ChatGPT vs Claude for Chrome vs Perplexity (James Dooley Discusses with Dennis Yu)

James Dooley: ChatGPT versus Claude versus Perplexity. Today I am joined with Dennis Yu, who has done a lot of split testing between the different LLMs. He works with a lot of different contractors on getting knowledge panels, improving local SEO, maps, search rankings, and doing almost everything related to digital marketing. So today, because I have seen you doing this test in depth with regard to pricing, the context window and what works best, when you are comparing ChatGPT, Claude and Perplexity, which one comes out on top and why?

Dennis Yu: Well, let me show you, James Dooley. I am going to share my screen, which I see very few people actually doing. So rather than give you an opinion, let me show you what is going on. We spoke earlier about how ChatGPT has this Atlas browser, which you can see here. Atlas is a Chromium browser, so as I am doing stuff like making Facebook posts or working, there is my calendar with meetings, we have a dashboard of different sites and so on, and I ask it to do stuff. What happens is this little window on the side opens up. Like here, we are chatting in LinkedIn, and I can open up this window on the side and tell it to do different things. All kinds of stuff. If it has access to my Google Photos and my email and whatever else, it can do all these different things, which is awesome.

My buddy is basically almost a professional athlete. Here I asked it how we make his site better because it built the site, right? It did a great job. It gathered all his videos and repurposed them. I constantly say, okay, what else? I speak to it and then decide whether I want it to do it or not. In agent mode, it does it. I have spent about $20,000 in the last six months burning credits on ChatGPT via Atlas because of agent mode. Sometimes I go into Pro mode, where when you first start a new project, you often want to go into Pro, and that basically costs about $150 in tokens. You can see the way the agent operates when you get the glistening dots. It starts to do different things, it starts to open tabs, and it starts to do work if it has the access and if you are clear about how you want it to do things. This has been very good for the last seven months or so. I was very loyal to ChatGPT, even though I was burning three or four thousand dollars a month. Then Claude came along, and I was one of those people thinking I do not want to do the Claude thing because I have got too many things going on. Now, look. Can you see how many tabs I have open?

James Dooley: Yes.

Dennis Yu: I have got maybe 70 tabs open here. This is 128 gigs. This is a fully stacked-out MacBook Pro. You can see I have got 128 gigs and it is using 108. In Chrome, it is about a gig per tab. So you see these groups, these lines, like this blue line. That is a tab group. So here I have it auditing a series of websites and then pulling the transcript, turning them into articles, writing emails to team members, paying invoices, and looking at themes. All of this stuff.

Claude is able to do the work in a slightly different way because it also starts opening tabs, but it is able to work longer. If we look at one of these, here is a book that we are writing that I have not finished. Let me find one of these other ones so you can actually see it at work. Okay, here. Just to give you context, I said, “Hey, I love the idea of building in public,” which is what we have been doing. So I said go and look at all the stuff on how we do that and write this article about how we build in public. Find any article I have written about that topic, about how we document what we do and how we do that, how we store things in different ways, and how that propagates across the team and clients and all that. I speak to it and, look, it tells me how many steps it took. It starts to document as it goes, it takes screenshots, and then it makes recommendations. Then I say, okay, how do you make that better? It says, “Well, you should do this, this and that.” Notice how it is more persistent. It takes 102 steps, and it might work for 40 or 50 minutes, and it is doing these things. Then it comes back and tells me what it did, and I will come back and say, “Awesome. That is great work. Can you go ahead and write a meta article based on our meta article guidelines where we document what we do?” Because obviously, if we are talking about how we build in public and we share exactly how we do everything, it would be hypocritical not to write an article about how we do that. Then it goes to town. If I open another tab, I can go to one of these other ones and say, “Hey, I forgot what we were saying about writing meta articles. Can you go and look at our article on how to write articles?” That has lots of examples of how we do that and how we define SOPs so that as our agents learn how to do stuff, they continue to update the document. It is recursive. They get better. They learn from themselves, and we send agents to these articles. At the same time, we also have an article on how to write a definitive article, which is how to tune an SOP and how these different SOPs are linked together based on what we do for SEO and PPC and so on. So the more experience we have, the more Zoom calls we have, the more data we see from Google Business Profile and Ahrefs and whatnot, the system just learns from itself. The agents learn from themselves. They document everything, and they document it in such a way that we can move everything to ChatGPT or to Gemini or to Grok or whatever because we have documented our knowledge outside of any particular system. So we are not locked in. Go ahead and audit the documentation. This is so meta of what we are doing, but go ahead and audit what we have done there and work out how we can make this system even better. If we need to write another article about this that does not conflict with what we have already said, then please go ahead and do that. Then this might work, James Dooley, for another hour or two. It will be very thoughtful and it will come back. Sometimes I even have a contest with myself, like how can I get the thing to work the longest? I have been able to get it to go up to about 700 steps and work for four or five hours, which you can imagine the token cost of, right? That is my favourite place to work.

James Dooley: So are you doing everything now? Have you moved everything over to Claude with regard to workplace? So no Perplexity, no ChatGPT?

Dennis Yu: We have got Perplexity here too. Perplexity has a Comet browser, and there are some pros and cons with Perplexity. One thing that Claude and ChatGPT cannot do very well is download and upload documents, but Perplexity kind of will. So here you can see when the blue dot is blinking, that means it is working, right? As I am getting it to do stuff, it is the same kind of thing where you see here where it says assistant. It is the same little thing on the side, where you are able to chat with it as you are doing work. You pop the thing open and say, “Hey, you see what I am doing here? Go ahead and do this thing.” So as I am sending emails, or I have it tracking where I have vouchers or flights that are changing or someone’s wedding or calendar meetings or all that kind of stuff, it will manage these things for me like a travel agent, like a web developer. Hey, rebuild this WordPress site, and so on. It will do all these things.

