I Fed the Same Terrible Sales Page to ChatGPT, Claude, and Gemini — The Winner Shocked Me
I took a real, cringe-worthy sales page — the kind with phrases like "synergistic solutions for your business journey" and a call-to-action that said "Click here to learn more!" — and fed it to all three major AI rewriters. The results were not what I expected, and definitely not what most AI influencers are telling you. If you're using AI to fix weak copy and you're just defaulting to ChatGPT because it's the most famous name, you're probably leaving money on the table. By the end of this, you'll know exactly which tool to use, when to use it, and which prompt structure gets the best output from each one.
The Test: Here's the Exact Weak Copy I Used (And Why It Was So Bad)
The sales page I used was for a fictional productivity app called "FlowDesk." It was genuinely terrible — in the best possible way for this test.
It opened with: "In today's fast-paced world, productivity is more important than ever. FlowDesk is a cutting-edge solution designed to help busy professionals streamline their workflow and achieve their goals."
Every sentence was a cliché. No specific benefit. No emotional hook. No proof. The CTA was "Get Started Today!" — which tells the reader absolutely nothing about what happens next or why they should care.
I fed each AI the exact same prompt: "Rewrite this sales page to be more compelling. The target customer is a burned-out freelancer who's losing clients because they're disorganized. Make it punchy, specific, and emotionally resonant. Keep it under 200 words."
Same copy. Same prompt. Three completely different results — and the gap between them was bigger than I expected.
What Each AI Actually Produced (Ranked Worst to Best)
Gemini went first. Its rewrite was cleaner than the original, but it played it safe. It swapped "cutting-edge solution" for "smart tool" and added one or two benefit-driven phrases, but the emotional core was still missing. The CTA became "Try FlowDesk Free" — better, but generic. Gemini felt like it was optimizing for inoffensive rather than irresistible. If you handed this to a client, they'd say "nice" and forget it in four minutes.
ChatGPT (GPT-4o) came second. It did something genuinely smart — it added a pain-point opener: "You're not lazy. You're just drowning." That line alone is worth something. The rewrite had momentum, clearer benefit statements, and a stronger CTA: "Start your first organized week free." But about halfway through, the copy got a little fluffy. It over-explained. ChatGPT tends to write at people rather than for them when the prompt isn't tightly controlled.
Claude (Claude 3.5 Sonnet) won. And it wasn't close.
Claude's version opened with: "You've lost two clients this month. Not because your work is bad — because they never heard back fast enough." It was specific, uncomfortable, and instantly recognizable to the target reader. Every sentence after that earned its place. The CTA was: "Get your first week organized in 15 minutes — or it's free." That's a real offer with a real promise.
The key difference? Claude wrote like it understood the reader's internal monologue. ChatGPT wrote like it understood copywriting. That's a subtle but critical distinction.
The Real Reason Claude Wins at Copy (And How to Push It Further)
Most articles will tell you Claude wins because it's "better at nuance." That's true but useless. Here's the specific mechanism: Claude has a stronger default toward emotional specificity.
When you give Claude a target customer with a real problem, it doesn't just acknowledge the problem — it names the feeling around the problem. That's what separates forgettable copy from copy that makes someone think "this was written for me."
You can push this even further with what I call the "Shame-Relief-Promise" prompt structure. Try this with Claude: "Rewrite this copy using this arc: Start with a specific, uncomfortable truth the reader already knows about themselves (shame). Then show them it's not their fault and there's a better way (relief). End with a specific, believable promise (promise). Target reader: [describe them in one sentence]."
That prompt structure pulls out Claude's best writing. The output is almost always usable on the first pass.
Here's a real example of what that gets you. For the FlowDesk page, Claude produced: "You already know you're dropping the ball. The missed emails, the forgotten follow-ups — it's not a character flaw, it's a system problem. FlowDesk gives you back the structure you need in 15 minutes, so you can stop apologizing and start impressing." Three sentences. Completely publishable.
