AI can write your first draft. It should not send it
What AI is genuinely good at in outreach, where it reliably fails, and why every draft here waits for a person.
AI drafting is genuinely useful for one specific job: getting past the blank page. It is bad at the job people keep trying to give it, which is representing you to a stranger without supervision.
Where it helps
- Turning a thread you have already read into a reply you can edit, in seconds.
- Matching the tone of the conversation, if you tell it what that tone is.
- Removing the stiffness that makes a written reply read like a template.
Where it fails
It invents specifics. A model asked to be personable will happily reference a project that does not exist, a mutual connection you have never met, or a detail from the wrong person's profile. In a sales conversation that is not a small error — it is an obvious tell that nobody real read the message, and it costs you the prospect permanently.
Our prompt explicitly forbids inventing facts, meetings, prices and links, and forbids placeholders. That reduces the failure rate. It does not eliminate it. No prompt eliminates it.
So every draft waits
Drafts are filed as pending review. Nothing is sent until a person reads it and approves it in the inbox. This is a deliberate design choice, not a missing feature — the value of AI here is speed of editing, and that value survives the review step. What does not survive is sending thirty unreviewed messages that each contain a small invention.
The tone presets and the sender profile in Settings exist to make the draft closer to right on the first pass, so the review is a quick edit rather than a rewrite. That is the honest promise: less typing, not less judgement.