AI for HR in 2027: What to Automate in Each Area, What It Costs and Where the Legal Risk Is

Ask any HR generalist at a 200-person company what they did last Tuesday. Rewrote a job posting that pulled zero qualified applicants. Answered the same PTO question four times on Slack. Read 140 applications for a warehouse supervisor role. Drafted a performance improvement plan the manager was supposed to write. Sat through a benefits renewal call. Tried, around 6 p.m., to finish the onboarding checklist for Monday's three new hires.

HR is the department that writes the most inside a company and has the least time to write well. That is exactly where generative AI earns its place. It is also the department where careless use turns into a charge of discrimination: a screening tool that quietly filters out older applicants, candidate data pasted into a personal chatbot account, a rejection nobody can explain. This guide covers AI for HR (or AI for human resources, if your search was the long version) area by area: what to automate, what I would use, what it costs and where the legal risk sits.

Two topics get their own posts: the resume screening agent for recruiters and the AI prompts for HR, ready to copy.

AI for HR: the map, area by area

I grouped the work the way a small HR team actually splits it. "Risk" here means legal and reputational risk, not the chance the AI writes something clumsy (that happens everywhere and you fix it by reading).

The rule I take from this table: AI writes, sorts and summarizes; a person decides about people. Anything that is "text and volume" can go to the machine. Anything that is "who gets hired, promoted or let go" stays with you, with AI at most organizing the information you look at.

Job descriptions: where the value (and the risk) shows up first

Here is a posting in a format that still shows up on Indeed every day. I asked the AI to rewrite it and flag problems first.

Original posting: "Seeking energetic digital native for Payroll Coordinator. Recent grads preferred, 2-5 years max experience. Must be able to lift 50 lbs. Clean-cut appearance. Salary DOE."

Rewrite this job posting for LinkedIn and our careers page.
First, list any phrase that could discourage or screen out applicants based on age, disability, sex, race, religion or national origin, and explain why in one line each.
Then deliver: job title, 2-line summary, up to 6 responsibilities, up to 4 true requirements (only what is actually required to do the job), nice-to-haves, benefits, work model and pay range.
Real details (use only these): Payroll Coordinator, on-site in Columbus, Ohio, Monday to Friday 8 to 5, US$ 52,000 to US$ 60,000, medical/dental/vision, 401(k) with 4% match, 15 PTO days, 340-employee logistics company, payroll in ADP Workforce Now.
Original posting: [PASTE]

Real output from that prompt (opening section):

Problem phrases: "Digital native," "recent grads preferred" and "2-5 years max experience" signal a preference for younger workers and can discourage applicants over 40, who are protected by the ADEA. "Must be able to lift 50 lbs" may screen out people with disabilities if lifting is not an essential function of a payroll role; keep it only if it is, and say how often. "Clean-cut appearance" is subjective and has been tied to race and religion discrimination claims (for example, hairstyles or head coverings). "Salary DOE" omits the pay range, which several states and cities now require in postings.

Payroll Coordinator (On-site, Columbus, OH)
Run accurate biweekly payroll for 340 employees in ADP Workforce Now and be the person employees trust with pay questions.

Two things to notice. The AI caught the age signals and the lifting requirement, which is the job you need from it. And it never invented a benefit, because the prompt gave it the real ones and said "use only these." Without that line, it will happily add "unlimited PTO" or "tuition reimbursement" you do not offer.

Pay transparency is the other trap. Colorado, California, Washington, New York, Illinois and others require a pay range in postings, with different rules on who is covered. AI does not know which ones apply to your remote role. Check your states yourself.

Recruiting: screening, interviews and rejection emails

Resume screening: the biggest gain and the biggest risk

A role with 300 applications costs a recruiter about 15 to 20 hours of real reading. With AI, it can drop to two. The problem: the most common way people do it ("score this candidate from 0 to 100") is the worst way. A score with no explanation cannot be audited, changes when you run it again and rewards whoever writes a polished resume.

The setup I trust separates two steps: the AI only extracts evidence for criteria you wrote before opening the role (yes, no or not stated, with the quote from the resume), and a fixed spreadsheet rule does the scoring. The full build, with prompts, cost and the human review step, is in the post about the AI screening agent for recruiters.

Structured interviews: the cheapest way to cut bias

Structured interviews (same questions, same rating scale, for every candidate) predict job performance better than casual conversations, and they make your hiring easier to defend. The reason most teams skip them is time: nobody wants to build a guide for every role. AI builds one in two minutes. You review it in ten.

Build a 45-minute structured interview guide for [JOB TITLE] assessing these competencies: [3 TO 4 COMPETENCIES].
For each competency: 2 behavioral questions (STAR format), 1 follow-up probe, what a strong answer sounds like, what a weak answer sounds like, and a 1-to-4 rating anchor.
Add a 5-minute opening and a closing for the candidate's questions. No questions about age, family, health, religion, national origin or anything not tied to the job.

Rejection emails people do not hate

Most candidates never hear back. A short, specific, respectful rejection takes the AI ten seconds and protects your employer brand on Glassdoor and LinkedIn. Keep the reason general ("we moved forward with candidates whose experience was closer to X") and never paste the interview notes into the email draft.

