AI for HR professionals: useful for drafting, careful with people decisions

HR sits between two facts. Much of the work is writing and organizing, which AI does well. And much of the subject matter is people, which is where legal exposure and trust live. Using AI well in HR means being generous about drafting and strict about anything that touches a decision about a person.

Where AI helps HR

Drafting. Job descriptions, offer-letter templates, onboarding checklists, policy first drafts, manager talking points and announcements all start faster from a draft. Give context: audience, tone, what must be included, and what must not. The prompting basics are in how to write better AI prompts.

Making sense of feedback. Engagement surveys and exit interviews produce hundreds of comments. An assistant can group them into themes and pull representative quotes. Spot-check the grouping against the raw comments, because it can overweight vivid ones.

Reporting. Headcount, turnover, time-to-fill and onboarding reports are often assembled by hand each month from an export. A repeatable workflow takes the export and produces the same report every time. See the weekly reports question.

Answering routine questions. A first draft of responses to common employee questions about leave, benefits enrollment or process can save time. Verify against the actual plan documents.

Where AI does not help

• Deciding about a person. Hiring, promotion, pay, discipline and termination decisions should not be delegated to a model.
• Legal conclusions. Employment law varies by state and country, and an assistant may state a rule confidently and wrongly. Treat its output as a prompt for your attorney, not advice.
• Sensitive conversations. Investigations and difficult performance discussions need your judgment and documentation.

Risks specific to HR

Bias and fairness. Models can reflect patterns in their training data. Using AI to screen, rank or score candidates or employees can create disparate impact even if nobody intended it. Some jurisdictions and cities have rules on automated employment decision tools, so check what applies to you and involve counsel before using AI in selection.

Employee privacy. Medical information, leave reasons, performance records, immigration status and pay are highly sensitive. Read is it safe to put company data in ChatGPT. If you must analyze people data, use a tool your company has approved and reviewed, and remove names where the task allows.

Consistency of documents. A policy drafted by AI may not match your actual practice or the law. Have counsel review anything binding.

You often write the AI policy. Employees will ask what they can use and how. Does my company need an AI policy covers the essentials HR is usually asked to lead.

Starter prompts

Job description check:

Here is a draft job description. Identify any wording that could
discourage qualified applicants from applying or that describes
personal characteristics instead of job requirements (for
example age-coded or gendered words). Suggest neutral
alternatives. Separate must-have requirements from nice-to-have.

Survey themes:

Below are anonymous survey comments. Group them into themes.
For each theme, give a count, two verbatim quotes, and what
action the comments imply. List any comments that did not fit.
Do not guess who wrote any comment.

Your first week

• Day 1: Find out which AI tools your company has approved. If none, pause on people data.
• Day 2: Use the job description prompt on one existing posting.
• Day 3: Summarize a set of already-anonymous feedback and check the themes against the source.
• Day 4: Draft one announcement and ask counsel or your lead to review the process, not just the result.
• Day 5: Write down what you will never enter into an AI tool, and share it with your team.

A worked example: rewriting a job posting

A hiring manager sends a posting with requirements like “digital native,” “recent graduate,” “must be able to lift 50 pounds” and fifteen must-haves. Ask the assistant to separate genuine requirements from preferences, flag phrases that could be age-coded or that describe the person rather than the work, and check whether the physical requirement is tied to essential job functions. Then ask for a rewritten version with a shorter must-have list.

You take the result back to the manager and discuss it. Your role does not change: you own the judgment about what is essential and whether the wording meets your legal obligations. The assistant saves the first pass and catches wording you may skim past.

Common mistakes to avoid

• Using AI to rank candidates. That creates fairness and legal exposure. Keep people in the decision.
• Pasting named performance reviews or medical information. Use approved tools and remove identifiers.
• Publishing a policy no lawyer has read.
• Assuming a summary of comments reflects the minority view. Read the outliers.

What the sources say

Occupational data for HR specialists and the one US city rule that sets a bias-audit requirement for automated hiring tools.

• The Bureau of Labor Statistics projects employment of human resources specialists to grow 6 percent from 2025 to 2035, faster than the average for all occupations.
• New York City Local Law 144 bars employers from using an automated employment decision tool unless it has had a bias audit within one year before use.
• New York City Local Law 144 also requires that information about the bias audit be publicly available.
• The New York City notice states that the Department of Consumer and Worker Protection will begin enforcement of Local Law 144 and its rule on July 5, 2023.

Learn it hands-on

Learn AI Fast is a two-day in-person class in Boston for non-developers, with sessions on AI risks, data privacy and company policy, plus a template for an AI use policy to take back to your team.

Sources

• U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Human Resources Specialists: “projected to grow 6 percent from 2025 to 2035, faster than the average for all occupations”
• NYC Department of Consumer and Worker Protection, Automated Employment Decision Tools (AEDT): “unless the tool has been subject to a bias audit within one year of the use of the tool”
• NYC Department of Consumer and Worker Protection, Automated Employment Decision Tools (AEDT): “information about the bias audit is publicly available”
• NYC Department of Consumer and Worker Protection, Automated Employment Decision Tools (AEDT): “DCWP will begin enforcement of this law and rule on July 5, 2023”