custom GPT, you know, is something you're going to do on a repeated basis for doing a particular function for you in CHAT GPT and actually having that level of detail and building on and expanding on it. I found you can get some really good results out of those things as well.
Kieran Rogers
It just worries me that we're all about to become like those characters in that Disney film Wally, where it just sits in a chair, it's like a big lump.
Daniel Rolls
I'm already like that already. So I wouldn't worry too much.
Louise Crossley
Very black mirror coded, isn't it?
Daniel Rolls
Yeah, I, I, I think the thing is here, did people think this when the abacus came out? Did people think this when the calculator arrived? Oh, you won't be able to do anything anymore. It's just that, But I don't think.
Kieran Rogers
You can argue that the calculator and the abacus had quite as.
Daniel Rolls
No, I'm not suggesting that. If you look at the Industrial revolution, right, when suddenly, you know, all the factories were automated, things like that, well, we'll just be sitting around all day doing nothing. The reality is it just evolved and we moved into doing different things and focusing on different tasks and I think that's already happening. But I do think the amount of tools that have arrived to automate your social media and it's like, whoa, whoa, whoa, you can take it too far and you're just going to pump out rubbish.
Kieran Rogers
Well, it's like I was saying, the job that we recently posted, we had hundreds of applicants and they all had almost identical CVs and like mission statements and like achievements and everything, everything was written in a certain, A certain way. And like, hats off to surfer SEO. They've probably worth including this in the show notes, but they've created a brilliant, like, was this written by AI tool? And it's free.
Daniel Rolls
Oh, really? Okay, because there's a few of them out there, but I haven't seen many free ones.
Kieran Rogers
But this is really good, really, really good. And I think because of their SEO background where they're, you know, looking at lots of kind of dot document frequency algorithms that look at like phrases that occur again and again. Like, I think theirs is a little bit more sophisticated. Like a lot of them try and just find phrases that, you know, ChatGPT is known to, known to use. This is a lot more sophisticated than that and it gives you a likelihood of how much something was written by AI. And I think it's a shame that tools like that even have a place. Right. But actually it's kind of key. I think for recruiters it's really key. There's nothing wrong with using AI, but are people being honest about it? You know, I think that's, that's the key question.
Daniel Rolls
Yeah, look, I mean, there's loads of tools out there that will do this. And the thing is, we did some research at Imperial College and they basically can use AI to detect AI generated content. And the answer is, if you're any good at using the tools, no, you can't, because you can take your content, put it through something else or add some manual editor to it. And then actually, because the AI is getting better and better at writing like a human actually, you know, it becomes, it becomes impossible to do it. So I think there's a skill of using these tools effectively and I think it's what it's eat, right? It's that experience, expertise, authoritativeness and trust. The Google thing, it's a similar kind of thing of saying, are you actually providing any value to this content? So I'm really interested in the fact that what Deepseek has kind of switched us onto Is this fact of the reasoning, the logic behind it. And there's a new model that's out from ChatGPT, that O3 model, and that's looking at reasoning, but doing it much more quickly as well. So this is, this is obviously, you know, a core focus because that, that ability to do critical thinking is, is still important for humans. But it just means we need to make sure that we're looking at the process that the AI is going through in order to create the content.
Kieran Rogers
Read them when you get them, because there are some, there are some odd statements that are made in there. Like one of the things it said to me was I asked it about a particular technique and it said, oh yeah, I remember reading about that in a lecture that I attended. And it's like, what, wait, what? Where's the training data for this come from? Is it actually pulled in and referenced like someone's memoirs?
Daniel Rolls
Well, I mean, that's. If it's looking at Reddit and it's looking at books that it's, you know, all those kind of things, then yeah, maybe, maybe.
Kieran Rogers
But it was all part of the, part of the conversation. But I've absolutely loved seeing its logic and seeing how it builds. It's like, well, before I answer that, I need to take a step back and look at all the different parameters. And it's definitely using anything that you've told it already. Like it will reference earlier questions within the same chat and consider whether that's relevant for the answer it needs to give you.
Daniel Rolls
I mean, from that, one of the things that we picked up from Keywords Everywhere. So Keywords Everywhere has got a load of prompt templates built into it. Even the free version, if you haven't installed the Chrome extension, it's one of our favorites. But they've got a new thing now that when you go through and you've got the paid for version, if you have a paid for account for ChatGPT or anything like that, what you can do is, is you just search in Google. It will show you all the related search phrases. But on the right hand side it's got a button to launch into Claude or into Gemini or in to ChatGPT and to go through and do you a keyword analysis report using the AI, which is a really nice little bit of functionality as well. So if you haven't checked out Keywords Everywhere, I should say we have no comm. We talk about this, this Chrome extension constantly, but we have no commercial association with them. Maybe we should have done. But yeah, yeah, just love it. And it's doing. And they keep on adding to it as well. So I'd take a look at that. I found that to be really useful.
Louise Crossley
It's, by the way, reasonably priced.
Daniel Rolls
It is really what is like 2.25amonth or something like that for the.
Louise Crossley
You just buy credits, but for what you pay, you get so many credits.
