I Gave ChatGPT, Claude, and Gemini the Same Cold Email Brief. The Results Were Not What I Expected.
Most people just pick one AI tool and stick with it — usually ChatGPT, because it's the one everyone talks about. But if you're using AI to write cold emails and you haven't tested the others head-to-head, you're probably leaving real money on the table. I ran the same cold email brief through ChatGPT (GPT-4o), Claude (Claude 3.5 Sonnet), and Gemini (Gemini 1.5 Pro) and scored each one on subject line quality, opening hook, personalization, clarity of the offer, and the call to action. The winner wasn't ChatGPT. It wasn't Gemini either. And the reasons why tell you something genuinely useful about how to write better cold emails — with or without AI.
The Exact Brief I Used (and What Each AI Did With It)
The brief was simple and realistic. Here's exactly what I sent to all three, word for word:
"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 The restaurant has an outdated website with no online ordering.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."
Same prompt. No extra coaching. No system prompts. Just the raw input so the comparison would be fair.
ChatGPT came back fast with something polished and professional. The subject line was: "Quick idea for [Restaurant Name]'s website." Clean. The body hit the pain point (outdated site, lost customers), made a specific offer, and ended with a soft CTA: "Would you have 15 minutes this week to chat?" Technically correct. Completely forgettable.
Gemini went a different direction — it opened with a compliment about the restaurant's food, which sounds nice but is actually a red flag in cold outreach. Compliments from strangers read as fake. The structure was fine, but the tone felt like it was written for a LinkedIn post, not a real human's inbox. It also clocked in at 187 words, blowing past the 150-word limit.
Claude did something neither of the others did: it led with a specific consequence instead of a feature. The subject line was "Your competitors are taking your takeout orders" — and the opening line was: "Most people searching for restaurants near you right now can't order from your site. They can, however, order from the place down the street." That's not just good copy. That's understanding what makes a cold email actually work.
Why Claude's Approach Reveals a Deeper Truth About AI and Persuasion
Here's the thing most AI comparison articles miss entirely: the difference between these outputs isn't really about which AI is "smarter." It's about which AI has a better internal model of human psychology in a sales context.
Cold emails fail for one reason above all others — they talk about the sender instead of the recipient's problem. ChatGPT's default mode is to be helpful and clear, which sounds great until you realize "helpful and clear" often produces writing that's easy to ignore. It answers the brief correctly but doesn't challenge the brief.
Claude's training — and Anthropic has talked about this publicly — puts heavy emphasis on understanding intent, not just instruction. When the brief said "get a 15-minute discovery call," Claude apparently asked itself: what emotional state does the recipient need to be in to say yes? The answer is mild anxiety about a real problem. That's why the output led with competitive loss, not with the designer's credentials.
This is a mental model worth burning into your brain: the best cold emails create a small, specific fear before they offer a solution. Not manipulation — just the basic structure of problem-aware persuasion. Claude nailed this intuitively. ChatGPT needed to be told.
You can actually test this yourself with a follow-up prompt to ChatGPT. Try: "Rewrite this cold email so the first two sentences make the reader feel like they're already losing business before I introduce myself." ChatGPT will execute that well — but you have to know to ask for it. Claude got there without the extra instruction.
How to Use This Comparison to Write Better Cold Emails Starting Today
You don't need to pick one AI and divorce the others. The smarter move is to use each one for what it's actually good at — and run a 20-minute workflow that combines all three.
Step 1: Use Claude for the first draft. Open Claude 3.5 Sonnet and use this prompt: "Write a cold email to [specific recipient type] from [your role]. Goal: [specific action]. Lead with the consequence of their problem before introducing any solution. Under 150 words. Include a subject line." Claude's first draft will almost always have the strongest emotional hook.
Step 2: Use ChatGPT to stress-test the logic. Paste Claude's draft into ChatGPT-4o and prompt: "Review this cold email. Does the offer feel specific and credible? Is the CTA frictionless? Rewrite any sentence that sounds like a template." ChatGPT is excellent at structural editing and spotting generic phrases.
Step 3: Use Gemini to check your subject line options. Gemini 1.5 Pro is surprisingly good at generating subject line variations quickly. Prompt: "Give me 10 subject line options for this cold email. Mix curiosity, specificity, and urgency. No clickbait." Gemini tends to produce volume better than the others — useful when you want to A/B test.
The whole workflow takes under 20 minutes and produces a cold email that has been shaped by three different AI reasoning styles. You're not relying on one model's blind spots.
The Part Most People Get Wrong
Most people treat cold email prompts like fill-in-the-blank templates: swap in the name, the industry, the offer — done. That's wrong, and here's exactly why it fails.
AI reflects the specificity you give it. A vague prompt produces a vague email. When you write "cold email to a business owner," every AI defaults to the most generic version of that person. But if you write "cold email to the owner of a 12-table Italian restaurant in a mid-size city who probably manages their own social media and hasn't updated their website since 2019," the output gets dramatically more specific and human.
The other mistake is accepting the first draft. Every single output I got in this test — including Claude's winner — needed at least one edit. Claude's subject line was punchy but slightly aggressive for some audiences. A/B testing matters. Treating any AI draft as final copy is how you end up with emails that sound like they were written by a robot, because they kind of were.
Don't skip the personalization layer either. AI can write a great structural email, but it can't know that the restaurant was featured in the local paper last month, or that the owner just opened a second location. That one real detail — added by you, in one sentence — will outperform any AI prompt trick every time.
Key Takeaways
- Claude 3.5 Sonnet: Consistently produces the strongest emotional hooks and leads with consequences — best for cold email first drafts.
- ChatGPT-4o: Excels at structural editing, logic-checking, and rewriting generic phrases — use it to sharpen, not start.
- Gemini 1.5 Pro: Great for generating high-volume variations fast, especially subject lines for A/B testing.
- The best cold email prompt: Always specify the recipient's problem and the emotional state you want to create — don't leave it for the AI to guess.
- The three-AI workflow: Draft with Claude → edit with ChatGPT → test subject lines with Gemini. Twenty minutes. Significantly better output than any single tool alone.
What to Do Right Now
Open Claude 3.5 Sonnet right now and paste this prompt: "Write a cold email to [your specific target audience]. Goal: [the one action you want them to take]. Lead the first two sentences with the consequence of a problem they're likely ignoring. Under 150 words. Include a subject line." Run the output through ChatGPT for a structural edit, then test two subject line variations from Gemini — and you'll have a send-ready cold email in the next 15 minutes that's better than 90% of what lands in people's inboxes.