I used to look for one “best” LLM. But after a lot of trial and error, I stopped comparing them and started treating them more like tools for different tasks.
So my current setup looks roughly like this:
Claude — brainstorming and strategy. I use it when an idea is still vague and I need someone to challenge it, ask questions, and help me find a direction. It’s especially useful for content ideas, structuring information from different sources, and planning community contests.
ChatGPT — writing, editing, and technical checks. I prefer ChatGPT’s way of thinking when it comes to actually writing and refining content. It’s my second pair of eyes for copy, plus my helper for CSS tweaks and JavaScript checks - small things that make me feel like a developer
In short: Claude for ideas, ChatGPT for polishing and technical help.
To be honest, sometimes I still wonder if I’m overcomplicating things by using two tools instead of just one But so far, it’s been the most efficient approach for me.
So, is anyone else juggling a couple of LLMs, or have you found one that checks all your boxes?
For quite a while I’ve been totally fine with ChatGPT and pretty happy with the results it gave me. It was only when I decided to test out Claude that I started comparing, second-guessing, and going back and forth
Right now I’m digging into Claude’s features (perhaps partly because of those stories about its rival messing with files ), but I cannot say for sure it’s 100% better than other LLMs.
Perhaps it really depends on what you’re trying to achieve. From my experience so far, Claude is especially good for big-picture thinking and strategic feedback.
Thank you for the valuable information. I particularly liked the examples you included, as they helped clarify the concepts. It would also be interesting to see a follow-up article covering advanced techniques.
Good afternoon. If you set up a specific Customer GPT, as we have for our company, it delivers great results for content as well as image creation. We have never used Claude, as we are happy with ChatGPT Results.
I like your perspective. Maybe the question isn’t always “which LLM is the best?”, but rather “how do we build a workflow around the tools we already have?”
Your comment made me curious — what kind of content are you creating with your Custom GPT? And what else do you find yourself using it for?
We built our own Corporate Ambassador AI Agent GPT and trained it with our Corporate Information, excluding sensitive internal financial data. The link is available publicly. With the correct setup, the GPT will use the trained data for answers together with Web Search results. It is similar to Gems, but with a huge difference (which I must mention) in the output when using the right prompts. This applies to creating blog posts, images, and specific visuals. The results with ChatGPT are incredible. Well, at least for us. We are on the $20.00 monthly paid version.
Interesting… I do it the exact opposite. CGPT 5.6-sol for content, build out and research - building a master framework knowledgebase file as we go, being structure and organization, then take it into Claude for all the code/html tweaks, fixes, logic correction, UI design/structure and final high-end premium polish using Opus 4.8 High/Extra. I also use Grok for specific research and medical as it gives you much better results than all the others. I find it much better for those tasks. I also use grok Imagine for things as well. So I do agree, all have their niche use cases.
I like the idea of having a dedicated GPT trained on your company information. It makes a lot of sense for content creation, especially when consistency and company context are important.
And I agree about prompts — sometimes the way you structure the request makes a huge difference in the final result.
That’s a really good example of how the same tools can end up being used in completely different ways depending on the workflow.
I was especially interested in the master framework knowledgebase file you mentioned. Do you keep it as a single evolving file, or do you build separate ones for different projects?
I was using projects, the main being a complete page by page website audit, re-write and optimisation for SEO/GEO/SERP, etc, adding new rules, guidelines, training info along the way using “add to knowledge base”, but after a while project would get very bloated and so slow I would have to start a new thread.
While it could retain its internal knowledge base for the most part, I found myself having to “remind” it of existing rules/guidelines. Eventually I asked it for a better way - a master framework was the solution.
I did have a few specific projects, but I recently had it build out a complete, merged master of all relevant data. I had it formatted and organized in the best way for it to use, not me, as it was looking more like a reference handout for me… and I never even look at it.
This new master is the complete knowledge base resource that I can now pull specific info out of for other projects and related tasks.
It is a actual doc that is continually added to, updated and revised… so before I leave a related chat thread, I just have it add everything relevant or updated to the master framework, and it drops it in my DropBox.
So now I dont even need to be in a project, as it retains most of that info internally anyway. I just upload the latest master or spin-off version in a new chat and off we go!
@HavasuLew Which LLM are you using with these projects? I wasn’t aware that they could get bloated. I use them in Claude and ChatGPT specifically to wall off one client project from another so the LLM doesn’t confuse “Greg’s” messaging with “Anne’s” from another client. That’s working, and the projects aren’t heavily loaded down yet, but they will get more loaded over time.
@Max - I’m so surprised to hear this. Personally, I find ChatGPT overly wordy and incredibly annoying lately. Claude is a much better writer, and better at mimicking voice. I use them specifically for ghost writing for multiple clients, so capturing voice is crucial. Claude is significantly better at this.
I’m not doing coding or building agents yet. But I tend to turn to ChatGPT for everyday tasks like analyzing my website for holes, telling me how to use a French washing machine when I was on vacation, and figuring out how to update specific settings in CloudFlare (although their chatbot is pretty good…should have started with that).
Love seeing how you set up the master framework, @HavasuLew!
Keeping everything centralized like that makes a lot of sense and clearly saves a ton of effort. A huge thank you for sharing your insights, we really appreciate it
@Max I follow a very specific, repeatable process. I’m usually trying to write 12 LinkedIn posts per session. I have to correct several things for the first 2, but then it gets better. I still do a bit of editing on each one. But it also depends on the day. It’s not always consistent. My bigger problem with Claude is running out of usage before completing the set of 12. I’ve been working through that issue and figuring out different ways to manage the context window. I never run into that problem with ChatGPT
I like the idea of having a repeatable process for each session. That’s probably where the real value comes from — not just the model itself, but having a workflow you can rely on.
The usage limit issue would definitely throw a wrench into that though
Thanks a lot for sharing your insights, much appreciated @Candyce_Edelen!
I have a bible based website that answers questions. ChatGPT stated a larger reasoning model (such as GPT-5.5 or a full GPT-5.4 reasoning model) would be best for my website. I love Elfsight for my website, I believe it does a pretty good job.
Also, I wonder if Elfsight will eventually have a built-in YouTube or Website Analyzer of web links.
But so far, it’s been the most efficient approach for me.
