My Experimentation with Claude and Gemini (and Why I’m Not Leaving ChatGPT)

As are most in the creative field, I was very leery when it came to AI advancements.

I will be the first one to admit that artificial intelligence has really enhanced our efficiency and ability to take on more projects. I will also state that AI has created better workflows and output. Some of our client blogs are far better researched, and we are able to turn these out in a fraction of the time. That’s what you call a win-win.

But since I’ve been connected at the hip with AI, I also see the horribly bad output that sometimes happens. Yes, sometimes it’s my lazy prompts. But there are other times that AI will just take a hard right, even though we’ve been working in the same way for months. Why does this happen? I’m sure the developers can give us some technical reasons, but we just mark it up to ChatGPT being ChatGPT.

ChatGPT was the tool we started with and have been working with primarily as a team. We’ve figured out how to make this a very effective tool, creating Custom GPTs, Projects, and building years of history into our conversations.

But we also pay for Google Workspace. We get Gemini with this subscription. It seems a bit foolish to pay for ChatGPT Pro when we could just use Gemini. So over the last few months, I started to pay a little more attention to Gemini.

From the start, I was slightly annoyed. Every single chat I had was showing up in my sidebar, and there was no real way to remove these chats.

See, I’m a pretty organized person. I have all of my ChatGPT conversations in Projects. If they aren’t in a project, I usually delete them. Having a running list of conversations in the sidebar is distracting, and I just couldn’t commit.

At the moment, Gemini doesn’t have an equivalent of a project. They have Notebooks, but it’s not the same. So that was a strike against Gemini.

In good news, they do have Gems. This was a must-have for me. I use Custom GPTs all the time. I did some general testing with the Gems, and that seemed to work just fine for me.

As I was watching and listening to experts, they were really hyping Claude. All of the YouTube videos were declaring winners, and Claude kept winning when it came to writing.

About 80% of our use is writing and researching. We are rarely just telling AI to “write a thing” for a client. Our process is pretty extensive, built around real writing, real voice, and refinement. It never comes out perfect, but 80% is good for us, and we take it the rest of the way.

But there’s still a lot of massaging with ChatGPT, especially lately. You can ask a million times to remove the em dash, and it still shows up. Annoying.

So I decided to test Claude on a real project instead of theory.

Soundstaged – The Podcast

I’ve been doing Disney and Universal podcasts for several years and working closely with Jim Hill Media. We landed on a concept built around the intersection of wrestling and Disney in the 90s.

I had already written the first two episodes and mapped out an eight-episode series before bringing Claude into the process.

I needed to expand a 10-minute script into a 20-minute episode, not with fluff, but with real story.

Claude delivered. It brought in relevant information, helped expand the narrative, and stayed pretty accurate. Sitting at my son’s baseball practice, I worked back and forth for over an hour and came away with a strong draft that actually sounded like me.

Over the next few practices, I refined it line by line. The result was solid.

The only issue - tokens. Claude has limits, and when you hit them, you’re done for a few hours unless you upgrade. That forces you to be more intentional with how you use it.

Claude Code - Renaming Files

Beyond writing, I started testing Claude Code.

For years, I’ve spent hours organizing and renaming photo files. Generic names don’t work when you need to find something quickly, especially when producing content.

With Claude Code, I built a naming system and tested it over a weekend. Running small batches, reviewing output, adjusting, and repeating.

The results were strong. Not perfect, but about 80% of what I would do manually, and significantly faster.

The long-term goal is to automate this and eliminate hours of repetitive work.

Where I’ve Landed

I’m not replacing ChatGPT. I’m expanding the stack.

Gemini has its place, especially for quick tasks like email, but it needs better organization.

Claude is powerful, especially for writing and automation, but comes with usage limits.

ChatGPT is still the most reliable all-around tool.

I still don’t love AI being everywhere, but I do appreciate the practical ways it can improve workflows without replacing the human element.

More to come.