Over the past year, I’ve been turning my cold-outbound playbook into writing instructions for AI.
Why I built Which Draft
I’ve had to revisit those instructions with every new generation of models. Does Astra write better outbound than Sol? How much does the model matter, and how much do the writing instructions matter?
I built Which Draft to test that, using Astra to help write the app. Give models the same prospect and instructions, then compare the messages side by side, including the follow-ups. You can also compare the same model with writing skills on or off.
In Battle, you choose between anonymous messages for the same recipient and sequence step. The leaderboard shows those votes alongside writing evaluations, where you can read the drafts and see why they received a particular judgment.
The work behind the examples
At Colabra, we build AI for M&A due diligence and sell to private equity firms. That’s why the saved examples use public information about investment and acquisition leaders.
Try it on your own writing
Explore the saved examples free, or bring your own API key to try your contacts, instructions and message sequences. You can also take the Personalization and Aoi voice skills and use them elsewhere.
I’d love to hear where the writing falls short, which instructions helped, or what you’d want me to add. Let me know on LinkedIn.
Compare the messages