To draft a job description with AI, begin with an approved job analysis created by people who understand the work. Give the model verified duties, outcomes, competencies, working conditions, and job-related requirements, then ask it to organize—not invent—the description. HR, accessibility, legal, compensation, and hiring owners should review the exact version before publication.

This is an employer-side workflow. Applicants looking for help should use the resume and cover-letter guide or job-interview preparation guide. The responsible AI guide explains the broader input and approval boundary.

Key Takeaways

Do the job analysis before asking AI to write prose.

Separate essential functions and outcomes from inherited preferences.

Make every qualification traceable to work evidence.

Do not use AI to infer the ideal personality or screen candidates.

Review accessibility, local law, pay, privacy, and publishing details with accountable specialists.

Freeze the role need and review context

Start with a short role-intake record. Name the business need, team, reporting line, location or location policy, employment type, decision owner, target publication channels, and applicable internal policies. State whether this is a new role, a replacement, or a materially changed role. Do not recycle an old description until the current work has been checked.

Also identify the jurisdictions and agreements that govern the hiring process. Requirements for disability accommodation, pay transparency, working time, privacy, union consultation, language, and record retention vary. The article’s U.S. and UK sources provide useful process controls, but they do not create a universal legal checklist.

Create an approval group before drafting. It normally includes the hiring manager, a subject-matter expert, HR or talent acquisition, and any required compensation, accessibility, legal, security, or works-council reviewer. The group—not the AI tool—owns the role definition.

Complete a job analysis with people who know the work

Job analysis identifies the tasks performed and the competencies needed to perform them. The U.S. Office of Personnel Management describes it as a foundation for developing assessments and emphasizes linking competencies to job tasks.［1］ Interview current role holders, supervisors, customers of the role, and adjacent teams where appropriate.

Collect evidence in a structured table:

Field	Evidence question
Role purpose	What outcome would be missing if the role did not exist?
Task	What observable work is performed?
Output	What artifact, service, or decision results?
Frequency	How often is the task performed?
Importance	What happens if it is done poorly or late?
Competency	What knowledge, skill, ability, or behavior enables the task?
Context	Which tools, environments, schedules, travel, or physical conditions apply?
Source	Who supplied or approved this evidence, and when?

Ask for examples and counterexamples. “Own reporting” is vague; “produce the monthly service report, validate the source table, explain material deviations, and obtain operations approval” is observable. Avoid copying current employees’ personal characteristics into the role definition.

Do not put names, health or grievance information, salary records, protected characteristics, identifiable performance records, or investigation evidence into an ordinary consumer AI service. Use only an organization-approved environment and the minimum de-identified fields needed to describe the role. Resumes, interview notes, and past selection decisions are also unnecessary at this stage. Review AI privacy risks before moving any internal material into an approved tool.

What counts as an essential job function rather than a preference?

An essential function describes work the role must perform, not the customary way one person happens to perform it. The EEOC’s employer guidance explains that written descriptions prepared before advertising or interviewing can be evidence of essential functions, while actual work, employer judgment, time spent, consequences, and available staffing also matter.［2］ Local law may use different terminology or tests.

For each proposed essential function, record:

the outcome or responsibility;
why it is essential;
evidence from the job analysis;
frequency, consequence, or operational dependency;
whether the result can be achieved in another reasonable way;
the reviewer who approved the classification.

Do not turn a convenient method into an absolute requirement. “Communicate decisions to distributed stakeholders” describes an outcome; “must communicate verbally in an open office” may unnecessarily prescribe a method. Ask accessibility and HR reviewers to examine wording about mobility, lifting, sight, hearing, speech, schedule, location, and travel.

Separate required from preferred qualifications. A required qualification should be demonstrably necessary at entry or required by a valid rule. A preferred qualification may improve ramp-up but must not become a hidden automatic screen. Degree, years-of-experience, credential, location, and schedule requirements deserve evidence because inflated proxies can exclude capable applicants.

Build an approved job description fact table

Before drafting prose, create one row for every claim the description may contain:

Section	Approved fact	Evidence ID	Required or preferred	Owner	Review status
Role purpose	`[approved outcome]`	JA-01	Required	Hiring manager	Approved
Essential function	`[observable responsibility]`	JA-07	Required	SME	Approved
Competency	`[job-linked competency]`	JA-12	Required	HR	Approved
Work context	`[location or condition]`	JA-18	Not applicable	Operations	Approved

Include role purpose, outcomes, essential functions, competencies, required and preferred qualifications, reporting line, location policy, work context, accessibility or accommodation route, version, and approver. Keep pay range, benefits, application deadline, sponsorship, background checks, and employment terms in fields owned by the relevant team; do not let the model infer them.

Use fact-checking practices for claims about credentials, regulations, tools, or organization policy. If a field has no approved evidence, leave it out or mark it for owner resolution outside the publishable draft.

How do you draft a job description with AI from approved facts?

Provide the fact table, style guide, channel limits, and a strict instruction:

Draft a job description using only the approved rows. Preserve the required/preferred distinction and evidence IDs in a separate claim ledger. Use observable outcomes and plain language. Do not add qualifications, years of experience, personality traits, working conditions, benefits, legal claims, or screening criteria. Do not rank or evaluate people. Mark missing sections as “owner input required.”

