I Fed the Same Terrible Cold Email to ChatGPT, Claude, and Gemini — Here's What Happened
Most people pick their AI tool by habit, not by results. They open ChatGPT because it's familiar, or Claude because someone on Twitter said it writes better, and they never actually test the difference side by side. I did the testing for you — I took one genuinely awful cold email and fed the exact same prompt to all three, and the results were night and day in ways you'd never predict. By the end of this article, you'll know which AI writes the most persuasive sales copy, which one plays it too safe, and which one pulled off something I didn't expect. Spoiler: the default choice isn't the best choice here.
The Terrible Email I Used (And the Exact Prompt I Gave Each AI)
Here's the original email. It's bad on purpose — but it's also the kind of email thousands of real salespeople send every day:
"Hi, I wanted to reach out because I think our product could really help your business. We offer a wide range of solutions that are tailored to your needs. Let me know if you're interested in a call. Thanks, Dave"
It's vague, it's self-centered, it has no hook, no proof, and no reason for the reader to respond. Perfect test subject.
I gave each AI this exact prompt: "Rewrite this cold email to make it more persuasive. The audience is a busy B2B decision-maker. Keep it under 120 words. Focus on their pain, not our product."
Then I sat back and watched three very different things happen.
The prompt itself is important — I wasn't asking for "a better email." I was giving each AI a specific constraint and a clear audience. That matters because vague prompts produce vague results across every tool.
What ChatGPT, Claude, and Gemini Actually Produced
ChatGPT (GPT-4o) came out swinging with structure. It opened with a sharp pain-point line — something like "Most [role] I talk to are losing 3-5 hours a week to [problem]." It followed a classic cold email formula: hook, proof, soft ask. The result was clean, professional, and immediately usable. It hit the word count almost exactly and felt like something a trained copywriter would produce on their third draft.
Claude (Claude 3.5 Sonnet) did something different. Instead of leading with pain, it opened with a specific, almost conversational observation — it felt warmer, more human, less "template-y." The line it wrote felt like something you'd actually say to someone at a conference, not something you'd read in a HubSpot playbook. It also added a subtle compliment to the recipient's company that made it feel personalized, even though it had no actual context to work with.
Gemini (Gemini 1.5 Pro) gave me three variations unprompted. That's actually a big deal — instead of one rewrite, I got a conservative version, a bold version, and a mid-range option. The boldest version had the strongest subject line of all three tools. But the body copy in all three felt slightly more formal, slightly less natural. Good bones, slightly stiff execution.
Here's the brutal truth: ChatGPT is the most reliable, Claude is the most human-sounding, and Gemini is the most structured in its output format — but none of them are perfect out of the box.
The Hidden Variable That Changed Everything: Follow-Up Prompting
This is the part most comparison articles completely ignore. They show you round one and call it done. But the real power gap between these tools shows up in round two — when you push back.
After the first rewrite, I followed up with each AI using this prompt: "The opening line is too generic. Make it feel like I actually researched this person's company. Give me 3 alternative opening lines that reference a real business problem a SaaS company at Series B might face."
ChatGPT delivered three solid options fast — they were specific, they referenced recognizable pain points like churn, CAC, and team scaling, and they felt plausible even without real research data. This is where GPT-4o's massive training breadth pays off.
Claude wrote opening lines that felt almost uncomfortably good. One of them was: "Scaling from 50 to 150 people usually breaks the processes that got you here — and the cracks show up first in [department]." That's not a template. That's a line a great salesperson would actually write. Claude seems to understand nuance at a level that still surprises me.
Gemini gave me accurate but slightly mechanical options. They were correct, useful, and would absolutely work — but they felt generated, not written. You'd likely edit them more before sending.
The mental model to take away: Don't judge an AI on its first answer. Judge it on how well it responds to refinement. The best AI for writing isn't the one that nails it immediately — it's the one that gets sharper the more you push it.
How to Run This Test on Your Own Emails Today
You don't need to take my word for any of this. Here's exactly how to run your own version of this test in under 15 minutes.
Step 1: Find the weakest email in your current outreach sequence — the one with the lowest reply rate. Copy it.
Step 2: Open ChatGPT, Claude, and Gemini in three separate tabs. Paste this prompt into all three: "Rewrite this cold email. Audience is [specific job title]. Keep it under 120 words. Lead with their pain, not my product. End with one low-friction call to action." Then paste your email below it.
Step 3: Read each output out loud. Seriously — out loud. The one that sounds most like a real person talking is usually the winner for cold outreach. Your ear catches things your eyes miss.
Step 4: Take the best version and go one round deeper. Ask: "Rewrite the opening line to be more specific to a company going through rapid hiring." Or whatever fits your actual prospect. See which AI responds best to the refinement.
Step 5: Take the winner and test it. Send 20 emails with the AI-rewritten version against your original. Give it two weeks. The data will tell you more than any comparison article ever could.
This whole process takes less than 20 minutes the first time. After that, you'll have a repeatable system for improving every piece of outreach you write.
The Part Most People Get Wrong
Most people treat AI like a vending machine — they put in a prompt and expect the first result to be ready-to-use. That's wrong, and it's why most AI-written emails still sound like AI-written emails.
The first output is a rough draft, not a finished product. Even the best prompt to Claude or ChatGPT gives you a 70-80% solution. The remaining 20-30% comes from your follow-up — your pushback, your specificity, your context. If you skip that step, you're leaving the best version on the table.
The other mistake: using the same tool for every task. Based on this test, I'd use Claude for emotional, relationship-driven copy — think cold outreach, LinkedIn messages, and nurture sequences. I'd use ChatGPT when I need fast, structured, reliable output I can edit quickly. I'd use Gemini when I want multiple angles fast and want to pick the strongest direction.
Don't be loyal to one tool. Be loyal to results.
Key Takeaways
- ChatGPT (GPT-4o): The most consistent and structured rewriter — great when you need a clean, usable draft fast.
- Claude (Claude 3.5 Sonnet): The most human-sounding output — best for emotional copy, cold outreach, and anything where tone matters most.
- Gemini (Gemini 1.5 Pro): Gives you multiple versions by default — useful for exploring angles, but body copy often needs more editing.
- Follow-up prompting: The real quality gap between tools shows up in round two — always push back with a more specific request.
- Tool selection: Match the AI to the job — no single tool wins every category, and switching between them is a skill worth building.
What to Do Right Now
Open Claude and paste this prompt with your worst-performing sales email: "Rewrite this cold email for a busy [job title]. Under 100 words. Start with their problem, not my solution. End with one easy ask." Read the output out loud, then follow up with: "Make the first sentence more specific to a company that is scaling fast." You'll have a better email in the next 10 minutes — and a process you can repeat forever.