Why I Switched from ChatGPT to Claude

I use ChatGPT everyday for 2 years, writing, researching, brainstorming, [music] coding with Codex. It was at the center of how I worked. And then I tried Claude and I canceled my OpenAI subscription the next day. I work in marketing, so I'm not switching tools for fun. I use this stuff for real work everyday. And if you're already paying for ChatGPT and wondering if it's worth the switch, I get it. Like, is it worth moving all my project instructions and chat history? Will I even notice the difference? Isn't Claude for coders and not for marketers? But there are actually four reasons I made the switch, each one bigger than the last. [music] And it starts with the one thing ChatGPT does that nobody talks about but everybody notices. No one needs a yes man. A few weeks ago, I had what I thought was a banger idea for a video. I pitched it to ChatGPT first. The response? Love it. Strong angle. Here's how to structure it. Then I pitched the exact same idea to Claude. Claude told me the hook didn't actually deliver on the title and that retention would go down because there was too much information in the intro. Once you start noticing this pattern, you can't unsee it. Ask ChatGPT a question, it [music] doesn't matter what. It could be about marketing. It could be about which bourbon to use in your cocktail. And every time, great question. Now, if you're like me and you you work in a job where real feedback is required to do your job well, you don't need an assistant that agrees with everything. You need one that will tell you when your idea isn't as good as you think it is. That's when I stopped [music] spending time with a cheerleader and started spending an unhealthy amount of time with a sparring partner. But pushback was just round one because when I tried using Claude for the thing ChatGPT had been failing me on for 2 years, I was shocked. And that's writing in my voice. When I was using ChatGPT, I was convinced that if I trained it long enough, it'd sound exactly like me. So, I uploaded old scripts, I pasted Slack messages, I probably fed it over 100,000 words of context, and it would do it for about two paragraphs, and then it'd slip up. The same phrases every single time. For marketing, that's not cosmetic, it's structural. It's not a feature, it's a philosophy. In the world of marketing, unlock this, game-changer that. And don't even get me started on the em dashes or the emojis in every sentence, even when you ask it to stop. No matter how many examples I fed it, no matter how much feedback I gave, [music] that style kept creeping back in. So, naturally, when I tried Claude, I went in with low expectations. I did the same things. I fed it with writing samples, I gave it strict guidelines. But this time, I did one thing differently. I told it not to write anything yet. Just analyze my writing style first. And what came back wasn't a generic, "Your tone is conversational and direct." It did say that, but overall, it was actually very specific. Things like [music] noticing that I avoid em dashes, that I use one-word sentences as punchlines, that I almost never use empty words like leverage, that my paragraphs get shorter when I'm building to a point. Some of it I agreed with, some of it I didn't, but I took the parts that felt right, and I dropped them straight [music] into Claude's custom instructions, and the difference was immediate. I've been using Claude almost exclusively now for a while, and I have a running style guide that gets updated regularly. So, it's getting better and better at sounding like me. The first drafts come faster, the edits come faster, too. No, it's far from [music] perfect, but it's close enough that I'm not rewriting every output like I was with ChatGPT. So, my experience Claude thinks better and it writes better. And honestly, these two alone were enough to make me switch. But the real reason I canceled ChatGPT, it has nothing to do with Claude's chatting abilities at all. It's the fact that Claude turned me into a capable developer. Years ago, I started an e-commerce website. This was before Shopify existed. [music] I was using a platform called Magento. And if you've never heard of Magento, just know it wasn't built for people who couldn't code. Any kind of customization was basically impossible unless you were a developer, which I was not. So, once I'd made a little bit of money, I reached out to a web development agency for a quote on a proper site. He came back at $25,000. I had like 250 bucks in the bank. So, I did what any broke but driven 20-something year old would do. I taught myself. I spent hours, days, weeks sitting in Starbucks reading web development books trying to teach myself how to code. And after many months, I got good enough to make small adjustments on my own. But nothing I was ever proud of. Fast forward to today, anyone can build an app just by describing what they want in natural language. But in my experience, the gap between Claude coding, Codex, the coding tool that comes with ChatGPT, was massive. More often than not, I'd ask Codex to [music] do something and it would just misinterpret what I was asking and then spend 10 to 15 minutes writing code. Then I'd open that file, and the app just looked bad and it didn't work at all, at least the way I wanted it to. That didn't happen so much with Claude code. And I think the reason why is because when When something complex, there are dozens of decisions to make. What should the architecture look like? How should the data flow? What happens for certain edge cases? [music] Codex never asked me anything. It would guess and then build. Claude Code does the opposite. Before it writes a single line of code, it has something called planning mode. It maps out the entire project with you. It asks questions. It surfaces the ambiguity instead of burying it. It's the difference between a contractor who sits down with you and goes through the blueprints versus one who just starts building based on their own assumptions. And the things I and other people within my company have built with Claude Code, stuff we just genuinely never thought would be possible. Like this. This is a full outreach and prospecting [music] tool that I built in a day. It follows