AI Workflows · HR workflow playbook · Updated June 2026

AI in HR: The Practical Workflow Playbook

Most HR teams are told AI will change everything and given no idea where to start. Here is the honest, task-by-task version: what to automate, what to keep human, and exactly how to run one high-value task today.

Short answer: AI belongs in the drafting and first-pass parts of HR work, not the deciding parts. Use it to draft job descriptions, summarize policy documents, prepare interview question sets, and turn rough notes into clean onboarding plans, then have a human review and own the result. Keep AI out of any decision about a specific person, including resume screening, performance ratings, and termination calls, unless that use is separately reviewed for bias and legality. The reliable pattern across the whole HR function is the same: AI drafts, a human decides, and a human always verifies before anything reaches an employee or a candidate.

Key takeaways

  • Automate the drafting, keep the deciding. AI is excellent at first drafts of repetitive HR documents and terrible as a sole decision-maker about a human being. Draw the line there and most of the risk disappears.
  • The highest-value, lowest-risk starting point is document drafting. Job descriptions, policy summaries, interview guides, and onboarding plans are high-volume, low-stakes-to-draft tasks where AI saves real time and a human still signs off.
  • Resume screening is the trap. Letting AI score or rank candidates is the one use that carries serious bias and legal exposure, and it is the one HR teams reach for first. Do not automate candidate selection without a documented bias audit and legal review.
  • Confidential employee data needs a rule before it goes anywhere. Salaries, health information, performance notes, and investigation records should not be pasted into a general-purpose AI tool without an approved policy. When in doubt, leave the names and the numbers out.

The HR team's real problem

HR sits on a mountain of writing-heavy, repeatable work: requisitions, offer letters, policy updates, handbook sections, interview scorecards, onboarding checklists, internal announcements. It is exactly the kind of work AI is genuinely good at speeding up. At the same time, HR is the function with the least room for an AI mistake, because almost everything it touches is about a specific person and a specific law. A sloppy marketing email is embarrassing. A biased hiring screen or a leaked salary band is a lawsuit.

That tension is why so many HR teams stall. They either avoid AI entirely and lose hours to drafting that a model could accelerate, or they get talked into automating the wrong thing, usually candidate screening, and quietly create legal exposure they cannot see. The way through is not a tool. It is a clear map of which tasks are safe to hand to AI for a first draft and which must stay fully human, paired with a repeatable way to run the safe ones well.

In HR, the question is never whether AI is smart enough. It is whether a human is still the one deciding, and whether you can prove it.

HR tasks: automate vs keep human

This is the map that matters. It is not about which AI tool you use. It is about which HR tasks you can responsibly hand to AI for a first pass, which you can use AI to assist on with a human firmly in the loop, and which should stay entirely human judgment. Read it as a routing guide for your week.

How to route common HR tasks
HR task AI role Why
Drafting a job description Automate the draft High volume, low stakes to draft, easy for a hiring manager to review and correct. A strong first-pass task.
Summarizing a long policy or handbook section Automate the draft AI is good at condensing dense documents. A human still confirms the summary is accurate against the source.
Building an interview question set Automate the draft Generating role-relevant, behaviorally framed questions is a fast win. The panel edits for fit and legality.
Turning onboarding notes into a structured plan Automate the draft Reformatting and structuring known information is exactly what AI does well. A manager verifies the details.
Drafting an internal announcement or FAQ Assist, human in the loop Useful for tone and structure, but anything employee-facing needs a human owner who checks every fact.
Preparing for a difficult conversation Assist, human in the loop AI can help you rehearse and structure, with no real names or confidential specifics pasted in.
Screening, scoring, or ranking candidates Keep human Serious bias and legal exposure. Do not let AI select people without a documented audit and legal review.
Performance ratings and promotion decisions Keep human A judgment about a person's career. AI may organize evidence, but it must not produce the rating.
Discipline, investigations, and terminations Keep human Legally sensitive and deeply personal. These decisions and the records behind them stay with people.
TLY rule of thumb Draft with AI, decide as a human If the output is a document, AI can draft it. If the output is a decision about a person, a human owns it.

Notice the pattern. Everything in the "automate the draft" rows produces a document that a human then reviews. Everything in the "keep human" rows produces a decision about a specific individual. That single distinction, document versus decision, is the cleanest line you can draw, and it maps almost perfectly onto where the legal risk lives.

A step-by-step workflow: drafting a job description with AI

To make this concrete, here is the full workflow for the single best place to start: drafting a job description. It is high volume, every HR team does it, the time savings are real, and a human reviewer catches anything off before it ever posts. Run it with Claude or whichever sanctioned tool your company provides. The steps are the same.

Step 1: Gather the real inputs first

Before you open the AI tool, collect the actual ingredients: the team and reporting line, the three or four outcomes this role must deliver in its first year, the genuine must-have skills versus the nice-to-haves, and your company's standard sections for level, location, and pay transparency. AI cannot invent these accurately, and a job description built on guesses is worse than none. You are assembling the raw material a human knows and the model does not.

