fav model for conversion copy?
- Thread starter YCR
- Start date
- Joined
- Oct 5, 2026
- Messages
- 1
- Likes
- 0
- Degree
- 0
In my case, when using various AI models, I combine them based on metrics such as anti-detection, "wateriness" (verbosity/filler content), and uniqueness: ChatGpt and Grok.
- Joined
- May 27, 2026
- Messages
- 32
- Likes
- 4
- Degree
- 0
I prefer OpenAI for landing page content and ad copy. It does a good job of creating clear headlines, persuasive copy, and different variations for testing. However, I always edit the output to match the target audience and brand voice.
- Joined
- Oct 7, 2026
- Messages
- 3
- Likes
- 1
- Degree
- 0
New here as well. I've tried pretty much all of them. Honestly, for short copy, the model barely matters anymore. The prompt does. Weak input gets you fluff every time. What works best for me: write an ugly rough draft with the actual offer and real numbers, then let the tool polish it. Starting from a blank page is how you get robotic junk.
eliquid
Digital Strategist
- Joined
- Nov 26, 2014
- Messages
- 1,237
- Likes
- 2,553
- Degree
- 4
Im gonna stick with Claude on copy.
I have a pretty well dialed in and detail skills, with excellent prompts, and very good databases of examples and teachings. Also, I have a very keen knowledge of copywriting ( over 26 years with copy and ads ) and I went through yesterday on a few client accounts with:
1. Claude
2. Codex/ChatGPT
3. Grok
4. Gemini
5. DeepSeek
6. GLM
7. Qwen
8. Kimi
9. few others on open router
And I selected different models in the same family ( opus, fable, luna, sol, etc ) and Claude always beats it hands down.
One thought though, and this applies to more domains than copywriting ( code, image generation, etc ) - if you know ZERO about copy/code/images/advice you want... sure anything you get from these models "looks good". You are starting from nothing and know nothing and anything will dazzle you and impress you and get you from A to B.
But when you actually know how to code and have been doing it for years, you see the issues with models that produce bad code. If you dont know what bad code is, then any code looks great from these models.
Same with copywriting.
Same with images.
Same with financial advice, etc.
The models for sure matter when you know what you need. No level of skill, prompt, database, or MCP will help ( right now at least ) if the model/llm is bad for it.
My skills, prompts and databases are excellent. Every model produces something different using them. Claude has historically ( and as of yesterday ) been the best.
Claude is the top spot right now for copywriting, bar none.
I've went through all the copywriting skills and prompts and MCPs on the market. I do this monthly as new ones are added. I have every one just about published. Most of them are garbage. Add this garbage to a weak model like ChatGPT, and you end up with more garbage.
Then add in something like my own skills and prompts and DBs and MCPs, on a weak model. Still garbage, but slightly better garbage.
This is why you always need to be testing and evaluating. But you can't eval good copywriting ( or code ) unless you know what good is.
I've also built my own model/llm based on copywriting too using Unsloth to train it. I would still prefer to use Claude than my own model.
I have a pretty well dialed in and detail skills, with excellent prompts, and very good databases of examples and teachings. Also, I have a very keen knowledge of copywriting ( over 26 years with copy and ads ) and I went through yesterday on a few client accounts with:
1. Claude
2. Codex/ChatGPT
3. Grok
4. Gemini
5. DeepSeek
6. GLM
7. Qwen
8. Kimi
9. few others on open router
And I selected different models in the same family ( opus, fable, luna, sol, etc ) and Claude always beats it hands down.
One thought though, and this applies to more domains than copywriting ( code, image generation, etc ) - if you know ZERO about copy/code/images/advice you want... sure anything you get from these models "looks good". You are starting from nothing and know nothing and anything will dazzle you and impress you and get you from A to B.
But when you actually know how to code and have been doing it for years, you see the issues with models that produce bad code. If you dont know what bad code is, then any code looks great from these models.
Same with copywriting.
Same with images.
Same with financial advice, etc.
The models for sure matter when you know what you need. No level of skill, prompt, database, or MCP will help ( right now at least ) if the model/llm is bad for it.
My skills, prompts and databases are excellent. Every model produces something different using them. Claude has historically ( and as of yesterday ) been the best.
Claude is the top spot right now for copywriting, bar none.
I've went through all the copywriting skills and prompts and MCPs on the market. I do this monthly as new ones are added. I have every one just about published. Most of them are garbage. Add this garbage to a weak model like ChatGPT, and you end up with more garbage.
Then add in something like my own skills and prompts and DBs and MCPs, on a weak model. Still garbage, but slightly better garbage.
This is why you always need to be testing and evaluating. But you can't eval good copywriting ( or code ) unless you know what good is.
I've also built my own model/llm based on copywriting too using Unsloth to train it. I would still prefer to use Claude than my own model.
Last edited: