Provenance marking

Added · Seed

Seeded from a cross-catalogue triage of six AI pattern libraries (references/ai-pattern-catalogues.md); not yet written from close product study.

Content carries a legible mark of what made it — a label on the AI’s turn, a badge on the generated block, metadata in the file — so origin survives the distance between maker and reader.

Forms

Recurring forms: disclosure labels on AI turns and features, visual conventions (colour, iconography) reserved for machine output, watermarks readable by humans or software, and per-span authorship records in mixed documents.

Resources & references

To-do

The unlabelled-becomes-a-claim force is the sharpest and least discussed in the catalogues; C2PA and platform-disclosure policy work would ground it. Whether identity marks (name, avatar, persona) are this pattern or a neighbouring one is open.

Consequences

  • origin is legible at the point of reading — labels, badges, marks, embedded metadata — so the reader's calibration has something to work from
  • a mark only works while it survives the content's travels; copy, paste, export, and screenshot strip exactly the layer the mark lives in
  • labelling the machine's output makes the unlabelled a claim of human origin — the default stops being neutral the moment anything is marked

Related patterns

Complements

  • Citation — who made it versus where it came from

Related

  • Living document — a document that keeps being regenerated blurs what was written and what was made
  • Collaboration — mixed human-machine authorship is where per-contribution origin matters most
  • Annotation — the mark is an annotation about origin rather than content