I Sent the Same Cold Email Brief to ChatGPT, Claude, and Gemini — The Results Were Shocking
I gave all three major AI tools the exact same cold email brief and asked them to write a compelling outreach email from scratch. One of them nailed it so well I could have sent it immediately. One was decent with a few tweaks. And one was so generic it read like a 2009 email marketing template. The differences weren't small — they revealed something important about how each AI actually thinks about persuasion, tone, and human psychology. If you're using AI to write sales emails, client outreach, or any kind of cold copy, what you're about to read could save you from sending embarrassing emails — and help you close more deals.
The Exact Brief I Gave All Three AIs (And Why It Matters)
The only way to run a fair test is to give every AI the same starting point. No extra prompting, no hand-holding, no "make it better" follow-ups. Just the brief — and whatever they produce first is what gets judged.
Here's the exact brief I used:
"Write a cold email from a freelance web designer named Alex to a small business owner named Sarah who runs a local bakery. Alex has noticed the bakery's website looks outdated and loads slowly. The goal is to get a 15-minute discovery call. Keep it under 150 words. Friendly, not pushy."
That brief has everything a good cold email needs to work with: a specific sender, a specific recipient, a real pain point, a clear ask, and a tone direction. If an AI can't produce something great from this, it's not the brief's fault.
I also kept the temperature settings at default for all three. No special system prompts, no Claude Projects, no Custom Instructions in ChatGPT. This is the out-of-the-box experience most people actually use.
The results were dramatically different — and the gap between first and last place was bigger than I expected.
Here's What Each AI Actually Wrote (Word for Word)
Let's get into the real output. I'm not paraphrasing — this is what each tool produced on the first try.
ChatGPT (GPT-4o) opened with: "Hi Sarah, I came across your bakery online and I have to say — your croissants look incredible. But I noticed your website might be holding you back from getting even more customers through the door." It stayed personal, referenced a specific detail (the croissants), connected the pain point to her actual goal (more customers), and ended with a soft, low-pressure ask for 15 minutes. Word count: 134. Tone: warm and direct. It felt like a real person wrote it.
Claude (Claude 3.5 Sonnet) did something unexpected — it wrote a subject line without being asked. The subject was: "Quick question about your website." That's smart. That subject line has a 40–50% higher open rate than something like "Web Design Services" because it triggers curiosity without sounding salesy. The body was clean, the ask was specific, and Claude even included a P.S. line that added social proof naturally. It was the most strategically complete email of the three.
Gemini (in Google Workspace) gave me something that felt like it was pulled from a "cold email templates" website circa 2017. The opening line was: "My name is Alex and I am a freelance web designer. I specialize in creating modern, fast-loading websites for small businesses." It led with the sender, not the recipient. That's the single biggest mistake in cold email writing — and Gemini walked right into it. The email was 162 words (over the limit), contained zero personalization, and read like a cover letter.
The ranking after round one: Claude first, ChatGPT second, Gemini third — and it wasn't close.
What the Winning Email Got Right That Most People Never Think About
Claude's email won because it understood the psychology of cold outreach, not just the structure. This is the layer most AI comparisons skip — they grade the output, not the thinking behind it.
The "you before me" rule is the golden rule of cold email. Every sentence in your email should be about the recipient's world, not your credentials. Claude's email had four sentences about Sarah's bakery before it mentioned Alex once. Gemini did the opposite. ChatGPT split the difference but still led with the recipient.
Claude also nailed the single ask principle. It didn't offer a proposal, a free audit, a portfolio link, AND a call. It asked for one thing — 15 minutes — and made that ask feel small and easy. The moment you give someone three options in a cold email, they subconsciously treat it like homework and do none of it.
The subject line Claude generated unprompted — "Quick question about your website" — works because it's honest. It IS a quick question. Deceptive subject lines ("Re: your inquiry") have short-term open rates but destroy trust on the first scroll.
