We Fed All Three AI Tools the Same Terrible Cold Email — Here's What Happened
We found the worst kind of cold email: generic opener, buzzword-heavy body, desperate close, zero personality. Then we handed it to ChatGPT, Claude, and Gemini with the exact same rewrite prompt and watched them work. The results weren't even close in the ways we expected — and the AI that "everyone" says is the best writer? It didn't win. What you're about to see will completely change how you pick the right tool for your copywriting tasks, and more importantly, how you prompt them to get results that actually convert.
The Original Email Was So Bad It Was Perfect for This Test
Here's the cold email we used as the raw material:
"Hi [Name], I hope this email finds you well. I wanted to reach out because I believe our cutting-edge solution could really help leverage your business to the next level. We help companies like yours achieve synergy and drive ROI. Would you be open to a quick 15-minute call this week?"
If you've ever worked in sales, you felt that in your bones. It hits every cringe checkpoint: hollow opener, zero specificity, vague value prop, and a close that sounds like a template from 2009 — because it is.
We gave all three AIs this exact prompt: "Rewrite this cold email to make it more personal, specific, and compelling. The target buyer is a marketing director at a mid-sized e-commerce brand. Keep it under 100 words." Same input, same instructions, same output format requested.
The goal wasn't just to see which one wrote prettier sentences. We wanted to see which AI understood sales psychology — specificity, relevance, and a reason to reply.
ChatGPT's Rewrite: Smooth, Safe, and Slightly Forgettable
ChatGPT produced a clean, professional rewrite in about four seconds. It swapped out the buzzwords, added a nod to e-commerce, and ended with a softer call to action. On the surface, it looks great.
Here's roughly what it gave us:
"Hi [Name], I noticed [Company] has been scaling its product catalog fast — impressive growth. Most e-commerce marketing directors I talk to hit a wall around email retention at that stage. We helped [Similar Brand] increase repeat purchase rate by 34% in 90 days. Worth a 15-minute conversation? I'll keep it tight."
That's genuinely better. It's specific, it leads with insight, the social proof feels real, and the close respects the reader's time. ChatGPT is excellent at this kind of structured persuasion writing because it's been trained on an enormous library of marketing and business content.
The weakness? It's predictable. If you've read a lot of "good" cold emails, you'll recognize the formula: observation → pain point → proof → soft ask. ChatGPT nails the formula. It doesn't surprise you.
Where ChatGPT truly shines in this workflow is iteration speed. You can follow up immediately with prompts like "Give me 3 versions — one aggressive, one casual, one ultra-short" and get usable options in under a minute. For volume output, nothing beats it right now.
Claude's Rewrite: The One That Made Marketers Stop Scrolling
Claude's output was the one that made a room of marketers go quiet — the good kind of quiet.
Here's what Claude produced:
"Hi [Name], Most cold emails promise to 'drive ROI.' This one won't. I'll just say this: e-commerce brands scaling past $5M usually have the same retention problem — customers buy once and vanish. We fixed that for [Brand] (34% more repeat buyers, 90 days). If that sounds familiar, I'd love 15 minutes. No deck, no pitch — just a real conversation."
Notice what Claude did differently. It broke the pattern — it acknowledged the format it was operating in and used that self-awareness as a hook. That's not a writing trick, that's a psychological move. It signals confidence and honesty in the same breath.
Claude, built by Anthropic, has a distinct voice that leans toward nuance and subtext. It's not just rewriting the email — it's thinking about why the reader would put their guard down. That's a different cognitive task, and Claude seems wired for it.
This comes from how Claude approaches instructions. Rather than executing the prompt literally, it tries to understand the intent behind the request. When you say "make it compelling," ChatGPT often interprets that as "make it well-structured." Claude often interprets it as "make it feel different."
The tradeoff is consistency. Claude's output can vary more run-to-run than ChatGPT. Sometimes you get magic. Sometimes you get something overly clever. The prompt "Be direct, subvert expectations, and write like a confident human — not a marketing tool" unlocks Claude's best work consistently.
Gemini's Rewrite: The Data-Aware Dark Horse
Gemini (Google's AI, especially Gemini Advanced) came in with an approach neither of the other two used — it structured the email around a question rather than a statement.
Here's what it produced:
"Hi [Name], Quick question — how's your repeat purchase rate looking right now? Most e-commerce marketing directors we talk to are seeing first-time buyers drop off after 60 days, even with strong acquisition numbers. We helped [Brand] solve exactly that with one workflow change — repeat purchases up 34% in three months. Open to hearing what they did? 15 minutes, your calendar."
The open-ended question opener is a real sales tactic with documented response-rate improvements, and Gemini reached for it without being asked. That's notable.
What Gemini does exceptionally well is research-informed writing. Because it's connected to Google's ecosystem, it tends to frame copy with patterns that reflect what actually performs in search and advertising — even when the task isn't SEO-related. That muscle shows up here in the structured, benefit-forward language.
Where Gemini struggled slightly was in voice. The email is effective, but it reads like it was assembled correctly rather than written confidently. It lacks the personality of Claude's version and the polish of ChatGPT's.
For the right use case — like writing email sequences at scale where testing and iteration matter more than a single perfect draft — Gemini's methodical approach is actually an asset. Pair it with a prompt like "Rewrite this email five ways, each testing a different psychological hook: curiosity, fear of loss, social proof, authority, and reciprocity" and you have a split-test arsenal ready to go.
The Part Most People Get Wrong
Most people treat AI copywriting like a vending machine: put in a prompt, take out a result, use it immediately. That's the wrong mental model — and it's why most AI-generated emails still feel robotic.
The real workflow is prompt → evaluate → refine → prompt again. Claude's first output wasn't the one that wowed the room. It was the third version, after a follow-up prompt: "Make it shorter, and add one line that shows we actually understand their specific problem — not just the category." That instruction pushed Claude from "interesting" to "I need to share this."
The other common mistake is using the same AI for every task because it's your default. ChatGPT is your best tool for fast volume and structured persuasion. Claude is your best tool when tone, originality, and voice matter most. Gemini is your best tool when you want methodical variation and data-aware framing. They're not interchangeable — they're specialists.
Finally, people forget to give AI the buyer's context. "Rewrite this email" is a weak prompt. "Rewrite this email for a marketing director at a $10M e-commerce brand who has tried three email tools and is skeptical of vendors" is a powerful one. The more specific you are about who's reading the email, the more targeted the output — across all three tools.
Key Takeaways
- ChatGPT: Best for fast, structured, high-volume email rewrites — follow up with variation prompts to unlock its full output power.
- Claude: Best when you need voice, subtext, and pattern-interrupting copy — prompt it to "subvert expectations" and it consistently delivers.
- Gemini: Best for methodical variation and psychologically structured drafts — ideal for building split-test sequences quickly.
- Specificity beats quality: A detailed prompt describing your exact buyer will outperform a vague prompt every single time, regardless of which AI you use.
- AI as a writing partner: The real output isn't the first draft — it's the third or fourth version after you've pushed back, refined, and challenged the tool.
What to Do Right Now
Open Claude and paste in the worst cold email in your current drafts folder — the one you haven't sent because you know it's not good enough. Use this exact prompt: "Rewrite this cold email for [specific job title] at [specific type of company]. Be direct, subvert expectations, and write like a confident human who genuinely understands this buyer's biggest frustration. Keep it under 100 words." Then run the same prompt through ChatGPT and compare. You'll immediately see the difference in how each tool interprets "compelling" — and you'll know which one to use for your next campaign.