I Gave ChatGPT, Claude, and Gemini the Same Cold Email Brief — and the Winner Wasn't Who I Expected
Most people just pick one AI tool and stick with it. That's a mistake, especially when cold email is involved — where a single word choice can be the difference between a reply and a delete. I ran a head-to-head test: same brief, same target audience, same goal, all three major AI tools. What came back was genuinely surprising, and it changed how I write cold outreach entirely. By the end of this, you'll know exactly which AI to use, when to use it, and what prompt gets the best result out of each one.
The Exact Brief I Gave All Three Tools
The setup was simple. I wrote one brief and copy-pasted it into ChatGPT (GPT-4o), Claude (Claude 3.5 Sonnet), and Gemini (Gemini 1.5 Pro) without changing a single word.
The brief: "Write a cold email from a freelance UX designer to a Series A startup founder. The goal is to land a 20-minute discovery call. Keep it under 100 words. Don't use a generic opener. Make it feel human."
That last instruction — make it feel human — is where things got interesting. All three tools technically followed the brief. But "technically correct" and "actually good" are very different things in cold email.
Here's what came back. ChatGPT wrote something clean and structured — subject line, punchy opener, one value prop, clear CTA. It was good. Like, genuinely good. If a junior copywriter sent this to a client, they'd be happy.
Gemini's version was longer, more formal, and opened with a question — "Are you struggling to convert users on your onboarding flow?" — which is exactly the kind of opener that gets emails deleted. It read like a marketing email pretending to be a personal one. Not what we asked for.
Why Claude Won — and What It Did Differently
Claude's email did something the others didn't: it felt like it came from a real person who had done 10 minutes of homework.
Where ChatGPT opened with a value statement, Claude opened with a specific observation. Something like: "I noticed you just closed your Series A — congrats. Your onboarding flow has two friction points that typically tank retention at this stage." It was specific, confident, and didn't smell like a template.
This is Claude's superpower: contextual specificity. When you give Claude a detailed brief, it reaches for concrete details rather than generalizations. ChatGPT tends to optimize for structure. Claude optimizes for voice. In cold email, voice wins.
The subject line Claude generated was also the best of the three: "Your onboarding has two problems" — short, curiosity-driven, slightly uncomfortable. That's a subject line people actually open.
Here's the key insight. Claude performs best when your brief has texture. The more context you give it about the recipient, the company, and the specific pain point, the sharper the output gets. ChatGPT is more resilient with a thin brief. Claude needs feeding — but when you feed it, it delivers.
The Prompt Technique That Changed My Results Completely
Most people prompt these tools the same way they'd Google something. That's why they get generic output.
The technique that unlocked dramatically better cold emails across all three tools is what I call the "recipient profile first" method. Before you ask for the email, you describe the recipient in detail — their role, their current stage, their likely biggest headache this quarter, and what they're skeptical about.
Here's the exact structure I use now: "The recipient is [Name/Role] at a [Stage] company in [Industry]. They're likely dealing with [Specific Problem]. They're skeptical of [Common Sales Pitch]. They respond well to [Tone/Style]. Now write a cold email that..."
When I ran Claude with this expanded brief, the output improved by about 40% — more specific, better subject line, tighter CTA. When I ran ChatGPT with the same expanded brief, it also improved significantly. Gemini? Still the weakest of the three, but noticeably better than its first attempt.
The lesson isn't just about which tool is "best." It's that your prompt is 60% of the result. The tool is just the engine. You're the driver.
How to Run Your Own AI Cold Email Test in Under 30 Minutes
You don't need to take my word for it. Run this yourself and see what works for your specific audience.
Step 1: Write one clear brief. Include: your role, the recipient's role and company stage, the goal of the email (call, reply, demo), a word count limit, and one instruction about tone. Keep the brief under 100 words.
Step 2: Open tabs for ChatGPT, Claude, and Gemini — all free tiers work for this. Paste the exact same brief into all three. Don't adjust anything.
Step 3: Read each output out loud. Seriously — read it out loud. You'll immediately hear which one sounds like a human and which one sounds like a robot in a blazer.
Step 4: Take the best one and run it back through Claude with this follow-up prompt: "Make this 20% shorter, remove any phrase that sounds like a sales email, and punch up the subject line." Claude is the best editor of the three — even if it didn't write the original.
Step 5: Send it. Don't overthink this. The best cold email is the one that actually goes out. You can optimize in the next round based on real reply data.
The Part Most People Get Wrong
Most people use AI to write the whole email from scratch and send it without editing. That's wrong, and it's costing them replies.
AI cold emails have tells. Phrases like "I hope this email finds you well," "I wanted to reach out," or "I'd love to connect" are AI fingerprints at this point. Experienced founders and sales leaders see these from the first sentence. The email is dead before the second line.
The right way to use AI in cold email is as a first draft engine, not a finisher. Let Claude or ChatGPT give you the structure, the angle, and the subject line. Then you spend 5 minutes making it sound like you — one personal detail, one thing you genuinely noticed about their company, one line that only makes sense in this specific context.
Also: stop treating "shorter" as an automatic win. The best cold email is the shortest one that still makes a specific, credible point. A vague 50-word email is worse than a sharp 90-word one. Brevity without specificity is just emptiness.
Key Takeaways
- Claude 3.5 Sonnet: Best for cold email when you give it a rich, detailed brief — it writes with the most natural voice and specificity.
- ChatGPT (GPT-4o): Most consistent with a thin brief — reliable structure and a clean CTA, even without a lot of context.
- Gemini 1.5 Pro: Weakest for cold outreach — tends toward formal, generic openers that feel like marketing copy, not personal email.
- The "recipient profile first" prompt: Describe the recipient in detail before asking for the email — this single change improves output across all three tools.
- AI as draft, not final: Always edit the AI output by reading it out loud and removing anything that sounds templated — that's where real reply rates live.
What to Do Right Now
Open Claude right now and paste this prompt: "I'm a [your role] reaching out to a [recipient role] at a [company stage] company. Their biggest challenge right now is [specific problem]. Write a cold email under 90 words with a subject line that creates curiosity without being clickbait." Read what comes back out loud, cut any phrase that sounds like it could apply to anyone, and add one line that's specific to their company. That's your email. Send it today.