I Gave ChatGPT, Claude, and Gemini the Same Cold Email Brief — Here's What Happened
Most AI comparisons are useless. They test generic tasks, declare a winner based on vibes, and send you back to using whatever tool you already had open. This one is different. I gave all three of the biggest AI tools — ChatGPT, Claude, and Gemini — the exact same cold email brief, scored them across five specific dimensions, and the results exposed a gap that most marketers never see coming. If you're using AI to write sales emails and you haven't run this test yourself, you're almost certainly leaving money on the table.
The Test: One Brief, Three AIs, Zero Mercy
Every AI got the same brief — word for word. No extra coaching, no system prompts, no hand-holding. Here's exactly what I gave them:
"Write a cold email from a freelance UX designer named Sara to a SaaS startup founder. The goal is to get a 30-minute discovery call. The startup recently launched a new mobile app. Keep it under 150 words. Subject line included."
That's a real-world brief. Specific sender, specific target, specific goal, real constraint. This is the kind of brief a freelancer or agency copywriter would actually use.
I scored each output on five dimensions: subject line strength, opening hook, personalization depth, call-to-action clarity, and overall tone. Each dimension scored out of 10.
Here's the short version of the results before we go deep.
ChatGPT scored highest on structure and CTA clarity. Claude wrote the most human-sounding email by a wide margin. Gemini was the most generic — and in cold email, generic is a death sentence.
But the raw scores aren't the interesting part. What each AI chose to prioritize tells you something much more important about how to use them strategically.
Claude Wrote Like a Human — And That's Not an Accident
Claude's output was the most surprising. Given no extra instructions, it opened with a line that felt like it came from a real person, not a template.
Its subject line: "Quick thought on your app launch" — not "Exciting Opportunity" or "Let's Connect." It avoided every cliché in the cold email playbook without being told to.
The opening line read: "Saw you just shipped your mobile app — congrats on the launch. I had a few ideas about the onboarding flow that might be worth a quick chat." That's specific, warm, and relevant. It signals research without being creepy about it.
This happens because Claude is trained with a strong emphasis on natural conversation and nuanced tone. When your task is inherently human — like a one-to-one email between two people — Claude's training gives it a structural edge. It defaults to sounding like a person, not a marketing department.
The lesson here is critical: use Claude when the tone of your writing needs to feel genuinely personal. Job application emails, investor outreach, partnership proposals, creator collabs — anything where sounding robotic will kill the opportunity before it starts.
ChatGPT Won on Structure — Here's the Hidden Reason That Matters
ChatGPT's email wasn't the warmest. But it was the most deployable — immediately.
Its structure followed a clean Problem → Relevance → CTA framework without being asked. The subject line was clear: "UX Help for Your New App — 30-Min Chat?" Not poetic, but functional. A/B test that subject line against ten others and it'll hold its own.
The CTA was specific: "Would Thursday or Friday work for a quick 30-minute call? Happy to work around your schedule." Two options, low friction, concrete ask. That's cold email 101 — and ChatGPT nailed it without prompting.
Why? ChatGPT has been trained on an enormous volume of sales copy, email marketing content, and conversion-focused writing. It's absorbed the structure of high-performing emails the way a senior copywriter absorbs frameworks after years of client work.
Here's the move most people miss: use ChatGPT when you need a solid first draft fast, then run it through Claude to humanize the tone. Give ChatGPT the prompt: "Write a cold email using a Problem → Solution → CTA structure. Outcome: book a 30-minute call." Take that output, paste it into Claude, and say: "Rewrite this so it sounds warmer and more personal — like one human talking to another, not a marketer." That two-step combo beats either tool alone.
Gemini Played It Safe — And That's a Problem You Need to Understand
Gemini's email was fine. Technically correct. Grammatically clean. Completely forgettable.
The subject line: "Freelance UX Designer Available for Your App Project." That's a résumé header, not a cold email subject line. No intrigue, no specificity, no reason to open it.
The opening went straight into credentials: "My name is Sara and I'm an experienced UX designer with over 5 years working with SaaS companies..." Nobody who receives cold email cares about your credentials in the first sentence. They care about themselves — their product, their problem, their launch.
This isn't a knock on Gemini as a tool overall. Gemini is genuinely impressive for research, synthesis, multimodal tasks, and pulling live information from the web. But in this test, it defaulted to safe, formal, and generic — the three things a cold email cannot afford to be.
The mental model to take away: match the AI to the cognitive task. Gemini excels at analytical and research-heavy work. Claude excels at human, emotional, nuanced writing. ChatGPT excels at structured, conversion-focused copy. Sending Gemini in to write persuasive one-to-one email is like asking your CFO to run a standup comedy set. Wrong tool, wrong room.
The Part Most People Get Wrong
Most people treat AI cold email writing as a one-prompt job. They type in a brief, copy the output, hit send, and wonder why their reply rate is 1%.
That's wrong. Here's why: every AI tool has a default voice — and that default voice is tuned for general audiences, not your specific prospect. The output you get in the first pass is the AI's best guess at average. Cold email that converts is never average.
The fix is simple but almost nobody does it: always give the AI a specific real-world detail about the recipient. Instead of "a SaaS founder," say "a SaaS founder who just launched a mobile app and tweeted that their onboarding drop-off rate is killing them." That one sentence of context transforms a generic email into something that feels researched.
The second mistake is not iterating. The first output is a draft — treat it like one. Run a follow-up prompt like: "Make the opening line more specific and cut the word count by 20%." Or: "Replace the CTA with something that feels less salesy and more curious." Three rounds of refinement on a Claude or ChatGPT draft will produce something that a human copywriter would charge $150 to write.
Cold email written by AI isn't the problem. Cold email written by AI and never touched again — that's the problem.
Key Takeaways
- Claude for tone: When your email needs to sound like a real human wrote it — not a template — Claude is your starting point.
- ChatGPT for structure: If you need a conversion-optimized draft fast, ChatGPT's default frameworks for Problem → Solution → CTA are hard to beat.
- Gemini for research, not persuasion: Use Gemini to research your prospect or industry before writing — not to write the email itself.
- The two-step method: Draft with ChatGPT for structure, then humanize with Claude for tone. This combo consistently outperforms a single-tool approach.
- Context is the multiplier: The more specific the real-world detail you give any AI, the better the output — every single time.
What to Do Right Now
Open ChatGPT and paste this prompt: "Write a cold email from [your name/role] to [specific type of prospect]. Goal: book a 30-minute call. They recently [specific thing they did — launched a product, posted a job, won an award]. Under 150 words. Subject line included." Take that output, open Claude, and say: "Rewrite this to sound warmer and more personal — like one real person reaching out to another." You'll have a better cold email in under 10 minutes than most people produce in an hour.