I Fed the Same Cringe-Worthy Cold Email to ChatGPT, Claude, and Gemini — The Results Shocked Every Marketer I Showed Them
Most AI comparisons are useless. They test AIs on trivia questions or ask them to write poems, then declare a winner based on vibes. This comparison is different — I gave all three of the biggest AI tools the exact same real-world task that marketers actually need help with: rewriting a terrible cold sales email. One AI rewrote it so well that a sales coach with 15 years of experience said she'd have paid a copywriter $300 for that output. Here's exactly what happened, what each AI got right and wrong, and which one you should be reaching for next time you need to fix a bad email.
The Awful Email I Used — and Why I Chose This Specific One
The email I used as the test case wasn't made up. It's the kind of cold email that lands in your inbox every single Tuesday at 9 AM from a SaaS company you've never heard of.
Here it is, word for word:
"Hi [First Name], I hope this email finds you well. My name is Jake and I'm reaching out because I think our product could really benefit your business. We offer a wide range of solutions that help companies like yours improve their efficiency and drive more revenue. I'd love to hop on a quick call to learn more about your needs. Let me know if you have 15 minutes this week. Thanks, Jake."
Every marketer reading this just cringed. It opens with a phrase no human has ever actually felt. It's vague. It's about Jake, not the reader. It has no hook, no specificity, no urgency, and no reason whatsoever to reply.
This email is so common precisely because most people write like this when they don't know what they're doing — which is exactly why it's the perfect test. If an AI can rescue this, it proves something real.
The prompt I gave all three AIs was identical: "Rewrite this cold sales email to make it more compelling, specific, and likely to get a response. Keep it under 150 words. Focus on the reader's pain points, not the product features."
What ChatGPT Did — Solid, Fast, and Dangerously Average
ChatGPT (GPT-4o) produced a rewrite in about four seconds. It was clean. It was professional. It checked all the basic copywriting boxes.
It opened with a pattern-interrupt line, pivoted to a pain point about "wasted hours on manual processes," and ended with a low-friction CTA — "Worth a quick look?" instead of "Do you have 15 minutes?"
That's genuinely better. A beginner copywriter would be proud of this output.
But here's what I noticed when I read it twice: it felt like a template. The pain point it chose — "wasted hours on manual processes" — was so generic it could apply to 10,000 different products. ChatGPT made educated guesses about the reader's problems without actually solving the vagueness problem. Jake was no longer the subject, but the email still wasn't really about anyone specific.
The output would work. It would probably perform better than the original. But if you sent it, your prospect would still sense they're one of 500 people who got that exact message today. ChatGPT gave me a better template. It didn't give me a human email.
What Claude Did — and Why Copywriters Are Quietly Obsessed With It
Claude (Claude 3.5 Sonnet) took a noticeably different approach, and this is the section most AI comparison articles completely miss.
Where ChatGPT fixed the structure, Claude interrogated the intent. Before giving me a rewrite, it said something unexpected: it flagged that without knowing the specific industry, the recipient's role, or the product's actual differentiator, any rewrite would still be generic. Then it gave me two rewrites — one using placeholder specifics I'd need to fill in, and one that demonstrated what the email could look like if the sender sold, say, project management software to agency owners.
The agency-owner version was jaw-dropping. It opened with: "Running an agency means your team is probably juggling 12 tools, three Slack threads, and a client who changed the brief this morning." That's not copywriting. That's empathy with a keyboard.
Claude's secret weapon is what I call the "specificity ladder." It naturally moves from vague → specific → personal, and it shows you exactly where your original was standing on that ladder. This is the technique that separates $50 copy from $500 copy — and Claude does it instinctively.
The final output Claude produced clocked in at 118 words and included a concrete result in the first line ("We helped a 12-person agency cut their client onboarding time by 40%"), a one-sentence pain-point acknowledgment, and a CTA so light it barely felt like an ask: "I wrote up a 3-minute case study if you want to see how — no call required." The sales coach I mentioned earlier read it once, looked up, and said: "That's the email I wish my clients would send."
What Gemini Did — and the Surprising Thing It Got Right
Gemini (Google's Gemini 1.5 Pro) is the one most people sleep on, and in this test it earned some genuine respect — just not for the reason you'd expect.
The rewrite itself was middle-of-the-road. Tighter than the original, warmer than ChatGPT's, but not as precise as Claude's. It leaned into conversational language well — less corporate, more human — but it didn't bring the structural sophistication that made Claude's version sing.
Here's where Gemini surprised me: it was the only AI that proactively suggested a subject line and a P.S. line without being asked. The subject line it suggested — "Quick question about [Company]'s onboarding" — is exactly the kind of curiosity-gap opener that gets emails opened. The P.S. it added ("P.S. — If now's a bad time, I'm happy to send over a quick resource instead. No call needed.") is a classic objection handler that most email writers forget exists.
No one asked Gemini to think beyond the email body. It just did. That kind of proactive thinking is genuinely useful if you're moving fast and want a complete send-ready asset without having to remember every element yourself.
The lesson here: don't use Gemini for your deepest copy thinking, but absolutely use it when you want a fast first draft that's already more complete than you expected. Pair it with Claude for a one-two punch — Gemini to build the scaffold, Claude to make it human.
The Part Most People Get Wrong
Most people use AI to polish bad emails instead of rebuild them. That's the wrong move, and it's why their results stay mediocre.
Feeding a weak email to an AI with a prompt like "make this better" is like asking a personal trainer to improve your form while you're doing the wrong exercise. The AI will smooth out the language, maybe fix the CTA — but it won't challenge the underlying assumptions. You'll get a prettier version of the same bad strategy.
The right approach is to tell the AI what success looks like before you ask it to write anything. Try this prompt instead: "Before you rewrite this email, tell me the three biggest reasons a busy [job title] would delete it without replying. Then rewrite it to solve those three problems." That single change will produce dramatically better outputs from all three AIs.
The other mistake? Picking one AI and sticking with it out of habit. This test proved that each tool has a genuine, specific strength. Claude thinks more deeply about audience. ChatGPT moves faster and handles volume. Gemini builds complete assets and surprises you with extras. The smartest workflow isn't ChatGPT vs Claude vs Gemini — it's knowing which one to call for which job.
Key Takeaways
- Claude wins on depth: When you need copy that actually sounds like it was written for one specific person, Claude's "specificity ladder" approach beats the other two.
- ChatGPT wins on speed: For high-volume tasks where you need 10 decent drafts fast, GPT-4o is your workhorse — just plan to add specificity yourself afterward.
- Gemini wins on completeness: It thinks beyond the task you gave it, often adding subject lines, P.S. lines, and extras that make your final asset more usable out of the box.
- Your prompt matters more than the tool: The prompt "tell me why this fails, then fix it" consistently outperforms "make this better" across all three AIs.
- The real test is specificity: Any AI-written email that could be sent to 1,000 different people will perform like an email sent to 1,000 different people — poorly. Force your AI to get specific or the output is worthless.
What to Do Right Now
Open Claude right now and paste in the worst cold email in your sent folder — or use the Jake email above as your test case. Use this exact prompt: "Before you rewrite this email, list the three biggest reasons the recipient would delete it. Then rewrite it in under 150 words, using a specific pain point, a concrete result, and a CTA that doesn't ask for a meeting." You'll have a genuinely better email in under three minutes — and you'll immediately see why the marketers I showed this to stopped using generic AI prompts forever.