I Sent the Same Cold Email Brief to ChatGPT, Claude, and Gemini — Here's What Happened
I gave all three AI giants the exact same task: write a cold email to a busy SaaS founder pitching a freelance copywriting service. Same brief, same context, same word limit. The results were so different they might as well have come from three completely different planets. One of them produced an email I'd actually send tomorrow. One produced something that read like a LinkedIn post from 2019. And the winner? It wasn't ChatGPT — which is what nearly everyone guesses first. Cold emails are one of the highest-stakes writing tasks you'll ever hand to an AI, so getting this comparison right could directly change how much money lands in your inbox.
I Gave All Three the Exact Same Brief — Here's What Came Back
The brief I used was specific on purpose. Vague prompts give vague results, and I wanted to test each model under real-world conditions.
The prompt was: "Write a cold email from a freelance copywriter to a SaaS startup founder. The founder's name is Alex. The copywriter specializes in onboarding email sequences. Keep it under 150 words. No fluff, no 'I hope this finds you well.' Make it feel human and get a reply."
ChatGPT (GPT-4o) delivered fast and clean. The subject line was punchy — "Your onboarding emails are losing you customers" — and the body was structured and confident. It followed best practices almost mechanically: one problem, one proof point, one CTA. It was good. Like, genuinely good. But it felt like it had read every cold email playbook ever written and averaged them together.
Gemini went a different direction. It wrote a slightly longer email with a warmer, more conversational tone. The problem? It ignored the word limit and opened with "I came across your product and I'm really impressed by what you're building" — which is exactly the kind of line that gets an email deleted before the second sentence. It also used the phrase "synergy" once, which should be illegal in 2024.
Claude (claude-3-5-sonnet) did something neither of the other two did: it wrote an email that sounded like a real person sent it at 9pm after actually researching the company. The subject line was "3 onboarding emails that usually tank trial-to-paid conversion" — specific, curiosity-driven, slightly alarming. The body named a real, concrete problem, hinted at a solution without giving it away, and ended with a question instead of a demand. It was under 140 words. Claude won this round and it wasn't particularly close.
Why Claude's Email Actually Works (And What the Others Are Missing)
Most AI writing tools are optimized to sound professional. Claude seems optimized to sound intentional — and that's a completely different thing when it comes to cold email.
The biggest difference came down to specificity of tension. ChatGPT's email identified a problem. Claude's email made you feel the problem. There's a line in Claude's output that read something like: "Most SaaS tools lose 60% of trial users in the first 7 days — and the onboarding emails are usually why." That's not just a claim. That's a stat that makes a founder's stomach drop a little.
This connects to a principle called "owned insight" — the idea that the best cold emails don't just name a pain point, they reframe it in a way the prospect hasn't quite articulated to themselves yet. Claude does this naturally. It doesn't just write what the problem is, it writes why that problem exists, which positions the sender as someone who actually understands the space.
ChatGPT, to its credit, gets close with the right prompting. If you add "Include a counterintuitive insight or reframe the problem in a way the reader hasn't heard before" to your ChatGPT prompt, you'll close most of the gap. But Claude does it without being asked, which matters when you're generating emails at volume.
Gemini's core issue is different. It's trained to be agreeable and collaborative, which makes it incredible for brainstorming and planning. But cold email is not collaborative. Cold email is a controlled interruption — and Gemini's instinct to warm up the room before making an ask works against it in this format.
How to Use This to Write Better Cold Emails Starting Today
Here's the exact workflow to get Claude-quality cold emails consistently, regardless of which tool you use.
Step 1: Build a "Context Bomb" before you write anything. Paste this into Claude or ChatGPT before your email request: "You are a senior B2B copywriter. You understand that cold emails fail when they sound like cold emails. You write like a thoughtful human who has done their research. Before writing any email, you will identify the single most specific pain point relevant to this prospect."
Step 2: Use the "What keeps them up at night?" framework. After your context bomb, add: "The prospect is [Name], [Title] at [Company]. Their company does [X]. Their biggest challenge right now is probably [Y]. Write a cold email under 150 words that opens with that challenge — not with who I am." This single shift — leading with them, not you — is responsible for most of the difference between emails that get replies and emails that get ignored.
Step 3: Ask for three subject line variations. Add "Give me 3 subject line options: one curiosity-based, one problem-based, one contrarian" to your prompt. Test them. The contrarian one almost always outperforms.
Step 4: Run a "delete the first sentence" test. This is the most underrated editing move in cold email. Take whatever AI writes, delete the opening sentence, and re-read from sentence two. If it flows better — and it usually does — sentence one was just the AI warming itself up. Delete it.
The whole process takes under 10 minutes per email once you have your context bomb saved. If you're sending 20 cold emails a week, this workflow alone is worth hundreds of dollars in improved reply rates.
The Part Most People Get Wrong
Most people treat AI like a cold email vending machine. They type "write me a cold email to sell X" and expect something that works. That's not wrong — it's just the floor, not the ceiling.
The real mistake is using the output as the final draft. Every email any of these three tools generates is a first draft. Claude's first draft is better than the others. But none of them know that your prospect just fired their head of marketing, or that they posted a frustrated tweet about their churn rate last week. That context, layered on top of a strong AI draft, is what turns a reply rate from 5% to 20%.
The second mistake is prompting for tone instead of prompting for strategy. Asking AI to "write in a friendly tone" is almost useless. Asking it to "write an email where the prospect feels like I've been observing their problem from the inside" is the kind of strategic brief that unlocks the model's real capability.
The third mistake is running all three tools and picking the best one every time. That's exhausting and inefficient. Based on this test: use Claude for cold email and outreach copy, use ChatGPT for email sequences and follow-up frameworks, and use Gemini for researching your prospect before you write anything. Each tool has a lane. Put it in its lane.
Key Takeaways
- Claude wins cold email: For persuasive, human-sounding outreach copy, Claude 3.5 Sonnet consistently outperforms ChatGPT and Gemini without extra prompting.
- Context Bomb prompt: Setting the AI's identity and constraints before making your writing request is the single highest-leverage move in AI copywriting.
- Lead with their pain, not your pitch: The best-performing cold emails open with the prospect's specific problem — not with who you are or what you do.
- Delete the first sentence: AI almost always uses the opener to warm up. Delete it, start from sentence two, and your emails will instantly sound more human.
- Each tool has a lane: Use Claude for cold email drafts, ChatGPT for building full sequences and frameworks, and Gemini for prospect research before you write.
What to Do Right Now
Open Claude right now and paste this: "You are a senior B2B copywriter. Write a cold email under 150 words to [Name], [Title] at [Company]. Lead with their biggest pain point — [insert pain point]. No 'I hope this finds you well.' End with a question, not a CTA." Send that email to one prospect today, then run the same prompt in ChatGPT and compare the outputs side by side — you'll see the difference immediately, and you'll never prompt an AI for cold email the same way again.