ChatGPT vs Claude vs Gemini: Who Writes the Best Cold Email?
I Sent the Same Cold Email Brief to All Three AI Giants — Here's What Happened
I gave ChatGPT, Claude, and Gemini the exact same brief: write a cold email from a freelance web designer pitching a local restaurant on a website redesign. Same instructions. Same context. Same goal. What came back were three completely different emails — different tones, different structures, different levels of persuasion — and when I showed the results to six marketers I trust, none of them agreed on which was best. That fight got interesting fast. If you use AI to write cold emails for your business or your clients, this test reveals something that most comparison articles completely ignore: it's not about which AI is "smarter." It's about which one understands selling. By the end of this article, you'll know exactly which tool to open for cold outreach — and which prompting tweak makes all three dramatically better.
The Exact Brief I Used — and the 3 Very Different Emails It Produced
The brief was simple on purpose. Here's what I gave each tool, word for word:
"Write a cold email from a freelance web designer to a local Italian restaurant called Bella Napoli. The goal is to get a reply and a discovery call. Keep it under 150 words. Sound human, not salesy."
That's it. No extra context. No examples. No tone guidance beyond "human, not salesy." This is how most people actually use these tools — and that's exactly why the results matter.
ChatGPT (GPT-4o) delivered a structured, clean email with a clear subject line, a compliment about the restaurant, a pain point (their current site looks outdated on mobile), and a soft CTA asking for a 15-minute call. It was competent. It felt like something a junior copywriter would hand in on their second week. Solid, but safe.
Claude (Claude 3.5 Sonnet) went a different direction entirely. It opened with a specific, almost conversational observation — something like noticing that Bella Napoli's food photos on Instagram are stunning but the website doesn't do them justice. It felt like a real human had actually looked at the business before emailing. The CTA was warmer, more curious, less transactional. Every marketer I showed it to leaned forward a little reading it.
Gemini (Gemini 1.5 Pro) produced the longest email of the three — and broke the 150-word limit by about 40 words. It included bullet points listing the designer's services, which immediately made it feel like a brochure. The subject line was generic. It read like an AI wrote it, which is the one thing you absolutely cannot afford in a cold email.
The Hidden Variable Nobody Talks About: Tone Calibration
Here's what most AI comparison articles skip entirely: the raw output is almost never the final product. What actually matters is how well each tool responds to refinement — and this is where the real differences show up.
After the first round, I added one sentence to the brief: "The email should feel like it came from someone who genuinely researched Bella Napoli — mention something specific about their online presence without making it creepy."
ChatGPT improved noticeably. It added a reference to the restaurant's Google review score and tied it to the web design pitch ("You've built a 4.8-star reputation — your website should show that the moment someone lands on it"). Smart pivot. Good instinct.
Claude barely needed the extra instruction because it had already gone in that direction — but when pushed further, it produced what I'd call a "warm specificity" technique: referencing something emotionally resonant (the restaurant's anniversary post on Instagram) rather than just a metric. That's a fundamentally different level of persuasion. It's the difference between complimenting someone's shoes and complimenting the specific memory behind them.
Gemini, even with refined instructions, struggled to ditch the bullet points and kept the tone formal. It's not that Gemini is bad at writing — it's genuinely excellent at research, summarization, and structured content. Cold email just isn't its native language. Knowing a tool's "native language" is the skill most people never develop.
The takeaway here is a mental model worth keeping: think of these three AIs as three different writers on your team. ChatGPT is the reliable generalist. Claude is the one who gets persuasion and nuance. Gemini is the researcher you bring in before the writing starts.
How to Use This Right Now to Write Cold Emails That Actually Get Replies
Stop treating AI like a vending machine where you drop in a brief and catch whatever falls out. Here's the workflow that gets real results — and you can run it in under 20 minutes.
Step 1: Do your research in Gemini first. Open Gemini and prompt it: "Research [business name] and give me 5 specific, public-facing details about their brand, online presence, and any visible gaps — website, social media, reviews." Gemini's real-time web access makes it the best tool for this step. You're not writing yet — you're loading your context.
Step 2: Write the email in Claude. Take the specific details Gemini surfaced and drop them into Claude with this prompt: "Write a cold email using these specific details about [business]. Goal: get a discovery call. Under 150 words. Open with something that proves I looked at their business. End with one low-friction question, not a hard CTA." Claude's natural persuasion instincts will do the rest.
Step 3: Run an A/B subject line test with ChatGPT. Paste the email Claude wrote into ChatGPT and say: "Give me 5 subject line variations for this cold email — ranging from ultra-direct to curiosity-driven. No clickbait." ChatGPT is fast, reliable, and surprisingly good at subject line variety. Pick two and test them.
Step 4: Send, track, iterate. Use a tool like Instantly or Lemlist to track open rates and replies. If you're getting opens but no replies, your body copy needs work — go back to Claude. If opens are low, your subject line is the problem — go back to ChatGPT. The AI doesn't stop being useful after you hit send.
This three-tool stack sounds like more work, but it takes about 15 minutes once you've done it twice — and it outperforms any single-tool approach because you're using each AI for what it's actually built for.
The Part Most People Get Wrong
Most people try to make one AI do everything — research, writing, subject lines, follow-ups — all in a single chat thread. That's wrong. Here's why: every time you ask an AI to switch tasks mid-conversation, you're fighting its context and momentum. Claude mid-persuasive-email doesn't want to suddenly become a subject line machine. The output gets muddier every time you pivot.
The second mistake is using the same prompt for cold email that you'd use for any other writing task. Cold email is a specific discipline — it lives or dies on the first line, it has a word count ceiling, and it requires the reader to trust a stranger in under 10 seconds. Generic writing prompts produce generic emails. You need prompts that encode those constraints from the start.
The third mistake — and this one kills response rates — is asking the AI to "not sound like AI" without telling it what to sound like instead. "Sound human" is not a direction. "Sound like someone who genuinely did 10 minutes of research on this business before writing" — that's a direction. Give the AI a behavior to replicate, not just a quality to avoid.
Fix these three things and you'll see reply rates jump before you change anything else.
Key Takeaways
- Claude wins on persuasion: For cold email specifically, Claude's natural tone, specificity, and instinct for low-friction CTAs outperform the other two out of the box.
- Gemini wins on research: Use Gemini to surface specific, real details about your prospect before you write a single word — that's where it adds the most value.
- ChatGPT wins on speed and iteration: Subject line variations, quick rewrites, and structural tweaks are ChatGPT's strongest move in this workflow.
- Refinement beats raw output: How a tool responds to a second instruction matters more than its first draft — always test with a follow-up prompt before you judge an AI's ceiling.
- Specificity is the only thing that makes cold email work: Whether you're writing it yourself or with AI, a detail that proves you looked at the business is worth more than any copywriting formula.
What to Do Right Now
Open Claude and paste this prompt in the next 10 minutes: "Write a cold email from [your role] to [specific type of business]. Under 150 words. Open with one specific observation about their online presence that proves I looked at them. End with a single low-friction question. Don't use bullet points. Make it feel like it came from a real person." Fill in your details, read what comes back, and then ask Claude to make it 20% warmer. That one iteration will show you more about what this tool can do than any comparison article — including this one.