How these are made

Deep Field episodes are produced with AI assistance. We think the honest way to say that is plainly, at the top — and then to show our work.

Every episode is built from published histories — books, archives, contemporary accounts — rather than from a model’s memory. We read the historians first and write from them, and each episode lists the main works it draws on. Where the histories don’t support a claim, it doesn’t go in.

“We write from the histories, not from memory — and we’d rather show a disagreement than fake certainty.”

AI helps with the parts it’s genuinely good at: gathering and organising the reading, drafting from that material, and narrating the final script. It works within the material we give it, and its drafts are checked against the books.

Editors choose the subjects, do the reading, check the draft against the histories, and sign off before anything publishes. Where historians disagree, we say so rather than flatten it into false certainty.

We get things wrong sometimes. When we do, we correct the transcript, note the change, and update the audio if it matters. Corrections are logged in the open. Write to us →