Prepare an AI writing workflow for a new user by defining the permitted writing tasks, approved accounts and sources, prohibited data, input template, expected output, claim-verification method, human review duties, and release authority. Give the user a practice assignment with non-sensitive material and require them to distinguish an AI draft from reviewed content. Do not provide publication access until they can identify unsupported claims and stop at the approval gate.
Define the job and its boundaries
List the specific tasks the workflow supports, such as outlining, summarizing supplied material, drafting low-risk copy, or formatting reviewed notes. Name the requester, operator, subject reviewer, editor, approver, and incident contact. State which data may not be entered, including secrets, private records, customer information, or copyrighted material without authorized use.
NIST's AI Risk Management Framework says organizations should define roles and responsibilities for human-AI configurations, document the tasks the system supports, and describe knowledge limits and human oversight (NIST AI RMF Core). Translate those expectations into a short procedure for the actual team.
Provide an evidence-first template
Give the new user an input form containing audience, purpose, required facts, source URLs, extracted evidence, tone, length, exclusions, and output format. Require every factual claim to point to a source that actually supports it. Separate source notes from generated prose so the reviewer can trace the draft.
Record the visible provider, product, model or configuration label, prompt version, date, and any retrieval or editing tools. Outputs can vary, so an old successful run is not a substitute for reviewing the current draft.
Teach review and risk checks
Use a practice draft containing at least one unsupported claim, stale date, ambiguous instruction, and disclosure requirement. Ask the learner to find and correct the problems without inventing evidence. The NIST Generative AI Profile offers suggested actions for governing, mapping, measuring, and managing risks specific to generative AI (NIST Generative AI Profile).
The checklist should cover factual support, missing context, misleading certainty, unsafe instructions, private information, copied phrasing, links, arithmetic, and house style. Define state labels such as draft, fact-checking, editorial review, approved, scheduled, and published.
Prove independent readiness
Have the new user complete one supervised draft and one independent draft from approved sources. Verify that they preserve evidence, use the required disclosure, escalate a risky request, and stop before release. Review their record of inputs and changes rather than judging only the final prose.
The workflow is prepared for a new user when access is least-privilege, sources are traceable, prohibited inputs are understood, review duties are explicit, and only the named approver can release work. If the user cannot reproduce the evidence trail, keep the status onboarding incomplete.
Sources
- AI RMF Core — National Institute of Standards and Technology
- Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile — National Institute of Standards and Technology, July 26, 2024