I Sent the Same Terrible Sales Email to ChatGPT, Claude, and Gemini — Here's What Happened
Most AI comparison articles test things that don't matter. I wanted to test something you'll actually use: rewriting a bad sales email into one that gets replies. So I wrote the worst cold email I could — generic opener, vague value prop, weak CTA — and dropped it into ChatGPT, Claude, and Gemini with the exact same prompt. The results were not equal. One AI gave me something I'd archive. One gave me something decent. And one gave me something I'd genuinely send tomorrow. Here's the full breakdown, and more importantly, here's what it tells you about which tool to reach for next time you're staring at a blank draft.
The Terrible Email I Used (And Why I Made It This Bad on Purpose)
Here's the original email I fed all three AIs, word for word:
"Hi, my name is Davide and I work at a company that offers software solutions for businesses. We have helped many companies improve their results. I think you might be interested in what we do. Would you love to jump on a quick call to learn more? Let me know. Thanks."
This is the kind of email that gets deleted before the second sentence. It has every classic failure: no specificity, no social proof with real numbers, a subject line I didn't even write, and a CTA so soft it practically apologizes for existing. I made it this bad deliberately, because the real test of an AI isn't how it handles a half-decent draft — it's how much it can rescue a total disaster.
The prompt I gave all three was identical: "Rewrite this cold sales email to make it more compelling, specific, and likely to get a reply. Keep it under 150 words."
No extra context. No persona. No additional instructions. Because most people don't give detailed prompts when they're in a hurry, and I wanted to see what each AI does with minimal input.
The constraint of 150 words was intentional too. A rewrite that balloons to 400 words isn't useful — it tells you the AI padded rather than sharpened.
What ChatGPT, Claude, and Gemini Actually Produced
ChatGPT (GPT-4o) delivered a clean, structured rewrite fast. It invented a plausible company name, swapped "improve your results" for something like "cut onboarding time by 30%," and rewrote the CTA to a specific ask: a 15-minute call on Thursday or Friday. The subject line it added — "Quick idea for [Company Name]" — is a proven format. It felt like a competent sales rep had touched it.
The problem? It felt slightly templated. Like ChatGPT has absorbed so many "how to write cold emails" blog posts that it produces the platonic ideal of a cold email — technically correct, but not surprising. If you've read sales email advice from the last five years, you've basically read ChatGPT's output here.
Gemini gave me the most polished-looking result at first glance. It formatted nicely, included a subject line, and organized the email with a clear problem-solution structure. But when I read it closely, it had reintroduced vagueness in different clothing — phrases like "transformative results" and "streamlined workflows" that sound good but say nothing.
Gemini also broke my 150-word limit by about 40 words without flagging it. That's a small thing, but it matters: following constraints is part of the job.
Claude (Claude 3.5 Sonnet) did something the other two didn't. It asked me a clarifying question before rewriting — or rather, it rewrote the email and added a note at the bottom explaining its assumptions and offering to adjust based on my actual industry, ICP, or tone preference. That's not just a rewrite. That's a thinking partner. The email it produced was the most specific of the three, the most human-sounding, and the only one I'd actually send without editing.
The Hidden Skill That Separates a Good Rewrite From a Great One
Here's what most comparisons miss: the difference between a rewrite and a transformation.
A rewrite fixes the words. A transformation changes the structure of persuasion. The original email was weak not just because the language was bad — it was weak because it led with me (the sender) instead of you (the reader). That's the real problem. And only Claude's version fundamentally inverted the structure, opening with the prospect's likely pain point before introducing any solution.
Think of it like this: a weak sales email is a selfie. A strong one is a mirror. The reader should see themselves in every sentence — their problem, their goal, their specific situation. Claude's rewrite opened with something like "If you're still manually tracking [X process] in spreadsheets, you're probably losing 3-4 hours a week to work that shouldn't exist." That's a mirror.
This matters because it's a mental model you can apply before you even open an AI tool. Before you prompt anything, ask yourself: does this email start with me or with them? If it starts with you, the sender, flip it. Then give that flipped version to Claude with the prompt: "Rewrite this email so every sentence either addresses the reader's problem or proves I can solve it. Cut anything that's about me."
The AI can fix words all day. Your job is to give it the right structural brief. The best output comes from combining your strategic thinking with the AI's language ability — not from outsourcing both.
How to Run This Test on Your Own Sales Emails Today
You don't need to take my word for any of this. Run the experiment yourself in the next 20 minutes and you'll learn more about these tools than any comparison article can teach you.
Step 1: Pull up your worst-performing cold email. Not a draft — an actual email you've sent that got zero replies. Real failures teach more than hypotheticals.
Step 2: Open Claude.ai (free tier works fine) and paste in this prompt: "Here's a cold sales email I wrote that isn't getting replies. Rewrite it to be more specific, more reader-focused, and more likely to get a response. Keep it under 150 words and add a subject line. After the rewrite, tell me what you changed and why."
That last sentence — tell me what you changed and why — is the move most people skip. When Claude explains its reasoning, you're learning a transferable skill, not just getting a fixed email.
Step 3: Take Claude's output and drop it into ChatGPT with this prompt: "Here's a sales email. Make it punchier. Tighten every sentence. Remove any word that doesn't do real work." ChatGPT is excellent at this kind of line-level editing — think of it as your copy editor after Claude acts as your strategist.
Step 4: Check the output against three questions: Does it open with the reader's problem? Does it include one specific, credible number or proof point? Does the CTA ask for something small and specific? If yes to all three, it's ready to test.
Run this workflow on five emails this week. By email five, you'll have internalized what "good" looks like — and you'll be faster and less dependent on AI for the first draft.
The Part Most People Get Wrong
Most people treat AI rewrites like a vending machine: put in bad email, get out good email, send it. That's wrong, and here's why it keeps failing them.
The AI has no idea who you're emailing. It doesn't know your prospect's industry, their company size, their specific pain point, or whether they've heard your pitch before. When you give Claude or ChatGPT a vague email and ask for a rewrite with no context, you're asking it to guess. It'll produce something that sounds good but fits nobody in particular — which is exactly the problem with the original email.
The fix is simple: give context before you ask for the rewrite. A one-sentence brief changes everything. Try: "My prospect is a Head of Operations at a 50-person e-commerce brand. Their biggest pain right now is fulfillment delays. Rewrite this email with that context in mind." That single addition will produce a more relevant output than any amount of prompt engineering tricks.
The second mistake is treating the first output as final. The best sales email writers I know use AI the same way they use a first draft — as raw material, not a finished product. Read it out loud. If it sounds like a LinkedIn automation bot wrote it, it's not ready.
Key Takeaways
- Claude for strategy: Claude produces the most structurally sound rewrites and explains its reasoning — use it when you need to understand why a change works, not just see the change.
- ChatGPT for polish: GPT-4o is the strongest line-level editor of the three — use it to tighten a draft that's already strategically sound.
- Gemini for formatting: Gemini produces well-organized, visually clean output, but watch for vague language sneaking back in — always read it critically before sending.
- Context is everything: Any AI rewrite is only as good as the context you give it — always include your prospect's role, their pain point, and the tone you want.
- The mirror principle: Strong sales emails open with the reader's problem, not your product — check your structure before you prompt anything.
What to Do Right Now
Open Claude.ai, paste in the worst cold email sitting in your sent folder, and use this exact prompt: "Rewrite this cold sales email to focus on the reader's problem first. Keep it under 150 words, add a subject line, and explain what you changed and why." Do it in the next 10 minutes while this is fresh — and when Claude explains its edits, read every word. That explanation is worth more than the rewrite itself.