Prompting Standards / Approved AI Use Case Guide
Provide a static guide for approved AI support patterns, source control, human review, automation boundaries, and evidence expectations.
Purpose
Provide a static guide for approved AI support patterns, source control, human review, automation boundaries, and evidence expectations.
When to use
Use before drafting prompts, approving AI-assisted support patterns, or deciding whether a workflow should use AI support at all.
Who should use this
Operations lead, HRB designee, AI governance owner, human reviewer, advisor, or documentation owner.
Review boundary
AI support should not make legal, tax, Medicare, COBRA, eligibility, coverage, reimbursement approval, or compliance determinations. Human review and source control remain required.
Resource sections
Approved AI support patterns
- Use AI for drafting, summarizing, classification support, checklist support, and plain-language internal preparation where approved.
- Require source control and human review before use.
- Keep sensitive or regulated topics within assigned boundaries.
Not-approved AI use patterns
- Do not use AI to make legal, tax, Medicare, COBRA, eligibility, coverage, reimbursement approval, or compliance determinations.
- Do not let AI send employee-facing communications without human approval.
- Do not store prompt history or run prompts in this static guide.
Source control
- Identify source materials, dates, versions, and approved references.
- Do not rely on unsourced AI output.
- Retain source/reference notes.
Human review requirements
- Name the reviewer and review checkpoint.
- Define when output may be used internally or externally.
- Route sensitive topics to qualified review.
Sensitive topic boundaries
- Flag legal, tax, Medicare, COBRA, HIPAA/privacy, eligibility, subsidy, coverage, payment, reimbursement, and compliance topics.
- Define safe language and handoff path.
- Retain review notes.
Automation boundary
- Define what AI may do and must not do.
- Use the automation boundary review for semi-autonomous or persistent support.
- Do not configure live automation here.
Stopping conditions
- Define when AI support must stop, such as uncertainty, high-risk topic, missing facts, employee-facing decision, or review trigger.
- Name escalation owner.
- Retain stopping-condition evidence.
Evidence expectations
- Retain approved use case note, source/reference note, human review note, automation boundary record, and evidence location.
- Connect to evidence capture.
- Do not upload files or store prompt history.
Escalation path
- Name escalation owner for uncertain, high-risk, or out-of-bound AI use.
- Connect to human review checkpoint or escalation protocol.
- Retain handoff note.
Evidence to retain
- Approved use case note - Retain with the audit file, evidence map, or review record as applicable.
- Source/reference note - Retain with the audit file, evidence map, or review record as applicable.
- Human review note - Retain with the audit file, evidence map, or review record as applicable.
- Automation boundary record - Retain with the audit file, evidence map, or review record as applicable.
- Evidence location - Retain with the audit file, evidence map, or review record as applicable.
- Escalation note - Retain with the audit file, evidence map, or review record as applicable.
Evidence is not saved in this static version. Assign an owner, repeat the process consistently, review exceptions, explain the rationale, and retain copies in the employer file or local project record.
Print, Copy, or Export This Resource
No database records or server-side files are created. Use these browser-only actions to retain a local copy.
AI support should not make legal, tax, Medicare, COBRA, eligibility, coverage, reimbursement approval, or compliance determinations. Human review and source control remain required.