Daniel Rolls
Yeah, yeah, it's great. Just before we kind of draw things together just to say, we have got a generative AI masterclass coming up and an advanced prompting one which I'm really excited about doing loads and loads of research for this as well. So. So as you're probably aware, Target Internet members have access to all of our masterclasses and they get all our up one hour update sessions as well. So if you want to see what we've got coming up, targetinternet.com live and you can see all the year ahead of all the sessions that we've got coming up. And they're really nice now because there's loads of people attending, but we're getting a really nice community of people now that are coming to them really regularly. And I should really give a shout out to Brian. Brian is one of our podcast listeners. He is an Australian and he joins these things at 2 o'clock in the morning almost every single time. So I think that shows pretty serious level of commitment. So thank you, Brian. Okay, so we're going to keep testing the different AIs against each other and we'll see how this evolves. And thank you for listening to Digital Marketing Podcast and you will find all of these links in the show notes. Targetinternet.com podcast for more episodes resources to leave a review or to get in contact, go to targetinternet.com podcast.
Podcast Information:
In this episode of The Digital Marketing Podcast, hosts Ciaran Rogers, Daniel Rowles, and Louise Crossley delve into the competitive landscape of generative AI tools, specifically comparing ChatGPT, DeepSeek, and Gemini. The discussion explores the capabilities, privacy concerns, and practical applications of each AI, providing listeners with a comprehensive analysis to determine which AI offers the best value for digital marketers.
Daniel Rowles initiated the conversation by highlighting the unexpected emergence of DeepSeek, a Chinese-based generative AI tool. Launched at a significantly lower cost and with reduced processing power compared to established models like ChatGPT and Gemini, DeepSeek swiftly disrupted the AI market. This disruption led to a notable decline in stock prices of several tech companies involved in AI development.
Daniel Rowles [00:23]: "DeepSeek... caused a huge crash in the stock prices of all those kind of technology companies as well."
However, skepticism soon arose regarding DeepSeek's rapid development. Allegations surfaced that DeepSeek might have utilized ChatGPT's API in its training process, although these claims remain unproven. Daniel shared an anecdote about a professor from Imperial College who initially invested heavily in Nvidia stock based on these rumors, only to see a reversal in fortune when the truth about DeepSeek emerged.
Daniel Rowles [02:15]: "But he's a little bit put out... it's very comfortable."
Ciaran Rogers brought attention to DeepSeek's privacy policy, which raised significant red flags. Matt, a collaborator of Ciaran's, identified that DeepSeek collects a vast array of user data, including text, audio, prompts, feedback, chat history, personal details, device information, crash reports, keystrokes, cookies, and IP addresses.
Ciaran Rogers [04:54]: "They're not saying they definitely do, but they're. They collect and may use that."
This extensive data collection is reminiscent of the privacy challenges faced by platforms like TikTok, leading to concerns about user data security and potential misuse.
The hosts conducted a practical comparison by prompting each AI with the same request: to write an article titled "The Best Online Digital Marketing Courses," including a review of English-language platforms, a comparison table of features, and mentions of TargetInternet.com and Google Garage.
Gemini's Performance:
ChatGPT's Performance:
DeepSeek's Performance:
Daniel Rowles [09:14]: "ChatGPT, the paid for version was very good, but the free version of DeepSeek is as good, in my opinion as the paid version of ChatGPT at the time I tested it."
The hosts discussed the implications of these findings for digital marketers. DeepSeek's free access and comparable performance to ChatGPT's paid version make it an attractive alternative. However, concerns about privacy and data security remain paramount.
Ciaran Rogers [12:28]: "But it's, it's, it's good. They would have gone through and checked all the security stuff on this as well."
Additionally, Daniel emphasized the importance of responsible AI usage, advocating for tools like 1Password to manage unique, strong passwords to safeguard against potential security breaches.
Daniel Rowles [05:09]: "So a little bit of a Shout out to 1Password because it is a great solution."
The conversation shifted to the broader impact of AI on human creativity and critical thinking. Ciaran expressed concerns that over-reliance on AI could diminish marketers' ability to think independently and creatively.
Ciaran Rogers [12:56]: "What happens if we have a whole generation of marketers who've never actually thought stuff through or don't actually know the theory."
Conversely, Daniel highlighted that AI tools, when used effectively, can enhance creativity by providing detailed reasoning processes that users can build upon.
Daniel Rowles [13:49]: "You can take your content, put it through something else or add some manual editor to it."
The hosts touched upon the emergence of AI detection tools, such as Surfer SEO's free tool, which assesses the likelihood of content being AI-generated. While these tools are becoming essential for recruiters and content creators to ensure authenticity, Daniel pointed out the challenges in accurately detecting AI-generated content as AI models become increasingly sophisticated.
Daniel Rowles [16:44]: "If you're any good at using the tools, no, you can't, because you can take your content, put it through something else or add some manual editor to it."
Wrapping up, the hosts promoted upcoming generative AI masterclasses and encouraged listeners to engage with the Target Internet community for continuous learning and updates.
Daniel Rowles [19:38]: "Just buy credits, but for what you pay, you get so many credits."
Ciaran also acknowledged the dedication of listeners like Brian from Australia, highlighting the podcast's global reach and committed audience.
In this insightful episode, The Digital Marketing Podcast provides a thorough comparison of ChatGPT, DeepSeek, and Gemini, revealing the strengths and weaknesses of each AI tool. While DeepSeek's rapid development and cost-effectiveness position it as a formidable competitor, concerns about privacy and data security cannot be overlooked. ChatGPT continues to offer reliable performance, especially in its paid version, making it a preferred choice for many marketers. Gemini, however, appears to be lagging, struggling with reliability and accuracy.
The discussion underscores the importance of balancing AI utilization with human creativity and critical thinking, ensuring that digital marketers remain innovative and intellectually engaged. As AI technology evolves, staying informed and adapting to new tools will be crucial for maintaining a competitive edge in the digital marketing landscape.
Notable Quotes:
For more insights and episodes, visit TargetInternet.com.