ChatGPT vs Claude vs Gemini: Who Writes the Best Cold Email?
I Sent the Same Brief to All Three AIs — One Outperformed the Others By a Mile
I gave ChatGPT, Claude, and Gemini the exact same cold email brief: pitch a freelance UX designer to a SaaS startup founder, keep it under 150 words, and make it feel human. Three AI tools. One identical prompt. Wildly different results. What I found didn't just surprise me — it changed how I use these tools for client work every single day. One AI wrote like a confident human. One wrote like a press release. And one gave me something I genuinely almost sent word-for-word. If you're using AI to write cold emails and you're not testing across multiple models, you're leaving serious money on the table.
The Exact Brief I Used (And Why the Prompt Setup Matters More Than You Think)
The prompt I sent to all three was identical:
"Write a cold email from a freelance UX designer pitching their services to a SaaS startup founder. The designer has 5 years of experience and has worked with two Series A companies. The email should be under 150 words, conversational in tone, and end with a low-friction CTA. No subject line needed."
Clean, specific, and with clear constraints. That's intentional. Vague prompts produce garbage from every AI — so I needed a level playing field.
ChatGPT (GPT-4o) came back fast with something polished and readable. It had a good opener, referenced the designer's experience naturally, and landed on a soft CTA: "Would it be worth a 15-minute call this week?" Solid. Felt like a competent copywriter on their third draft.
Claude 3.5 Sonnet did something different. It opened with a specific pain point — "Most SaaS founders I talk to say their onboarding drop-off is a UX problem they haven't had time to fix" — before introducing the designer at all. That's a real copywriting technique called leading with the wound. It immediately felt like someone who understood sales psychology, not just writing.
Gemini 1.5 Pro was the most disappointing. The email was grammatically fine, but it read like a LinkedIn connection request. Generic phrases like "I believe my skills could be a great fit" and "I'd love to connect" killed any sense of personality. If you sent this email cold, it would get ignored in under two seconds.
The gap between Claude and Gemini here wasn't small. It was the difference between a reply and a delete.
What the Outputs Actually Revealed About How Each AI Thinks About Persuasion
This is the part most comparisons skip, so let's go deeper.
ChatGPT writes to satisfy the prompt. It's extremely good at hitting your stated requirements — word count, tone, structure. But it defaults to "competent" rather than "compelling." The email it produced was correct in every way. It just wasn't dangerous. By dangerous I mean: the kind of email that makes someone stop scrolling and actually feel something.
Claude writes to move the reader. This sounds vague, so here's what I mean: Claude seems to apply something closer to a persuasion framework by default. It front-loads value. It identifies a likely objection before you even ask it to. When I ran a follow-up prompt asking Claude to "rewrite the email assuming the founder gets 50 cold pitches a week," it completely restructured the opening without me explaining why that matters. It got it.
Gemini writes to sound professional. That's its default setting — safe, polished, forgettable. For internal documents, summaries, or research tasks, Gemini is genuinely strong. For cold outreach where personality is your competitive edge? It consistently underdelivers compared to the other two.
Here's the deeper insight: each AI has a default emotional register. ChatGPT defaults to neutral-confident. Claude defaults to curious-empathetic. Gemini defaults to formal-polished. Cold email lives and dies on emotional resonance — which is exactly why Claude had the edge in this specific test.
Knowing each model's default register means you can either choose the right tool for the job, or you can prompt against the default — which is where things get really interesting.
How to Use This Information to Write Better Cold Emails Starting Today
Here's exactly how to put this into practice in the next 30 minutes.
Step 1: Start with Claude for your first draft. Open Claude 3.5 Sonnet and use this structure for your prompt:
"Write a cold email from [your role] to [specific recipient type]. They receive [X] cold pitches a week, so the first sentence must immediately address a real problem they face — not introduce me. Keep it under [word count]. End with [specific CTA]. Tone: [one word — e.g. direct, warm, confident]."
The key difference from a basic prompt is the instruction to lead with their problem, not your pitch. Claude responds to that constraint brilliantly.
Step 2: Take that draft into ChatGPT for a tightening pass. Paste the Claude draft and say: "Make this 20% shorter without losing any of the impact. Don't add new information — just cut." GPT-4o is excellent at trimming. Claude can be slightly verbose. This two-model workflow consistently produces cleaner output than using either alone.
Step 3: Read your final email out loud. Not to Gemini. Not to another AI. Out loud, yourself. If you stumble on a word, rewrite it. If it sounds like an email, it's not good enough. It should sound like something you'd actually say to someone at a conference.
Step 4: A/B test your subject lines using ChatGPT. Ask it for "10 subject line variations — 5 that sound like they're from a real person, 5 that lead with a specific outcome the reader wants." Then pick the two strongest and split test them in your outreach tool.
Cold email is still the highest-ROI outreach channel for B2B. Getting your AI workflow right here compounds fast.
The Part Most People Get Wrong
Most people treat all three AIs like they're interchangeable spell-checkers with extra features. They paste in a prompt, accept the first output, and send it. That's wrong — and it's the reason most AI-written cold emails read exactly like AI-written cold emails.
Each model has a distinct strength profile. Using Claude for persuasive outreach and GPT-4o for structural cleanup isn't just splitting hairs — it's the difference between a 4% reply rate and a 12% reply rate. The model you choose for the job matters as much as the prompt you write.
The second mistake is over-prompting for tone. Telling an AI to "sound human" or "make it conversational" almost never works. These are vague instructions that produce generic results. Instead, give the AI a specific constraint — "write like you're texting a colleague, not emailing a stranger" or "use one contraction per sentence" — and you'll get dramatically more natural output.
The third mistake is not editing the output at all. AI gets you 80% of the way there faster than any human. But that last 20% — a personal reference, a specific data point from their LinkedIn, one line that only you could write — is what turns an email from "decent" to "actually responded to."
Key Takeaways
- Claude 3.5 Sonnet: The strongest performer for cold email because it defaults to persuasion-first structure, not just competent writing.
- ChatGPT GPT-4o: Excellent for trimming, refining, and generating subject line variations — best used as a second pass, not a first draft.
- Gemini 1.5 Pro: Strong for research and professional documents, but consistently produces generic output for personality-driven writing tasks.
- Default emotional register: Every AI model has one — knowing it lets you choose the right tool or prompt against the default for better results.
- Two-model workflow: Draft in Claude, tighten in ChatGPT — this combination consistently outperforms using either tool alone for outbound copy.
What to Do Right Now
Open Claude 3.5 Sonnet and paste this prompt, customized for your actual use case:
"Write a cold email from [your role] to [specific recipient]. Lead with a real problem they face — don't introduce me until sentence two. Under 120 words. End with a one-line CTA that requires minimal effort to respond to."
Send the output to ChatGPT for a tightening pass, read it out loud, add one line only you could write — and you'll have a cold email worth sending in under 15 minutes.