Onboarding and employee questions

A 30-60-90 day plan for a new hire is a perfect AI task: structured, repetitive, and different enough per role that templates get stale. Give the AI the role, the manager's real priorities and the tools the person will use, and ask for weekly goals plus who they should meet.

Employee questions are where I would be careful. "How many PTO days roll over?" or "Am I eligible for FMLA?" have answers that depend on your handbook, your state and the person's tenure. A general chatbot will answer confidently and sometimes wrong. If you want AI here, use a setup that answers only from your own documents: a custom GPT or a Claude Project loaded with the handbook and benefits guide, Microsoft 365 Copilot pointed at your HR SharePoint, or Gemini inside Google Workspace with access to the HR drive. Tell it to reply "check with HR" when the answer is not in the documents, and keep leave eligibility decisions with a person.

Performance reviews and PIPs

Managers hate writing reviews, so they write two vague lines or paste last year's text. AI helps if you feed it facts: goals, results with numbers, specific examples, peer comments. Ask it to organize those into strengths, gaps and next-period goals, in the manager's voice. Without facts, you get "consistently demonstrates strong communication skills," which tells the employee nothing and helps nobody in a later dispute.

For a performance improvement plan, AI is good at structure: measurable expectations, timeline, support offered, check-in dates. It is bad at judging whether a PIP is fair or whether it lines up with a recent leave request or complaint. That timing question is exactly what turns a PIP into retaliation evidence. A person checks it, ideally with employment counsel.

Engagement surveys: reading 900 comments in an afternoon

An engagement survey with 600 responses usually has 800 to 1,000 open comments. Nobody reads them all, so the report becomes "people mentioned communication." AI can code every comment into themes, count them and pull representative quotes in one afternoon.

Two rules. Remove names, team names and any detail that identifies the person before uploading. And never report a theme for a group smaller than five people: with three people in a team, "my manager plays favorites" identifies the writer, and the trust in your next survey is gone.

The legal side of AI in HR (US)

There is no single federal AI hiring law in the US yet, but that does not mean there are no rules. The rules that already apply to hiring apply to hiring with AI, and you, the employer, remain responsible for the tool's output even if a vendor built it.

• Title VII, the ADEA and disparate impact. A screening tool that rejects a protected group at a much higher rate can create liability even if nobody meant to discriminate. The "four-fifths rule" from the federal Uniform Guidelines is the usual first check: if one group's selection rate is below 80% of the highest group's rate, look closer.
• The ADA. AI tools can screen out people with disabilities: a timed game, a video analysis of facial expressions, a gap in employment history. You must offer reasonable accommodations in the process and tell candidates how to ask for one.
• The EEOC. The agency has said for years that anti-discrimination law covers algorithmic decisions. In early 2025 it removed its AI technical assistance documents from its website, but Title VII, the ADA and the ADEA did not change, and private lawsuits continue.
• NYC Local Law 144. If you use an automated employment decision tool for candidates or employees in New York City, you need an independent bias audit within the year before use, a public summary of the results and notice to candidates at least 10 business days in advance.
• Illinois. The Artificial Intelligence Video Interview Act requires notice, an explanation of how the AI works and consent before AI analyzes video interviews. Illinois also amended its Human Rights Act to cover discriminatory use of AI in employment decisions.
• Candidate data. California applicants have privacy rights under the CCPA as amended by the CPRA. Everywhere, keep resumes out of personal chatbot accounts: use a business plan where your data is not used for training.

State laws on AI in hiring keep changing (Colorado passed one, others have bills moving). If you hire in many states, put a 30-minute check with employment counsel on the calendar every six months.

Tools for AI in human resources and what they cost

If I ran HR at a 100 to 500 person company and had to start Monday with US$ 50 a month, I would buy one business seat of ChatGPT or Claude for HR and use the AI already inside the ATS. Nothing else until the team uses those every day.

Common mistakes with AI in HR

• Pasting resumes and employee files into a personal account. Use a business plan with training turned off, and strip Social Security numbers, dates of birth and medical details first.
• Letting the AI rank candidates by "fit." "Fit" scores reward writing style and can mirror the people you already hired. Ask for evidence per criterion instead.
• Trusting AI on FMLA, overtime or state leave law. It mixes states and outdated rules. Use it to draft the explanation after you confirm the rule.
• Buying a tool before checking the audit. Ask any vendor that scores candidates for its bias audit, the data it was tested on and who is liable. If the answer is vague, walk away.
• Using AI text as the manager's voice without editing. Employees recognize generic reviews. The AI drafts, the manager rewrites at least the examples.

Where to start this week

Pick the three "start here" rows from the first table: job descriptions, structured interview guides and rejection emails. They are low risk, save hours right away and get the team used to writing prompts with real details and clear limits. The AI prompts for HR post has those prompts ready to copy. When you have a role with more than 100 applicants, read the screening agent build before you try it.

Advertisement

Want to skip this part?

Ready-made AI agents for HR: nothing to build, no prompts to write, they already work for you.

• Screens hundreds of applications with quoted evidence per criterion and a shortlist for your review
• Sends respectful, on-time replies to every candidate who did not move forward
• Answers candidate questions about the role and the process around the clock

I want the ready-made agents