the exact workflow I'd use myself down to the quirky little details like finding emails based on a priority list of job titles. George, he's created some incredible video editing tools that have improved efficiency by a ton. And all of this was as easy as sending a few messages. The point is, I'm prototyping and shipping things in hours [music] that used to take weeks and cost thousands of dollars, all thanks to Claude. I'll tell you this. The kid sitting in Starbucks trying to learn PHP would lose his mind. But here's the thing. Building tools is just one level. The next level is when those tools stop needing you to run them because Claude made it super simple to create AI agent files. Going back to that outreach tool I just showed you. It was great, but every time I ran a campaign, I was still doing the same steps. Finding prospects, qualifying them, writing the messages, following up. But everyone was talking about AI agents. So I thought, why not try to fully automate cold outreach in a non-spammy of course. So, I tried it with ChatGPT first. It was a disaster. It didn't understand the actual [music] workflow. It skipped steps. It did things in the wrong order. To the point where I thought fully automated outreach just wasn't [music] possible. So, I scrapped it, and I rebuilt the whole thing in Claude. Now, I like to think that I'm pretty good at operations, building systems that are repeatable and scalable. That's kind of the thing I obsess over. But, the annoying part has always been the documentation, which it turns out is basically what AI agents need to work well. I quickly discovered that Claude is incredible at creating documentation. So, here's what I did. I pitched my workflow to Claude and walked through how I do outreach. I talked through the edge cases, and Claude turned it into structured workflows in the form of markdown files that I could plug and play with an AI agent tool. After some back and forth, Claude and I built a fully automated outreach system. Something I genuinely thought was impossible. In fact, I documented exactly how I did this in a video, which I'll link up in the description. Now, here's where everything I've talked about in this video comes together. Agents make decisions on your behalf. They're handling complex scenarios and branching paths. If the AI you're using is a yes-man that just guesses at everything and never asks clarifying questions, it builds agents that make bad decisions. The pushback, the clarifying questions, the "Are you sure?" before it acts. Those aren't just nice when you're chatting. They're essential when you're building systems that think and act on their own. The model is the brain, and when you're building something that makes decisions for you, the brain matters more than anything else. So, is Claude perfect? No, far from it. And I need to be real with you here, because I don't want this to sound like a sponsored video, because it's not. First of all, it's more expensive. Now, you might be thinking, "Well, Sam, they're both 20 bucks, so how is that true?" What you'll find when you use them side by side is that it's much easier to burn through your usage in Claude than ChatGPT. I'm on the $20 plan, and I frequently run out of my hourly usage to the point that I'm seriously considering upgrading to the $100 max plan for my own personal usage, even though I already have an enterprise plan through work. Another thing is the downtime. I found Claude to have noticeably more downtime than ChatGPT. There are times when I'm in the middle of work, and it just dies on me. It won't send messages, and the only thing I can do then is wait. And third, ChatGPT has a way better voice mode. I used it constantly when driving. [music] I'd learn new skills, plan video outlines, structure ideas. It was genuinely fun. The last time I used Claude's voice mode was pretty bad. So, my time in the car has gotten a lot less productive. But here's the thing. No tool is going to be better than every other tool in every other way. OpenAI will get better. Gemini will get better. Anthropic will get better. It'll be a tug-of-war between a handful of companies for the next several years, which is exactly why we did something different at Ahrefs. Marketers are using ChatGPT or Claude for marketing work, but we all hit the same walls. These models are great at thinking and writing, but they don't have access to the data marketers actually need in order to [music] make good decisions. Things like keyword search volume, backlink profiles, keyword and traffic data, all the stuff that lives inside marketing platforms like Ahrefs. So, we asked ourselves, "What if you could combine the best AI models with unrestricted access [music] to all of Ahrefs data?" What would that unlock? It was then that we realized it would help people to go from chatting about marketing to actually having AI do it for them. So, we built Agent A, which lives inside our own AI chat tool. [music] And it's not just a chatbot that can generate reports with Ahrefs data. Yeah, it can do that really, really well, too. But, it can [music] actually build tools, run workflows, and execute real marketing tasks. So, you can focus on higher-level strategy and execute [music] at 10 times the speed that you're doing now. For example, I built a tool that automates keyword research for YouTube. [music] You just drop in your competitor's YouTube channel URL, and it pulls every video from that channel. Then it runs each one through Ahrefs to get the Google traffic estimations and tells you which videos are quietly killing it through Google SEO. I built a content decay detector for an affiliate site operator who's absolutely killing it. It monitors his pages, watches for ranking drops, and it can even be configured to send a Slack message when a post needs updating. I mean, if he really wanted to, he could have Agent A update the content and push it to his WordPress site. If you can describe a marketing workflow, Agent A can plan it, it can build it, and it can run it for you. Sure, the model is the brain, but the brain still needs the right fresh data to actually go beyond consultant-type work and to be an executor. Now, getting back to the ChatGPT versus Claude decision, for me, that's Claude every time.