I Tested ChatGPT, Claude, and Gemini on the Same Cold Email Brief — One Model Wasn't Even Close
I gave all three of the biggest AI tools the exact same cold email brief and told them nothing else. No extra context, no system prompts, no hand-holding — just a raw task and a blank text box. What came back surprised me, and honestly, it should change how you think about which AI tool you're actually using for your business writing. One model produced an email I'd send right now. One produced something that would get you instantly ignored. And one landed somewhere frustrating in the middle — technically correct, but completely forgettable. By the end of this article, you'll know exactly which tool to use for cold outreach, why the differences matter more than you think, and how to get the best possible output from whichever one you already have.
The Brief Was Identical — Here's Exactly What I Gave Each Model
The test had to be fair, so I kept the input identical across all three tools: ChatGPT (GPT-4o), Claude (Claude 3.5 Sonnet), and Gemini (Gemini 1.5 Pro).
Here's the exact prompt I used — word for word:
"Write a cold email from a freelance web designer named Alex to the owner of a local bakery called Sweet Layers. Alex wants to offer website redesign services. The bakery's current site is slow and outdated. Keep it under 150 words. Subject line included."
No templates. No tone instructions. No "make it friendly but professional." I wanted to see what each model defaulted to when left to its own devices — because that's how most people actually use these tools.
The results were wildly different in three specific ways: opening line quality, personalization depth, and call-to-action clarity. These are the exact three things that determine whether a cold email gets a reply or gets deleted in 0.3 seconds.
One model nailed all three. Two did not. Let's break it down.
ChatGPT Wrote a Solid Email — But It Sounded Like an AI Wrote It
ChatGPT's output was clean, structured, and competent. Subject line: "Quick idea for Sweet Layers' website." The body opened with a compliment, mentioned the slow load times, pitched the redesign, and closed with a calendar link offer. Technically, it hit every beat.
Here's the problem: it felt assembled, not written. The opening line was "I came across your website and noticed it could use some improvements." That sentence has been in a thousand cold emails. It's the written equivalent of a firm handshake from someone who took a sales course in 2009.
The call to action was also vague. It said something like "I'd love to chat if you're interested." That puts all the work on the recipient to figure out what "chat" means and how to make it happen. Cold email experts will tell you — a weak CTA kills your reply rate faster than anything else.
ChatGPT's email would pass a basic review. But it wouldn't stand out in a flooded inbox, and for a bakery owner who gets three of these a week, "doesn't stand out" means "deleted."
Verdict on ChatGPT: Competent, generic, forgettable. A good starting point, not a finished product.
Claude Did Something the Other Two Didn't Even Attempt
Claude's output was different from the first sentence — and not by accident.
The subject line it produced: "Your croissants deserve a better website." That's a real subject line. That's the kind of line a clever human copywriter charges $150/hour to come up with. It's specific, it's warm, and it immediately signals that this isn't a copy-paste blast email.
The opening line matched that energy: "I stopped by Sweet Layers' site after a friend mentioned your almond cake — and while the cake sounds incredible, I noticed the site takes about 8 seconds to load on mobile, which is likely costing you online orders." There's a specific number. There's a human detail. There's a consequence — not just a problem, but what the problem is actively costing the business owner.
This is what specificity as trust-building looks like in cold email. Claude didn't just identify a pain point — it quantified it and connected it to revenue. That's a fundamentally different level of persuasion.
The CTA was sharp too: "Would a 15-minute call this week work?" Specific time commitment. One clear question. Easy to say yes to.
Claude produced an email I would send with minimal edits. That almost never happens on a zero-context prompt.
Verdict on Claude: The clear winner. Personalized, specific, conversion-ready.
Gemini Tried Too Hard and Ended Up Saying Nothing
Gemini's email had the most words and the least impact — which is a brutal combination in cold outreach.
The subject line: "Elevate Your Online Presence with a Fresh Website Design." That phrase — "elevate your online presence" — appears in so many bad marketing emails that most spam filters have probably started flagging it. It sounds like it was written by someone who has heard marketing buzzwords but has never actually done marketing.
The body followed the same pattern. Lots of "I specialize in creating visually stunning, user-friendly websites" and "I would love the opportunity to collaborate." These phrases mean nothing because they could apply to any freelancer emailing any business on the planet. There's zero signal to Sweet Layers that Alex has thought about them specifically.
Gemini also went slightly over the 150-word limit — a small thing, but it tells you something about how each model handles constraints. Claude came in at exactly 143 words. ChatGPT hit 148. Gemini landed at 167. When a client brief says "under 150 words," that's a hard rule, not a suggestion.
Verdict on Gemini: Bloated, generic, and full of the exact phrases a busy bakery owner rolls their eyes at before hitting delete.
The Part Most People Get Wrong
Most people think the AI tool is the problem when their cold emails don't work. It's not. The prompt is the problem.
Every single output I showed you above came from the same prompt — and Claude still won. But here's what's important: even Claude's great email could have been made dramatically better with one additional sentence in the prompt. Something like: "Write it in a warm, conversational tone. Reference a specific detail about their business that makes it feel personal." Claude was already doing this instinctively — imagine what it produces when you explicitly ask for it.
Most people treat AI like a vending machine. They punch in the minimum, expect a finished product, and then blame the tool when the output is mediocre. The real skill isn't picking the right AI — it's knowing how to brief it like a creative director would brief a copywriter.
If you'd given Gemini Claude's level of instruction — specific tone, specific detail requirements, specific word count enforcement — the gap would have closed significantly. The tools have different ceilings, yes. But most people are operating way below even Gemini's ceiling because their prompts are too vague.
Use the right tool and write better prompts. Both matter. Only focusing on one of them is leaving serious performance on the table.
Key Takeaways
- Claude 3.5 Sonnet: Currently the strongest default performer for cold email — it leads with specificity and writes like a human copywriter without being told to.
- Subject line quality: This is where models diverge fastest — Claude's subject lines feel personal; Gemini's feel like newsletter spam.
- Specificity = trust: The best cold emails name a specific number, detail, or consequence — and Claude does this instinctively better than the other two.
- Prompt quality matters as much as tool choice: Even the best AI produces generic output from a vague brief — give it context, tone, and constraints.
- ChatGPT is a strong second: With a more detailed prompt, GPT-4o can close most of the gap with Claude — it just needs more direction to get there.
What to Do Right Now
Open Claude.ai, paste in the exact prompt I used above, and then run it again with this addition at the end: "Make the subject line hyper-specific to their business. Include one concrete detail or statistic in the first two sentences. End with a single, low-friction yes/no question." Compare the two outputs side by side — you'll see in under 10 minutes exactly why prompt precision is the real skill worth mastering.