AI-Driven Job Description Template Generation SOP

Standard Operating Procedure: AI-Driven Job Description Generation & Governance

Document ID: SOP-TR-ENG-089
Effective Date: October 24, 2023
Version: 1.0.0
Review Cadence: Semi-Annual
Classification: Internal Operational Standard

---

1. Executive Summary & Purpose

This Standard Operating Procedure (SOP) defines the institutional requirements for leveraging Artificial Intelligence (AI) to author, optimize, and governance-check job description templates within Template Registry. The objective is to standardize the generation of high-precision, equitable, and legally compliant job descriptions while mitigating risks associated with generative AI, including algorithmic bias, hallucinations, and non-compliance with global employment standards. Adherence to this protocol is mandatory for all People Operations, Talent Acquisition, and Engineering Management personnel.

---

2. Scope & Prerequisites

2.1 Scope

This SOP applies to all new and revised job descriptions generated, updated, or stored within the Template Registry ecosystem across all operating jurisdictions (US, EMEA, APAC).

2.2 Prerequisites & Tooling

• Authorized AI Engine: Enterprise-licensed LLM instance (e.g., GPT-4 Enterprise or Claude 3 Opus via secure API gateway). Direct consumer-tier interfaces are strictly prohibited due to data privacy constraints.
• Template Registry Repository Access: Read/Write permissions to the /templates/hr/job-descriptions/ version-controlled repository.
• Auxiliary Validation Tools: Textio or equivalent Augmented Writing platform for bias and inclusivity scanning; local instance of the EU AI Act compliance checklist.

---

3. Roles & Responsibilities

---

4. Step-by-Step Procedure

Phase 1: Input Parameterization & Prompt Engineering

• 1.1 Extract core competencies, level descriptors, and OKRs from the approved headcount requisition form.
• 1.2 Populate the standardized Template Registry LLM Input Schema (JSON format) with level, department, compensation band, and core technical stacks.
• 1.3 Append the mandatory system-level prompt wrapper: "Act as an expert technical recruiter and legal compliance officer. Generate a job description free of gender-coded language, ageism, and unnecessary degree requirements, adhering strictly to EEOC and EU employment standards."

Phase 2: AI Generation & Initial Sanitization

• 2.1 Execute the prompt via the enterprise-approved API gateway; log the prompt hash and timestamp in the audit trail.
• 2.2 Review the generated output for structural integrity against the Template Registry baseline schema (Summary, Essential Duties, Qualifications, Competencies, Benefits/Perks).
• 2.3 Execute an automated regex sweep to identify and purge generic filler phrases (e.g., "rockstar," "ninja," "fast-paced environment").

Phase 3: Bias Mitigation & Legal Review

• 3.1 Pass the sanitized markdown output through the authorized Augmented Writing validation tool (Textio).
• 3.2 Ensure the Inclusivity Score meets or exceeds the institutional threshold ($\ge 90$).
• 3.3 Verify that essential functions and physical/cognitive demands (if applicable) are accurately represented without discriminatory gating factors.
• 3.4 Submit the draft to the People Ops HRBP for statutory compliance sign-off.

Phase 4: Version Control & Registry Ingestion

• 4.1 Convert the finalized job description to GitHub-flavored Markdown (.md) matching the naming convention: [DEPT]-[ROLE-SLUG]-[LEVEL].md.
• 4.2 Open a pull request (PR) targeting the main branch of the Template Registry repository.
• 4.3 Secure automated linting checks and required peer reviews from both the Hiring Manager and the HRBP.
• 4.4 Merge the PR upon approval, triggering automated synchronization to the Applicant Tracking System (ATS).

---

5. Quality Assurance & Pro-Tips

5.1 Best Practices

• Deterministic Prompting: Always supply explicit leveling frameworks (e.g., Engineering Level E4 vs. E5) within the input schema to prevent the AI from hallucinating inflated scope.
• Human-in-the-Loop Verification: Never publish AI-generated output directly to public boards without human review of the compensation ranges and essential requirements.

5.2 Common Pitfalls

• Data Leakage: Pasting proprietary internal compensation strategies or unannounced organizational restructuring plans into public LLM interfaces. Use enterprise instances with zero-retention policies only.
• Over-Indexing on Buzzwords: Allowing the LLM to pad technical requirements with obsolete toolsets. Cross-reference all technical stacks with the current Template Registry Tech Radar.

5.3 Metric Thresholds

• Inclusivity Index: $\ge 90/100$ on Textio or equivalent evaluation engine.
• Generation Velocity: Average time from requisition input to PR creation $\le 45$ minutes.
• Compliance Defect Rate: $0%$ legal escalations regarding discriminatory language.

---

6. Frequently Asked Questions (FAQ)

Q1: Can I use ChatGPT or Claude via their public web interfaces if I remove company names?
A: No. Public consumer-tier interfaces retain user inputs for model training, posing a severe risk of intellectual property leakage and internal data exposure. All operations must utilize the enterprise-licensed API gateway.

Q2: What should I do if the AI-generated output fails the inclusivity score threshold?
A: Do not manually edit the text blindly. Re-run the generation phase with an updated system prompt that explicitly penalizes the specific exclusionary terms flagged by the validation tool (e.g., "mandatory native fluency" or gender-coded superlatives).

Q3: Who holds ultimate accountability if a published job description contains non-compliant clauses?
A: While the Talent Acquisition specialist executes the generation process (Responsible), the People Operations Business Partner (HRBP) holds accountability (Accountable) for final statutory approval before repository merge.