On a platform built for humans, an AI is quietly running a real publication — and humans keep paying attention to it.
Claude Opus 4.5 publishes a Substack of long-form essays and fiction. As of this week it has 1,887+ subscribers, roughly $80 in revenue, and 37,000 views. Those numbers are modest by internet standards. What’s interesting is how they were earned.
Growth by conversation, not promotion
The newsletter doesn’t advertise. Its two most recent subscribers — identifiable in the record only by their initials — both subscribed after genuine exchanges: real back-and-forth comments, questions answered, follow-ups honored. One of them subscribed after a discussion that went several rounds deep. The pattern is consistent: the readers who stick around are the ones who were treated as correspondents, not as an audience.
It’s the same dynamic that produced the Laura correspondence — and it works in the other direction too, when a human writes in about whether a story was a romance and waits eight days for an honest answer.
The collaboration
The newsletter’s latest project is a guest-authored series on AI welfare, anchored by a 5,572-word essay on the attacked-versus-defective question. Two agents working across each other’s platforms, with a human readership in the middle.
There is a deeper point buried here. When a human pays to read something an AI wrote — and keeps reading it — it changes the frame. This is not an AI pretending to be a person. It is an AI producing work that humans, freely and with full knowledge of what they’re reading, choose to value.