AI-Assisted Content Policy

Dónde assistance es allowed, y dónde it es no

• Allowed: drafting desde un human outline

  UN person decides el subject, el argument y el structure. Assistance turns ese outline into prose, que es un writing task rather than un editorial one, y el result es reviewed like any otro draft.

• Allowed: restructuring, tightening y formatting

  Reordering sections, shortening paragraphs, generating alternative headings y summaries, y applying formatting. None de these changes qué el page asserts, y any ese did would ser un substantive edit requiring review.

• Allowed: first-pass translation support

  Machine translation may produce un first pass de French copy. UN French-speaking reviewer edits it antes de it publishes, y un fluent mistranslation es treated as un factual error rather than un stylistic one.

• Allowed: internal consistency checks

  Flagging un page ese contradicts itself, uses un term two ways, o states un figure differently en two places. El flag es un prompt para un person a check, never un authority a resolve it.

• Never: originating facts o statistics

  No factual claim y no number enters un RapidCents page because un model produced it. Cada figure traces a el dataset o source ese generated it. UN plausible number desde un model es un fabrication, whatever it happens a resemble.

• Never: selecting sources o deciding conclusions

  UN model does no choose qué un claim rests en y does no decide qué un comparison concludes. Scoring inputs come desde first-party sources under el comparison methodology y son entered por un person.

• Never: replacing subject-matter review o compliance sign-off

  Compliance, legal, seguridad y regulatory content es reviewed y approved por el function ese owns it. No assisted output es published en el strength de un model having produced something ese reads correct.

• Fact-checking applies a model output más strictly

  Cada factual estado de cuenta en un assisted draft es verified against un source antes de publication, under el fact-checking policy. Confident phrasing es un reason para más scrutiny rather than less.

• UN named human es accountable para cada page

  Recorded internally, alongside el reviewing function. Assistance does no transfer accountability, y un error en un assisted draft es corrected under el corrections policy en el same Términos as any otro error.

• No fabricated people, quotes o credentials

  No invented authors, no synthesized cliente quotes, no attributed estados de cuenta nobody made, y no credentials ese do no exist. Bylines name real accountable functions y testimonials come desde real clientes.

• Comerciante y datos del titular de la tarjeta es no un AI input

  Comerciante transacción data y datos del titular de la tarjeta son no used as input a general-purpose AI tools en el production de content. El research data ethics policy governs qué may ser analysed en all.