I tested ColdCraft vs ChatGPT on AI detection. The results are embarrassing for ChatGPT.
I built ColdCraft, an AI cold outreach generator for freelancers and solo founders.
The most common pushback I get is: "Why would I use this when ChatGPT exists?"
Fair question. So I ran the test.
Same input. Same goal. I asked both tools to write a 3 message cold email sequence for a freelancer trying to get clients. Then I ran both outputs through Phrasly, an AI detection tool with 5 third party detectors built in.
Here are the results.
ChatGPT: 100% AI detected. 13 out of 15 sentences flagged. Every single detector failed. Originality.ai, ZeroGPT, Turnitin, Copyleaks, GPTZero. All red.
ColdCraft: 8% AI detected. 0 out of 10 sentences flagged. Every detector passed. All green.
This matters because the person receiving your cold email is not reading it with fresh eyes. Their inbox filters flag AI written emails before they even open them. And the ones that get through, people can feel when a robot wrote it.
Cold outreach only works if it sounds like a real person sent it.
ColdCraft is free to try. 3 sequences a month, no card needed.
What tool are you currently using to write cold outreach?
The interesting challenge is moving from "less detectable AI" to "better outreach outcomes."
What would convince you that AI detection scores are the strongest proof of value, rather than metrics like reply rates, booked calls, or conversions from real campaigns?
Fair challenge. Detection score isn't the end goal, you're right. But it's the entry fee.
An email scoring 100% AI gets filtered or ignored before reply rate even becomes measurable. You can't track conversion on an email nobody opened.
But here's the bigger point. ColdCraft isn't built for sending a thousand emails and hoping something sticks. It's built for the person who has done the research, knows exactly who they're reaching out to and why, and wants the message to land like a human wrote it specifically for that person. The more specific the input, the better the output.
That combination, real research plus human sounding output, is what moves reply rates. Detection score is just proof that the human part is working.
Campaign data will come. Product launched 11 days ago.
Appreciate the context.
I actually reached out via email on your other thread a few days ago — could you check your inbox/spam folder once?
Looking forward to continuing the conversation there.