ChatGPT vs Claude vs Gemini: Who Writes the Best Cold Email?
We Gave Three AIs the Same Cold Email Brief — Only One Actually Sounded Human
Cold email is one of the most brutal tests you can give an AI. There's no room for fluff, no hiding behind "great content" — either the person replies or they don't. We handed ChatGPT, Claude, and Gemini an identical brief: write a cold email from a freelance UX designer pitching a SaaS startup whose onboarding flow was visibly broken. Same brief, same context, same word limit. The results were shockingly different — and one of these tools produced an email that could genuinely close deals, while the others produced exactly what you'd expect from a robot. By the end of this article, you'll know which AI to use for cold outreach, why the other two fall short in ways most people never notice, and how to prompt any of them to perform at their ceiling.
The Test: Same Brief, Three Very Different Emails
The brief we used was deliberately specific. Here it is word for word:
"Write a cold email from a freelance UX designer named Maya to the founder of a B2B SaaS startup called Flowdesk. Their onboarding completion rate is publicly listed as 34% on their website. Maya wants to offer a UX audit. Keep it under 120 words. Sound like a real person, not a pitch deck."
That last line — sound like a real person — is where everything fell apart for two of the three tools.
ChatGPT (GPT-4o) delivered a clean, competent email. It opened with the 34% stat, made a clear offer, and had a soft CTA. But it still felt like it was written by someone who had read 500 cold emails and averaged them together. Phrases like "I'd love to connect" and "I'm confident I can help" crept in. It was safe. It was fine. It wouldn't get deleted, but it probably wouldn't get replied to either.
Gemini was the biggest surprise — in the wrong direction. It wrote a longer email, ignored the 120-word limit by about 40 words, and opened with a compliment about Flowdesk's "innovative approach to workflow management." Nobody talks like that. More importantly, nobody buys from an email that talks like that. It felt like a LinkedIn connection request from 2019.
Claude wrote something different. It opened with: "Your onboarding page says 34% of users complete setup. That means 66% are leaving before they ever see your product work — and most of them aren't coming back." No intro. No "my name is Maya." Just a fact that hurts a little, followed by a sentence that made the founder feel the problem. That's a cold email.
Why Claude Won — And the Specific Writing Technique Behind It
Claude didn't win because it's "smarter" in some abstract way. It won because it defaults to a specific cold email structure that the other two tools don't naturally reach for.
The structure is called Problem-Amplify-Solve (PAS), and Claude hit it almost instinctively. It stated the pain (34% completion rate), amplified it (66% of users gone forever), and then positioned Maya's audit as the specific fix. ChatGPT used a softer AIDA structure (Attention, Interest, Desire, Action) that works for marketing copy but dilutes cold email. Gemini used no recognizable structure at all.
The deeper reason Claude excels here is that it's been trained with a heavy emphasis on avoiding what Anthropic calls "sycophantic outputs." In plain terms: Claude is less likely to write something that sounds impressive but means nothing. That training difference shows up most clearly in high-stakes, short-form writing where every word has to do a job.
Here's the prompt that pushed Claude to its best output:
"Write a cold email under 120 words. Don't introduce yourself in the first line. Open with a specific fact about the reader's business that suggests a problem. Make the reader feel the cost of that problem before you offer anything. End with one low-friction question, not a call to action."
Try that exact prompt with all three tools and watch the gap widen. Claude will follow those constraints almost perfectly. ChatGPT will partially follow them. Gemini will add a subject line you didn't ask for and ignore the word count.
How to Use This Today: A 15-Minute Cold Email Workflow
You don't need to be a copywriter to send cold emails that get replies. You need a system, the right tool, and one prompt that does the heavy lifting.
Step 1: Gather one specific, public piece of data about your prospect. This could be a G2 review mentioning slow onboarding, a job posting that reveals a gap in their team, or a stat from their own website. The more specific, the better. If you can't find one, use ChatGPT with the prompt: "Find three publicly visible signals that [Company Name] might need help with [your service]. Use their website, job listings, and any press mentions."
Step 2: Open Claude and use this prompt framework:
"Write a cold email under 120 words from [Your Name], a [your role], to the [founder/head of X] at [Company Name]. The specific pain point is: [your data point]. Don't open with my name. Open with the pain point. Amplify the cost of that problem in one sentence. Then offer [your specific service] as the fix. End with one low-friction question like 'Is this something you're actively working on?'"
Step 3: Read the output out loud. This is the step people skip. If you stumble over a phrase, rewrite it. If it sounds like something a chatbot would say, it sounds like something a chatbot would say. Use Claude's follow-up capability — just type: "Rewrite the second sentence. Make it sound more like something I'd say to a friend who runs a startup."
The whole process takes 15 minutes. The email you end up with will be better than 90% of cold emails hitting people's inboxes right now — because most people are either not using AI at all, or they're using it with zero-context prompts like "write me a cold email."
The Part Most People Get Wrong
Most people treat AI cold email tools like a vending machine. They put in a generic prompt and expect a great email to come out. That's wrong, and here's exactly why it fails: AI writing tools are context engines, not creativity engines. The quality of the output is almost entirely determined by the specificity of the input.
When you type "write a cold email for my design agency" into any of these tools, you're asking it to write for everyone — which means it writes for no one. The result is an email that could have been sent to 10,000 people, and the person reading it can feel that.
The fix is to front-load your prompt with three pieces of specific context: who you are, what specific problem you've noticed about this prospect, and what exact outcome you're offering. That's it. Not your entire backstory. Not your pricing. Three specific things.
One more thing most people miss: subject lines matter more than the email body. Claude is better at body copy, but for subject lines, test ChatGPT with this prompt: "Write 10 cold email subject lines for this email. Make them feel like they were written by a human who did their homework, not a marketing team. No question marks. No emojis." Use the subject line from ChatGPT and the body copy from Claude. That combo is hard to beat.
Key Takeaways
- Claude for cold email body copy: It defaults to problem-first structure and avoids filler language better than the other two tools.
- ChatGPT for subject lines and research: Use it to mine public data about your prospect and generate subject line variants to A/B test.
- Gemini for cold email: Skip it — it ignores constraints, adds fluff, and opens with compliments that nobody believes.
- Prompt specificity is everything: A vague prompt produces a generic email; a specific prompt with three concrete context points produces something that gets replies.
- The PAS framework (Problem-Amplify-Solve): This is the structure that separates cold emails that convert from ones that get deleted — build it into every prompt you write.
What to Do Right Now
Open Claude right now and paste this prompt, swapping in your own details:
"Write a cold email under 120 words from [Your Name], a [your role], to [prospect's title] at [Company Name]. The specific pain point I've noticed is: [one specific thing]. Don't open with my name. Open with the pain point. End with one low-friction question."
Send the first version to yourself, read it out loud, then spend five minutes tightening it. That's your new cold email template — built in under 10 minutes.