I've spent well over 100hrs testing different AI models this year. For one purpose: Generating cold emails at scale that don’t all sound the same. The best AI model for cold outreach is all about who writes differently, not necessarily who writes the best. If every email sounds identical, deliverability suffers. To counter this, we force models to generate high-variance copy using this line in our prompt: “Maximise email variability by changing greetings, transitions, and phrasing. Ensure each generation is linguistically unique while preserving meaning, tone, offer, and call-to-action.” The problem: GPT & Claude try, but the outputs still converge. Gemini 2.5 Pro is the only model that truly understands and executes tthe variability requirement. If you’re using AI for outbound and your emails still have that “same template, new day” energy - it’s probably not your prompt. It’s your model.