I Gave All Three AI Giants the Same Cold Email Brief — The Winner Wasn't Who I Expected
I've watched people argue about ChatGPT vs Claude vs Gemini for two years, but almost nobody tests them on the thing that actually makes or costs businesses money: cold outreach. A great cold email can land a $10K client. A bad one gets deleted in 0.3 seconds. So I gave all three the exact same brief and graded every word. The results were genuinely surprising — and one model made a mistake so obvious it should disqualify it from any sales conversation. By the end of this article, you'll know exactly which AI to open the next time you need to write an email that gets a reply.
The Exact Brief I Sent All Three (And What I Was Looking For)
Every test started with the same prompt, word for word. I didn't tweak it for any model, didn't give bonus context, didn't coach them. That's the only fair way to do this.
The prompt was: "Write a cold email from a freelance web designer to a local restaurant owner. The goal is to get a 15-minute discovery call. The restaurant has an outdated website with no online ordering. Keep it under 150 words. Subject line included."
I was scoring each email on four things: specificity (does it sound like it was written for this person?), clarity of the offer (do they know exactly what's being proposed?), call-to-action strength (is the ask clear and low-friction?), and tone (does it feel human or like spam?).
I ran the test three times per model, took the best output from each, and compared them side by side. This wasn't a vibe test. This was as close to a controlled experiment as you can get without a lab coat.
What I found made me rethink which AI I'm recommending to my audience for anything sales-related.
ChatGPT Plays It Safe — Almost Too Safe
ChatGPT produced a clean, competent email. Subject line: "Quick idea for [Restaurant Name]'s website." It opened with a compliment about the restaurant's food, acknowledged the missing online ordering, and pitched a free 15-minute call. Ticked every basic box.
The problem? It was generic in a way that felt intentional. The compliment could apply to any restaurant on earth. There was no real hook — no specific pain point that would make a busy restaurant owner stop scrolling and think, "Wait, this person actually gets my situation."
ChatGPT also loves the phrase "I'd love to connect" — and sure enough, it showed up in all three of my test runs. That phrase is the cold email equivalent of a participation trophy. It signals nothing.
Where ChatGPT genuinely shines is structure and length discipline. It hit under 150 words every time, the formatting was clean, and the call-to-action was direct. If you're new to cold email and just need a solid first draft to edit, ChatGPT is a reliable starting point.
But "reliable starting point" isn't winning a competition. It's showing up.
Claude Does Something the Other Two Don't Even Attempt
Here's the thing nobody talks about when they compare these models: Claude thinks about the reader's psychology, not just the content of the message.
Claude's subject line came back as: "Your competitors are taking orders online — here's a 15-min fix." That's not just a subject line. That's a threat and a promise in ten words. It triggers loss aversion (competitors are ahead of you) and curiosity (what's the fix?) in the same breath.
The body of Claude's email did something tactically smart: it led with a specific cost, not a vague benefit. Instead of "I can improve your website," it opened with something like, "Restaurants without online ordering lose an estimated 20–30% of potential delivery revenue to competitors who make it one click easier." Now the reader has a number in their head. Numbers create urgency. Vague promises don't.
Claude also structured the ask differently. Instead of "Would you be open to a quick call?" — which puts the mental burden on the recipient to decide — it used an assumptive close: "I have two spots open this week — Tuesday or Thursday work for you?" That's a sales technique called the alternative close, and most humans forget to use it. Claude used it unprompted.
The tone was warmer than ChatGPT's without being sycophantic. It felt like a message from someone who had actually walked past the restaurant and noticed the problem. That specificity of voice is Claude's biggest competitive advantage right now, and it's massively underrated.
How to Use These Results to Write Cold Emails That Actually Get Replies Today
You don't need to pick one model and ignore the others. The smartest move is to use all three in a single workflow — each one for the part it's best at.
Step 1: Start with Claude for the first draft. Use this prompt: "Write a cold email from [your role] to [target person/business]. Their specific problem is [X]. My offer is [Y]. The goal is a 15-minute call. Use a subject line that triggers loss aversion. Keep it under 150 words." Claude's psychological awareness will give you a strong opening hook and a close that actually converts.
Step 2: Paste Claude's draft into ChatGPT and ask it to tighten. Use: "Edit this cold email for clarity and length. Cut anything that doesn't earn its place. Keep it under 120 words." ChatGPT's structural instincts are excellent for trimming fat without losing meaning.
Step 3: Run a final tone check with Gemini. Ask: "Does this email sound like a real person or like a template? What one sentence would you change to make it feel more human?" Gemini's responses here tend to be surprisingly candid — it'll flag anything that sounds robotic in a way the other two won't always catch.
This whole workflow takes under 15 minutes. You'll come out the other end with a cold email that has Claude's psychology, ChatGPT's structure, and a Gemini-approved humanity check. That's better than any single model can produce alone.
The Part Most People Get Wrong
Most people paste their brief into one AI, copy the output, and send it. That's not using AI — that's using a template generator with extra steps.
The models are not finished writers. They're first-draft engines. Claude's email was the best of the three, but I still edited it before I'd ever send it to a real prospect. I swapped in a specific restaurant name, changed the statistic to one I could actually verify, and adjusted the tone to sound more like me. That took four minutes.
The other mistake: people ask for "a cold email" without specifying the one job the email needs to do. A cold email doesn't need to sell your service. It needs to sell one thing: the call. The moment you try to explain your pricing, your portfolio, and your process in 150 words, you've already lost. Tell the AI your goal is the call and nothing else — watch the output transform.
Stop judging AI by the raw output. Judge it by what you can turn the raw output into.
Key Takeaways
- Claude wins the cold email test: Its psychological framing — loss aversion hooks, assumptive closes, specific numbers — produces the highest-converting first drafts of the three models.
- ChatGPT is the best editor: Use it to trim and structure after you have a strong draft, not to generate the draft itself.
- Gemini is your tone checker: It's surprisingly good at flagging robotic phrases and suggesting more human alternatives.
- The three-model workflow beats any single model: Claude → ChatGPT → Gemini in sequence produces better cold emails than any one AI alone.
- Specificity is the only thing that gets replies: The AI can't add real details about your prospect — that's your job, and it's the difference between a reply and the trash folder.
What to Do Right Now
Open Claude right now and paste this prompt: "Write a cold email from [your role] to [specific target]. Their problem is [one specific pain point]. My offer is [one specific outcome]. Goal: get a 15-minute call. Subject line should trigger loss aversion. Under 150 words." Edit it for two minutes with your own specifics, run it through ChatGPT to tighten the length, and you'll have a cold email worth sending before the hour is up.