Request two outputs. First, a claim ledger mapping each sentence to an evidence ID. Second, the candidate description. The ledger makes it easier to catch attractive but unsupported additions such as “fast-paced,” “rock star,” “native speaker,” or “must thrive under pressure.”

A practical order is:

role title and one-sentence purpose;
three to six outcome-oriented responsibilities;
essential functions and work context;
job-linked required competencies and qualifications;
genuinely preferred qualifications;
reporting line, location, schedule, and application process;
accessibility or accommodation route and required employer statements.

Avoid overloading the opening with culture claims. Candidates need to understand the work, decision scope, expectations, and conditions before promotional language.

How do you review for exclusion, automation, and accessibility risks?

Run a human line review. For each requirement, ask: What task supports it? Is it needed on day one? Could another credential or experience demonstrate the same competency? Does the wording prescribe a method instead of an outcome? Could it screen out a protected group without a job-related reason?

The EEOC’s AI and ADA materials warn that automated employment technologies can create disability discrimination risks, including screening out qualified people or failing to support accommodation.［3］ This drafting workflow must not become an automated-selection workflow. Do not ask AI to score the description against applicant profiles, infer disability, rank candidates, or generate “culture fit” traits.

Check language for coded or subjective claims. Replace “young and energetic” with an actual work outcome, “digital native” with a specific competency, and “excellent communication” with the audiences and deliverables involved. Do not promise that a word-scanning tool can prove a description is unbiased or legally compliant.

NIST’s Generative AI Profile treats privacy, confabulation, harmful bias, information integrity, and human oversight as lifecycle risks.［4］ Use approved tools, minimize inputs, preserve evidence, and record who approved the final text.

Approve, publish, and maintain one version

Before publication, the hiring manager verifies role accuracy; the SME verifies tasks and competencies; HR checks selection consistency and policy; compensation checks pay information; accessibility and legal reviewers check their required areas; and the publishing owner verifies channel fields and dates.

Use this release record:

Field	Value
Description version	Stable version ID
Job-analysis version	Source evidence version
Hiring manager	Name and approval date
HR reviewer	Name and approval date
Specialist review	Accessibility, legal, compensation, or other required review
Channels	Approved publication destinations
Review trigger	Role change, policy change, recurring review date, or hiring feedback

Publish the same approved core description across channels. Channel-specific shortening should preserve essential functions, qualification distinctions, location, accessibility route, and employment terms. If a channel requires a different field, route that change back to the owner rather than asking AI to improvise.

After hiring, collect process feedback without importing applicant profiles into the description prompt. Re-run the job analysis when duties, tools, reporting lines, location, or legal requirements materially change. Do not silently update a live description while interviews are underway; version the change and decide how current applicants will be treated.

Summary

Base the draft on current job analysis, not an inherited description.
Tie duties and qualifications to observable work evidence.
Separate essential, required, and preferred fields.
Use AI only to organize approved facts and surface gaps.
Keep applicant and sensitive employment data out of the prompt.
Require named HR, role, accessibility, and legal review as applicable.

FAQ

Can AI write a job description from a job title alone?

It can produce generic prose, but that prose is not evidence of your actual job. Complete a current job analysis and provide approved facts before drafting.

Should I paste an old job description into AI?

Only after owners have checked which parts remain accurate and removed personal, sensitive, outdated, or unsupported content. Treat the old description as one input, not the truth.

Can AI decide which qualifications are required?

No. SMEs, the hiring manager, HR, and required specialists must establish job-related requirements. AI can flag missing evidence or inconsistent labels.

What is the difference between an essential function and a qualification?

An essential function is work the role must perform. A qualification is evidence that a person can perform or learn that work; the relationship should be documented rather than assumed.

Can AI check whether wording is legally compliant?

No general model can certify compliance across jurisdictions. Use it for drafting questions and consistency checks, then obtain the review required by local law and policy.

Should the description include physical requirements?

Include only accurate, job-related conditions and outcomes reviewed under applicable rules. Avoid describing one customary method when the essential result can be achieved differently.

Can I use AI to compare the description with applicants?

That is a separate, high-risk selection use. This workflow does not authorize automated scoring, ranking, disability inference, personality inference, or candidate screening.

How often should a job description be reviewed?

Review it when the work, tools, reporting line, location, policy, or legal context changes and on the organization’s recurring schedule. Preserve versions when a hiring process is active.

Disclaimer: This article provides general information, not employment, disability, compensation, privacy, or legal advice. Requirements differ by jurisdiction, policy, and collective agreement; obtain qualified review.

Sources:

U.S. Office of Personnel Management, Job Analysis — https://www.opm.gov/policy-data-oversight/assessment-and-selection/job-analysis/
U.S. Equal Employment Opportunity Commission, The ADA: Your Responsibilities as an Employer — https://www.eeoc.gov/publications/ada-your-responsibilities-employer
U.S. Equal Employment Opportunity Commission, Artificial Intelligence and the ADA — https://www.eeoc.gov/eeoc-disability-related-resources/artificial-intelligence-and-ada
NIST, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile — https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence

Sources checked 29 August 2026.

Related reading:

How to Draft a Job Description with AI