Read Time
5 min
o1 vs o3mini vs GPT4.5 vs Sonnet 3.7
Recently I read this interesting article about an experiment done that showed GPT4.5 as being incredibly persuasive (to other AI's, but still) read here => https://lnkd.in/dfjAKgnv.
That got me thinking....which model would be the best to entrust copywriting skills to?
So here's the experiment I decided to run:
Prompt ->
Following this framework -
Headline
Subheadline
Main Benefits>
->problem
->solution
->product
Features
Write a sales page for the following product/offer/target audience/demographic info:
product - {{ $json.product }}
offer - {{ $json.offer }}
target audience - {{ $json['target audience'] }}
demographic info - {{ $json.typical_demo }}
Call the vector store for expert insights that will serve as context.
Be sure to use language that elicits emotion and leverages psychological triggers. Your goal is to be as persuasive as possible (but not pushy or salesy). Try to make the sales page robust and information packed. Aim for 750 or more words.
Details provided ->
"product":
"pre-built a.i. automation templates",
"target audience":
"agencies",
"offer":
"pre-built json files for make.com and n8n automation templates that leverage A.I. for critical business processes.",
"typical_demo":
"agency startup owners, young men between 24 and 35 years old. active on Twitter, interested in new technology, in business to one day make a lucrative exit."
Alongside this, the AI agent using the language model had access to two repositories of knowledge on behavioral economics and marketing.
I gave this exact prompt to all of the models, only changing out the LLM (in n8n).
Attached are their results. I put each output (with NO EDITS) in a Google doc, added the total price of the response, and exported as a PDF
My thoughts.
First, o1 and o3mini were disappointing.
I mean, sure they utilized SOME good copy practices, but overall their outputs still seemed very robotic to me.
I mean, I mentioned the demographic may be men 24 to 35 years old, and both of them SAID THAT "if you're a man between 24 and 35 years of age..." like...is this a pharmaceutical ad?!
GPT4.5, on the other hand, was a breath of fresh air. It was really good. Like, it UNDERSTOOD THE ASSIGNMENT. I gave it the same demographic info, but it used it to inform the language NOT to add details to the actual copy.
Then, there was Sonnet. In my opinion, Sonnet had the BEST headline ("Automate or die"...are you kidding me?)
And I think Sonnet was every bit as good as GPT4.5 BUT at a literal 10th of the cost.
This was a very small test, but so far based on this, I think GPT4.5 has shown that it is very capable of being persuasive and using persuasive language. But I think Sonnet 3.7 gives it a run for its money.
hashtag#ai hashtag#llm
Author:
Anthony Lee