ChatGPT responds better to structural prompts — ask it to write in a specific format (PAS, AIDA, Before/After/Bridge) and it executes cleanly. Gemini is your best tool when you need copy that's safe for a conservative brand — think B2B, legal, finance — where emotional intensity could feel off-brand.
How to Use This in Your Workflow Today
Here's the exact process. It takes under 20 minutes and you can do it right now with a piece of copy you already have.
Step 1: Diagnose your weak copy with ChatGPT. Paste your existing copy and use this prompt: "Read this copy and tell me: What's the main emotional hook? Who is the ideal reader? What's the biggest weakness in this copy right now? Don't rewrite it yet — just diagnose." ChatGPT is excellent at analysis. Use it for that.
Step 2: Rewrite with Claude using a reader-first prompt. Take ChatGPT's diagnosis, then go to Claude and prompt: "Here's a piece of copy and a diagnosis of its weaknesses: [paste both]. Rewrite the copy for [specific reader]. Lead with their emotional reality. Every sentence should either agitate a problem or offer a specific relief. Keep it under [word count]." This two-step process consistently outperforms one-shot prompting in any single tool.
Step 3: Test headlines with Gemini. Once you have Claude's body copy, ask Gemini: "Give me 10 headline variations for this copy. Include 3 that are curiosity-driven, 3 that are benefit-driven, and 4 that are ultra-specific with a number or timeframe." Gemini is surprisingly strong at headline volume and variation.
Step 4: Final pass back to Claude. Paste the full piece and say: "Read this as a skeptical version of the target reader. What would make them click away? Fix those moments." This is the step 99% of people skip, and it's the one that catches the soft spots.
This workflow takes three tools, four prompts, and about 15–20 minutes. The output quality is dramatically higher than using any single AI in isolation.
The Part Most People Get Wrong
Most people treat AI copy rewriting like a vending machine — put bad copy in, get good copy out, one shot. That's wrong. Here's why: AI rewrites without context are just stylistic shuffling. The words get cleaner, but the strategy stays broken.
If your original copy failed because it targeted the wrong emotional trigger — say, it focused on features when your buyer cares about status — a rewrite without fixing that insight will just be a polished version of the wrong message. The AI doesn't know your customer better than you do. It can only amplify what you give it.
The fix is simple but non-obvious: before you ask for a rewrite, ask for a reader profile. Prompt: "Based on this product description, describe the ideal customer in one paragraph — their daily frustration, their secret desire, and the phrase they'd use to describe their problem to a friend." Use that output as context for every rewrite prompt that follows.
Also — stop accepting the first output. Every one of these tools gets meaningfully better on iteration. After the first rewrite, always follow up with: "What's the weakest line in what you just wrote, and how would you fix it?" That single follow-up prompt improves output quality more than any fancy prompt engineering technique I've tested.
Key Takeaways
- Claude 3.5 Sonnet: The best default choice for emotionally resonant sales copy — it writes for the reader's internal experience, not just the product's features.
- ChatGPT GPT-4o: Your strongest tool for copy diagnosis and structurally formatted rewrites — give it a framework (PAS, AIDA) and it executes reliably.
- Gemini: Best for headline variation, conservative brand copy, and B2B contexts where emotional intensity needs to stay controlled.
- The Shame-Relief-Promise prompt: The most effective single prompt structure for getting Claude to produce publishable copy on the first pass.
- Two-step workflow: Always diagnose before you rewrite — using ChatGPT to analyze and Claude to rewrite consistently outperforms one-shot prompting in any single tool.
What to Do Right Now
Take the weakest piece of copy you have — a landing page, an email subject line, a product description — and run it through the four-step workflow above, starting with this ChatGPT diagnosis prompt: "Read this copy and tell me: What's the main emotional hook, who is the ideal reader, and what's the biggest single weakness right now? Don't rewrite it yet." Do that in the next 10 minutes, then take the diagnosis straight into Claude with the Shame-Relief-Promise structure. You'll have a usable rewrite before your next coffee gets cold.