The beauty is when I do this in Perplexity, which is a model of models, I can choose which model I want. Because of business, and because these things change all the time as the models leapfrog each other, right now 4.6 is the best one for people that do SEO. So when I have maxed out my Claude, because you have a session limit of five hours, and here in this session I have used 57 per cent, and of my weekly limit I have used 24 per cent. So if I happen to reach this before the end of the week, or before the end of the session, rather than going to extra usage, which will burn you, because that is how you know how expensive it actually is, I will switch over to Perplexity and then I will continue to use 4.6 out of Perplexity. So that is how I am probably burning $50,000 a month in tokens and I am still paying $400 to $500 a month in total. I am like the guy who goes to the buffet and just eats steak and lobster. I bankrupt the buffet. That is what is going on.

James Dooley: That is a good way of putting it, to be fair. But if people, let us say a business owner or a contractor, are looking for simple measures, when they are comparing ChatGPT, Perplexity and Claude, let us say they go all in on Claude and they are not going to be using the full usage. They are going to be paying the $200 a month, but they are not going to be using the full usage.

Dennis Yu: Yes. So you could start with the $20 a month plan, which is the basic Claude Pro, and then they have a 5x plan, which is only $100 a month. What you do is start with the $20 a month plan, and if you run out of credits, move up to the $100 plan, and then move up to the 20x plan, which is their biggest individual plan. That gives you 20 times the credits as the regular Claude Pro. That is the $200 a month one. If you are able to use that up, and a lot of my friends that I have got on to it have lost sleep. They are like, “Dennis, damn you. I stayed up till 3am because what you showed me was so good and now I am hooked,” which is great. We are hooking people on the good drugs.

But yes, just start out there. Start out with things that are repetitive and well defined in your agency. Things that you do not like to do but you need to do, like sending out weekly reports or pulling data together or auditing websites. Stuff that you have done before. You can upload all your previous examples. If you have a project management system like we do via Basecamp, I have it go there. I probably have 700,000 emails in my Gmail, and then in Basecamp we have 3,000 client projects that we have had over the course of however many years, and we just let it go to town on that. There is so much work that can be done. So now you are just thinking, anytime you have an idea. This is how my life has changed, James Dooley. Anytime there is something I want to do, all I need to do is think of it and just get it done. I do not even have to worry about prioritising. I just do it.

James Dooley: Yes. That is crazy. And with regard to simplicity, let us say repurposing content or social media, how easy is it for someone to be pre-logged in to Twitter, LinkedIn and all the others and then just have it posting each day?

Dennis Yu: In fact, I will just show you. Do you want me to use ChatGPT or Claude or Perplexity?

James Dooley: Use Claude.

Dennis Yu: Claude. Okay. So Claude inside Chrome, which is where I am now. Let us just say I will do a search here. Watch this in real time. So I open up, I am just browsing Chrome, right? But then I think I want to pull in Claude and then I say, and you could also use Whisper Flow, which is great at dictation:

“James Dooley and I have made a lot of content, as you can see here, and it has been on YouTube and it is being partially repurposed, but I know that we can repurpose things multiple ways because the same content can be repurposed from my point of view and his point of view. So I want you to write multiple blog posts, you choose, on content that he and I have written together because we are helping SEOs and agencies and digital marketing people stay on top of the latest trends. Right now we are talking about the differences between Claude and ChatGPT and Perplexity and Grok and Gemini and so on. I want you to write an article that fits and meets our BlitzMetrics article guidelines. For example, embed the YouTube videos, choose proper categories and tags inside WordPress, and all that stuff that you already know you need to do. Go ahead and do that and make it better. Pull out the key quotes. If you need to create a diagram explaining a concept just to make it visually easier instead of just a bunch of text, go ahead and do that. You do not need my permission. And then I want you to write a meta article explaining what you did. How many steps did it take? How long did it take? How many tokens would you estimate that we used? What would it have cost if we had to use the API to do that versus just using our Claude 20x Max account? Then estimate how many hours a human would have had to do that, whether it was a VA from the Philippines or a staff member in London, and do that thoroughly. You are going to publish a number of these articles that are repurposing what we have already made. You are going to create a meta article about all these different pieces. Then I want you to cut out some of these snippets and I want you to post on my Facebook and my LinkedIn about what we did here. Then I want you to go into my LinkedIn and send a message to James Dooley explaining what we did in detail, and feel free to admit that you are the Claude AI that is doing it.” So now it will go to work for a couple of hours.

James Dooley: That is crazy.

Dennis Yu: So you talk to it, and a lot of people do not realise that talking to it unlocks a whole new level because of the amount of context you can provide, right?

James Dooley: Yes, for sure. Anyone who is watching this, look at the differences Dennis Yu is sharing there. There are a lot of people out there on YouTube or sharing blogs saying why one might be better than the other. Dennis Yu here shows the practical ways of being able to use Claude over Perplexity and ChatGPT, or alongside them. If you want any other videos like this, check out the links in the description. There are a dozen different videos where Dennis Yu explains all the different LLMs and what he is using them for. He dives straight into which AI platform is working best at present. He believes Claude for Chrome to be the best. Now, that could change in one to two months. So make sure you subscribe over to Dennis Yu. And Dennis Yu, it has been an absolute pleasure.

Dennis Yu: Hey, man. Guys, jump in. There are so many changes and all that, but I am telling you, it is amazing. You will not believe how much you will love this stuff.