Show-your-work provenance

Show-your-work provenance is the practice of surfacing, on every AI-generated draft, the inputs that produced it — which voice fingerprint was referenced, which brief, which past posts inspired the rhythm, what intent the strategy expected.

Most AI writing tools present drafts as opaque outputs: the user sees the text but not the reasoning. Provenance turns the output legible: this draft used voice fingerprint version X, was generated against brief Y, references past post Z for rhythm, and was meant to fulfill intent W. The user can then sanity-check the inputs rather than only judging the output.

Implementation is largely a metadata problem: capturing the inputs at generation time and persisting them so the UI can display them. The hard part is choosing what to surface — too much provenance buries the draft in metadata; too little makes the output feel like a black box. The right balance is one collapsible panel with the four or five inputs that actually matter.

AI workflows that don’t show their work require pure trust. AI workflows that do invite real review — which surfaces problems early and builds confidence over time.