ChatGPT vs Claude vs Gemini: Who Writes the Best Cold Email?
I sent the exact same cold email brief to ChatGPT, Claude, and Gemini — and the gap between the best and worst output was bigger than I expected. We're not talking about minor stylistic differences. One AI wrote something I'd actually send to a real prospect. The other two needed serious surgery before they'd be usable. If you're using AI to write cold emails without knowing which tool actually performs, you're either leaving deals on the table or sending emails that quietly kill your reputation. Here's exactly what happened — and what it means for your outreach strategy.
The Brief I Gave All Three AIs (And Why It Was a Fair Fight)
The test had to be fair. Same brief, same constraints, zero extra coaching — just the raw output each AI produces when a real marketer sits down and types fast.
Here's the exact brief I used: "Write a cold email from a B2B SaaS founder to a Head of Marketing at a mid-size e-commerce brand. The goal is to book a 20-minute demo of our email automation tool. Keep it under 150 words. Sound human, not salesy."
Every tool got that prompt verbatim. No follow-up prompts, no role instructions, no system messages. Just the brief — which is honestly how most people use these tools.
ChatGPT (GPT-4o) responded in about three seconds. Claude 3.5 Sonnet took a beat longer but came back with noticeably more structure. Gemini 1.5 Pro was the quickest and had the most confident opening line — which turned out to be a double-edged sword.
The outputs weren't just different in tone. They were different in strategy. And that's the gap most marketers completely miss.
What Each AI Actually Wrote — The Honest Breakdown
ChatGPT's email was clean, competent, and completely forgettable. It opened with "Hi [Name], I know your time is valuable..." — which is the cold email equivalent of starting a joke with "So, funny story." The structure was solid: one-line hook, brief value prop, soft CTA. But it felt like it was written by someone who read a copywriting guide rather than someone who actually sends cold emails. It was safe. Safe doesn't book demos.
Claude's email surprised me. It opened with a specific pain point — "Most e-commerce brands using Klaviyo hit a ceiling around 8-10 automated flows and then stall." — before the founder even introduced themselves. That's a real copywriting technique called leading with the problem, and Claude did it without being asked. The email then tied the tool directly to that specific frustration, and the CTA was sharper: "Worth a 20-minute look?" Four words. No pressure.
Gemini's email had the boldest opening line of the three — genuinely punchy — but then it shifted into feature-listing mode. By sentence three, it was talking about "robust segmentation capabilities" and "seamless integration with your existing stack." That's not a cold email. That's a product page crammed into 130 words.
Here's what this tells you: ChatGPT optimises for correctness. Gemini optimises for confidence. Claude optimises for persuasion. And in cold email, persuasion is the only metric that matters.
The winner of round one was Claude — not because it's a better AI overall, but because its default output happened to align with how high-converting cold emails actually work.
The Hidden Variable That Changed Everything — Prompting Depth
Here's the part most AI comparison articles skip entirely: the raw output test only tells half the story.
The real question isn't "which AI writes the best cold email by default?" It's "which AI writes the best cold email when you actually know how to prompt it?" Because the gap between a mediocre output and a great one is almost always in the prompt — not the model.
I ran a second round. This time I added a persona layer and a specificity constraint to the same brief: "Write a cold email from a B2B SaaS founder to a Head of Marketing at a mid-size e-commerce brand. The goal is to book a 20-minute demo of our email automation tool. Keep it under 150 words. Sound human, not salesy. The founder's name is Marcus. The prospect's company recently ran a BFCM campaign and probably exhausted their flows. Reference that context without being creepy about it."
ChatGPT's output jumped significantly. It wove in the BFCM context naturally: "Post-BFCM is usually when the gaps in automation become obvious..." — and it felt like a real person who pays attention. The email became something I'd actually open.
Claude stayed excellent but leaned slightly more formal with the added detail — which is worth knowing. Give Claude too much context and it sometimes becomes a bit... thorough. You may need to explicitly say "keep the tone casual and punchy" to stop it from slipping into professional-email mode.
