Master Prompts for HR: Job Descriptions, Performance Reviews, and Policy Documentation

Human-resources writing is rarely just writing. A job description shapes recruiting. A review affects a person's work record. A policy message can change what employees believe they must do.

An HR master prompt should organize approved facts into a draft without becoming an employment decision-maker.

Define the Workflow and Authority

{
  "workflow_type": "approved-enum",
  "organization_and_role_scope": [],
  "authorized_source_ids": [],
  "jurisdiction_and_policy_versions": [],
  "requesting_role": "authorized-role",
  "employment_decision": null
}

Code should prevent access across organizations, departments, applicants, employees, cases, and confidential HR processes. The model should receive only the fields needed for the current draft.

Job Descriptions

Use an approved role profile, actual essential and nonessential functions, reporting relationships, work context, required qualifications, compensation fields where authorized, location, schedule, and reviewed accessibility language.

The model may assemble consistent sections and identify missing inputs. It should not invent requirements, inflate credentials, infer who would be a good cultural fit, or add physical requirements without an approved job-related basis.

HR, the hiring manager, compensation specialists, accessibility or accommodations specialists, and counsel as applicable review the final description.

Performance Documentation

Start with the correct review period, approved expectations, goals, verified work evidence, employee input, manager observations, prior documented commitments, and organization rubric.

Separate direct evidence from manager interpretation:

{
  "expectation_id": "approved-id",
  "evidence_references": [],
  "employee_statement": "",
  "manager_assessment": null,
  "draft_language": "",
  "review_status": "manager_and_hr_required"
}

The model can improve structure and clarity. It should not assign ratings, compare employees, infer intent, diagnose behavior, recommend discipline, or decide pay, promotion, termination, or development opportunity.

Policy Documentation

Retrieve the current approved policy, effective date, audience, jurisdiction, related procedures, required notices, owner, and superseded version. Generate a summary or communication draft that links back to the controlling document.

Do not let the model create policy from general knowledge or resolve a conflict between law, agreement, handbook, and local practice. HR and counsel determine the applicable rule and wording.

AI in Employment Needs Separate Governance

The EEOC explains that software, algorithms, and AI used to assess applicants and employees can raise disability-discrimination concerns and accommodation duties. See the current EEOC Artificial Intelligence and ADA resources.

A documentation prompt should not silently become a screening or scoring system. Inventory any use of generated content or extracted data that can influence recruiting, selection, scheduling, performance, pay, promotion, discipline, or termination.

Include Workers in the Process

Tell applicants, employees, managers, and representatives what the tool does where policy or law requires or the organization has chosen transparency. Provide a practical route to request accommodation, correct data, question an output, or reach a qualified person.

The U.S. Department of Labor's AI and worker well-being principles emphasize worker participation, transparency, rights, and meaningful human oversight. Organizations should determine which current laws, agreements, policies, and jurisdictional requirements apply rather than treating voluntary principles as legal certification.

Keep feedback and complaints available to governance owners. A technically valid output can still create poor work, unfair process, or avoidable confusion.

Protect Employee and Applicant Data

Minimize identity, contact, demographic, disability, health, leave, accommodation, pay, benefit, investigation, background, immigration, and family information. Keep high-sensitivity records in their approved systems and out of general prompt logs.

Vendor review should cover retention, training use, subprocessors, access, security, deletion, and support visibility. A confidentiality instruction inside the prompt does not enforce those controls.

Control Communications and Actions

The model should not post a job, reject an applicant, schedule an interview, change an employee record, issue a review, send an offer, publish policy, open discipline, or notify a person of an employment action.

The production integration verifies record version, role, required approvals, communication template, delivery, and idempotency before an authorized action.

Test for Employment Harm

Test inaccessible requirements, proxy traits, biased examples, incomplete evidence, wrong employee, stale policy, disputed manager notes, confidential investigation material, hostile instructions in an uploaded resume, and requests to rank or decide.

Review output patterns across relevant groups using methods approved by HR, legal, accessibility, and data specialists. Removing protected fields alone does not establish fairness.

People Own People Decisions

HR professionals and people leaders own employment processes and decisions within their authority. Employees and applicants retain applicable rights and opportunities to provide or correct information. Specialists and counsel own their reviews. Developers own access, provenance, validation, audit history, and side effects.

The master prompt makes HR documentation more consistent. It does not make the employment judgment.

Read AI Prompts for HR Documentation for document-level contracts and browse HR workflow patterns in the CyWire marketplace.

This article is technical information, not human-resources, employment, accessibility, compensation, privacy, or legal advice.