Editorial guidelines for AI-assisted reporting (without losing trust)

Accountability stays human

Assistants accelerate drafts; editors remain accountable for facts, fairness, and proportionality. Newsroom policies should state plainly where automation assists research formatting versus substantive claims.

Source tracing

Require citations to primary documents or named publications—not model memories. When assistants summarise filings, editors must reopen originals before asserting interpretations.

Disclosure calibrated to impact

Transparency need not mean clumsy boilerplate every paragraph. Disclose materially—especially where audiences might otherwise infer omniscient reporting—while avoiding gimmicky badges that substitute for rigour.

Corrections workflow

AI-introduced errors still deserve prominent corrections when substantive. Version histories should capture meaningful edits without exposing risky internals like leaked prompts.

Training nuance

Using reader submissions or copyrighted texts to train proprietary models raises distinct consent questions—separate from everyday drafting assistance.

Closing thought

Trust accrues from predictable behaviour after mistakes, not from novelty tooling announcements alone.

---

Policies should align with your jurisdiction’s press norms and union agreements.