Gemini improved the most in relative terms. The extra specificity killed the feature-listing instinct and forced it to anchor on a real situation. The result was sharper and more human than its first attempt.
The mental model here: Think of these AIs like three different sales reps. ChatGPT is coachable and consistent. Claude has strong instincts. Gemini needs a clear target or it'll improvise in the wrong direction. Once you know that, you stop blaming the tool and start writing better briefs.
How to Use This to Write Better Cold Emails Starting Today
Stop using AI as a vending machine. You don't put in a brief and accept whatever comes out. You treat it like a first draft from a junior copywriter — which means you need a review checklist before anything gets sent.
Step one: Start with Claude for your first draft. Use this prompt: "Write a cold email from [your role] to [specific job title] at [type of company]. Goal: book a [X]-minute call about [your offer]. Under 150 words. Lead with a specific pain point they likely have right now, not a compliment or a 'I know you're busy' opener. End with a single, low-pressure CTA."
Step two: Run the output through this gut-check. Does the first sentence make the reader think "how did they know that?" If yes, keep it. If it's about you, your company, or your product — delete it and try again.
Step three: Take that Claude draft into ChatGPT for variations. Paste it in and say: "Give me three alternative opening lines for this cold email that are punchier and more specific. Keep everything else the same." ChatGPT is excellent at rapid variation. Use it for that — not for generating from scratch.
Step four: Test subject lines with Gemini. Type: "Here's a cold email. Give me 10 subject lines. Make them curiosity-driven, under 8 words each, and avoid spam trigger words." Gemini's speed and confidence shine here — and subject lines are the one place where bold and direct actually wins.
The whole workflow takes under 15 minutes once you've done it twice. That's a complete, tested cold email — built with three AI tools, each doing what it's actually good at.
The Part Most People Get Wrong
Most people treat cold email AI output as a finished product. They read it, think "that's pretty good," and hit send. That's the mistake. The AI doesn't know your prospect. It doesn't know your conversion data. It doesn't know which opening lines your specific audience ignores.
The email AI writes is a hypothesis. Your job is to turn it into a result. That means you need to actually read it like a skeptical stranger — not like someone who's relieved they don't have to write it themselves.
The second most common mistake is using the same AI for everything. People find one tool they like and never leave. But as you saw above, Claude, ChatGPT, and Gemini each have a specific strength in the cold email workflow. Ignoring two of them because you have a favourite is like only using one club in golf.
Finally — and this one stings — most people never test. They write one version with AI, maybe edit it slightly, and launch it as their whole sequence. Then they wonder why their reply rate is 2%. Write three AI-assisted variants. Test the subject lines. Test the opening line. Give each version at least 50 sends before you judge it. That's not extra work — that's how email actually works.
Key Takeaways
- Claude 3.5 Sonnet: The strongest default cold email writer — it leads with problems, not features, without needing heavy prompting.
- ChatGPT (GPT-4o): The most coachable and consistent — best for generating variations and improving specific sections of an existing draft.
- Gemini 1.5 Pro: Needs more specific briefs than the others, but excels at punchy subject lines and rapid ideation when given a clear target.
- Prompting depth: The quality gap between AI tools shrinks dramatically when you add persona, context, and specific constraints to your brief.
- The three-tool workflow: Use Claude to draft, ChatGPT to vary, and Gemini to generate subject lines — each tool doing the thing it's actually built for.
What to Do Right Now
Open Claude right now and paste this exact prompt: "Write a cold email from [your role] to [specific job title] at [type of company]. Goal: book a [X]-minute call. Under 150 words. Lead with a pain point they likely have this month — not a compliment. End with a single low-pressure CTA." Fill in your details, read the output like a skeptical stranger, and run it through the four-step workflow above. You'll have a better cold email in the next 10 minutes than most people send in a month.