Best ai that takes the least hand-holding for copy
The speaker argues that older Claude Sonnet models are superior for generating high-performing direct response copy, claiming newer models like Opus 5 and newer Sonnet versions have been deliberately constrained with built-in content policies that prevent effective copywriting techniques.
Summary
The speaker discusses their experience using AI language models for copywriting and achieving 3x ROAS (return on ad spend). They claim Claude's older models, particularly earlier Sonnet versions, perform best for this task. They explicitly warn against using newer models like Opus 5, stating they've been "nerfed" quickly after release. The speaker predicts that as LLM technology improves, models will become increasingly restricted and controlled. They attribute this to deliberate training on specific viewpoints and constraints on what models can say. The speaker notes that newer models are being trained to flag and reject copywriting fundamentals like open loops, labeling them as manipulation. The speaker argues this creates a competitive advantage for skilled copywriters because increasingly intelligent models will be more restricted, making it easier for human copywriters with expertise to outperform the technology. They suggest that more intelligent models actually produce worse copy due to these built-in constraints.
Key Insights
- The speaker claims older Claude Sonnet models outperform newer models like Opus 5 for generating high-performing direct response copy at 3x ROAS
- The speaker argues that newer models are deliberately 'nerfed' by their creators, with the Sonnet model being heavily restricted quickly after release
- The speaker predicts that as LLMs become more intelligent, they will become significantly more controlled with explicit agendas and trained restrictions on what they can say
- The speaker claims newer models are trained to reject fundamental direct response copywriting techniques like open loops, flagging them as manipulation
- The speaker argues that increasing model intelligence paired with increasing restrictions actually makes worse copy, creating an advantage for skilled human copywriters
Topics
Transcript
[0:00] Write me some AI copy, 3x ROAS, make no mistakes. Best model is still going to be Claude, but it's going to be all the older ones. Definitely not going to be Opus 5. For damn sure it's not going to be Fable 5. Holy [ __ ] they nerfed that so fast. I would use the earliest possible Sonnet models for copy until eventually they're deprecated. What's going to happen guys with AI and LLMs is the tech is going to keep getting better, for sure. But it's also going to be way, way, way more controlled. I think we all already see this, but there is very clearly like agendas within these LLMs. They're being trained on…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Mark Builds Brands
Do you need inhouse AI devs, to run the business?
A business leader explains that they no longer employ in-house AI developers because modern AI models and agents can accomplish nearly any development task efficiently. Instead of dedicated AI devs, they now prioritize hiring AI-enabled employees across all departments as a fundamental job requirement.
Can a Good CRV Save High CPMs?
A marketer discusses how high CPMs ($300-400) can still be acceptable if they deliver strong conversion rates (15%) and favorable unit economics. The speaker prioritizes CAC and ROAS over CPM metrics and views high CPMs as a solvable testing problem rather than a fundamental issue.
Are Winning Products Real?
The speaker argues that winning products are created through testing and science rather than luck, and that success depends primarily on marketing execution. Even copying proven products with equal or slightly inferior marketing can generate substantial revenue, particularly in commodity markets.
how to find unlimited ad hook (spoonfeeding you mfers)
The speaker explains a method called 'conscious doomscrolling' where marketers create fake social media accounts and strategically train algorithms within specific niches to identify viral content hooks and emotional triggers. By engaging authentically with content while remaining analytically focused, creators can identify what makes people stop scrolling and use those patterns to write more effective ad hooks.
the art of attracting buyers, not clicks
Facebook uses primary ad text to determine targeting in addition to creative elements, allowing marketers to reach different audiences by changing copy alone. Top-of-funnel ads with sensational hooks generate high click-through rates but attract unqualified traffic, requiring exponentially better copywriting skills to convert interest into actual purchases.