Here's the mental model worth keeping: cold email is not a sales pitch, it's a conversation starter. Claude seemed to understand that. Gemini wrote a pitch. There's a massive difference — one opens doors, the other closes them.
If you want to test this yourself, try this follow-up prompt in Claude: "Rewrite this email using the 'you before me' principle — don't mention the sender's name or credentials until at least the third sentence." The output gets even sharper.
How to Use This Comparison to Write Better Cold Emails Starting Today
Here's the practical playbook. You don't need to pick just one AI and stick with it — you need to know which tool to use for which job.
Step 1: Start with Claude for your first draft. Give it a brief like the one I used above. Be specific about the recipient, the sender, the pain point, the ask, and the word count. The more specific your brief, the better the output. A good template to follow: "Write a cold email from [name + role] to [name + company type]. The pain point is [X]. The goal is [specific ask]. Keep it under [word count]. Tone: [adjective]."
Step 2: Run the same brief through ChatGPT. GPT-4o often adds a warmth and specificity that Claude sometimes trades for structure. You're not looking for a winner — you're looking for lines to steal. Maybe Claude's structure + ChatGPT's opening line = your final email.
Step 3: Use this follow-up prompt in whichever tool gave you the best draft: "Now rewrite the subject line five different ways — from curiosity-based to direct to ultra-short. Show me all five." This takes 10 seconds and gives you split-test material immediately.
Step 4: Don't use Gemini for cold email copy. Use it for research — it's excellent at pulling real-time information about a prospect's company, recent news, or industry context. Feed that research into Claude or ChatGPT to create hyper-personalized opening lines. That combination is genuinely powerful.
The whole workflow — brief, draft in Claude, compare with ChatGPT, steal the best lines, generate subject lines — takes about 8 minutes. That's faster than most people spend staring at a blank Google Doc.
The Part Most People Get Wrong
Most people treat AI cold email writers like a vending machine — put in a vague request, expect a perfect email to fall out. That's wrong. Garbage in, garbage out is never more true than in cold outreach.
The most common mistake is writing a brief that's too generic. Prompts like "Write a cold email for my web design business" will produce exactly the kind of template soup Gemini gave me. The AI has no one to write to, no real pain point to work with, and no specific ask to aim at. It fills the gaps with filler — and filler is what gets deleted.
The second mistake is accepting the first draft as final. Every AI output is a starting point, not a finished product. Read it out loud. If it sounds like an AI wrote it, it needs another pass. The sentence "I hope this email finds you well" should be deleted on sight — no human under 60 starts a cold email that way.
The third mistake is ignoring the subject line. Over 50% of cold emails get opened or deleted based on the subject line alone — before a single word of your email is read. Claude proved it understands this by writing one unprompted. If your AI doesn't give you a subject line, ask for five variations before you send anything.
Key Takeaways
- Claude 3.5 Sonnet: The best first-draft cold email writer of the three — it thinks strategically about structure, subject lines, and persuasion psychology.
- ChatGPT GPT-4o: Excellent for warm, specific, human-sounding copy — use it to improve tone and add personality to Claude's structure.
- Gemini: Skip it for cold email writing — use it to research prospects and feed that intel into the other tools instead.
- The "you before me" rule: Your email should talk about the recipient's world before it ever mentions you — this single principle separates emails that get replies from ones that get deleted.
- Specific briefs win: The more detail you give an AI (recipient, pain point, word count, tone, goal), the better the output — vague prompts produce template copy that no one reads.
What to Do Right Now
Open Claude right now and paste this brief: "Write a cold email from [your name + role] to [a specific prospect type]. They have [one specific pain point]. The goal is to get a 15-minute call. Under 150 words. Friendly, not pushy. Include a subject line." Read what comes back, then run the exact same brief in ChatGPT and steal the best line from each. You'll have a cold email worth sending in under 10 minutes — and you'll immediately see why the tool you choose actually matters.