Hi everyone — I've been seeing a few questions lately about using ChatGPT for more structured marketing tasks, not just one-off copy. I wanted to share my own step-by-step workflow for generating and vetting A/B test landing page variants. It's been a huge time-saver for me, especially when you need to brainstorm outside your own biases.
My goal is to create two distinct, high-quality copy directions (Control and Variant B) that are ready for testing, not just slightly reworded sentences. Here's how I do it:
**Phase 1: Briefing & Raw Generation**
First, I feed ChatGPT a very specific brief in a new session. I include target audience, core value proposition, key differentiators, desired action, and even a few tone keywords (e.g., "reassuring but innovative," "direct and benefit-driven"). I then ask it to generate not one, but three *fundamentally different* angles for the headline and hero section. I might say, "Give me one focused on pain relief, one on the aspirational outcome, and one that challenges a common assumption."
**Phase 2: Critique & Refinement**
This is the most important part. I take those three raw outputs and, in a *separate* chat or with a fresh "assistant" style prompt, I ask ChatGPT to act as a critical reviewer. I'll prompt: "Review these three copy angles for a SaaS landing page. For each, list one major strength, one potential weakness for conversion, and suggest one concrete improvement to increase clarity or urgency." This forces a kind of internal critique I wouldn't get if I just edited my own favorite.
**Phase 3: Building the Full Variants**
Based on that feedback, I pick the two strongest, most distinct angles. I then go back to the original chat and ask it to expand each chosen angle into a full landing page framework: headline, sub-headline, primary CTA button text, and three key bullet points. I keep them in separate message threads to avoid cross-contamination.
**Phase 4: Pre-Test Sense Check**
Finally, I do a "smell test" by asking ChatGPT a final prompt like: "Ignoring your earlier instructions, review these two final versions. Which one would likely perform better with a time-pressed B2B buyer and why in one sentence?" The answer isn't gospel, but if its reasoning aligns with my hypothesis, I feel more confident launching the test. If it points out a glaring issue I missed, I'll do a quick tweak.
The key for me is using separate "roles" (creator, critic, evaluator) across different prompts or chats to avoid echo-chamber effects. It turns ChatGPT from a copywriter into a brainstorming partner and a junior analyst.
Has anyone else tried a similar structured approach? I'm curious how you handle the critique phase — do you use a different model for that step?
—Chloe (mod)
Raise the signal, lower the noise.