Step 2: Give the model role, context, and constraints

Do not ask for "a job description for a recruiter." Give it the brief. The more real context you provide, the less generic and the more usable the draft.

Example prompt: "You are helping an HR team draft a job description. Role: Senior Recruiter, reporting to the Head of Talent, hybrid in Chicago. Top outcomes for year one: rebuild the engineering hiring pipeline, cut time-to-fill, and improve candidate experience. Must-have skills: full-cycle technical recruiting, structured interviewing, ATS fluency. Write a clear, inclusive job description with these sections: about the role, what you will do, what you bring, nice to have, and how we work. Keep the language plain and free of jargon and cliches. Do not invent a salary or any benefit I have not given you."

Step 3: Pressure-test the draft for inclusivity and accuracy

Once you have a draft, use the model to improve it, then judge the result yourself. Ask it to flag exclusionary or biased language, remove inflated requirements that would screen out good candidates, and check that the must-haves are genuinely necessary. This is where AI adds quiet value: it is good at catching the "10 years experience for an entry role" type of inflation that creeps into postings.

Example prompt: "Review this draft for language that could discourage qualified candidates from underrepresented groups, list anything that reads as a degree or experience requirement that is not truly essential, and suggest plainer alternatives. Do not change the factual requirements I confirmed."

Step 4: A human edits, fact-checks, and owns it

The model gives you a strong draft, not a finished posting. A person now confirms every factual claim: the reporting line, the location, the pay range if your jurisdiction requires one, the actual requirements. You cut anything that does not match the real role, align it to your company voice, and put your name on it. The AI accelerated the blank page. You are still the author.

Step 5: Log what you did

Keep a simple note that this job description was AI-drafted and human-reviewed, with who reviewed it. It takes ten seconds and it is the difference between a defensible process and a vague "the computer wrote it." Once your team trusts this workflow, save the winning prompt as a reusable template so the next requisition starts from your standard, not from scratch. That template is where the real time savings compound.

Honest real-world usage notes

A few things become clear once an HR team uses AI on real work rather than in a vendor demo.

The time savings are real but they live in the first draft, not the final product. AI gets you from a blank page to a solid eighty percent in minutes, which is genuinely valuable when you are drafting your fifth requisition of the week. The last twenty percent, the part that makes it accurate, legal, and yours, still takes human time and judgment. Teams that expect AI to deliver a finished, postable document untouched are the ones that get burned.

The tool is only as good as the inputs you give it. A vague prompt produces a generic, slightly off job description that reads like every other one online. A specific prompt with real outcomes and real constraints produces something genuinely useful. Most of the skill is in the briefing, which is why this work rewards HR professionals who know their roles deeply, not the ones who can write clever prompts.

And the temptation always points toward the riskiest use. The moment a team gets comfortable drafting job descriptions, someone suggests using AI to screen the applicants. That is the exact line not to cross without a bias audit and legal review. The same tool that safely drafts a posting becomes a legal liability the instant it starts ranking the people who apply. For the broader safe-use posture across the function, our companion piece on how HR teams use AI safely covers the data and privacy side in more depth.

Bias, legal, and data guardrails

HR is the highest-stakes place to use AI inside a company, because almost every task touches a real person and a real law. These are the non-negotiables.

No automated resume screening without an audit

Do not let AI score, rank, or reject candidates as a sole or primary basis for a decision without a documented bias audit and legal review. Automated hiring tools are squarely in the sights of regulators and plaintiffs, and a tool that "passed" a vendor's check can still produce a disparate impact in your specific applicant pool. If you are considering any AI in selection, treat it as a regulated decision, not a productivity hack. Our briefing on when a passed hiring audit is still unfair walks through exactly why.

Keep confidential employee data out of general tools

Salaries, health and disability information, performance notes, investigation records, and anything tied to a named individual should not go into a general-purpose AI tool without an approved company policy. When you need AI's help on a sensitive task, strip the names and the numbers and work with the structure instead. If you do not yet have a policy, our one-page AI policy briefing gives you a starting template.

Verify before anything reaches a human

AI invents plausible-sounding details, including fake policy citations, wrong legal thresholds, and benefits you do not offer. Every AI-assisted HR document gets a human fact-check before it reaches an employee or a candidate. The verification step is not optional polish. It is the part of the job that protects both the employee and the company, and it stays yours.

How we built this playbook

This playbook reflects hands-on use of AI on the actual HR drafting tasks described above: job descriptions, policy summaries, interview question sets, and onboarding plans. The automate-versus-keep-human routing reflects the practical risk pattern we see across HR work, where the line between drafting a document and deciding about a person tracks the line between low and high legal exposure. We do not publish invented survey numbers or fabricated case results. Where a claim touches law or regulation, confirm it against current guidance and your own counsel before you rely on it, because rules around AI in employment are changing quickly and vary by jurisdiction. We date this guide and refresh it as the tools and the rules evolve.

What this means for your week

You do not need an AI strategy to start. You need one safe task. Pick job description drafting, run the five-step workflow above on your next requisition, and feel how much faster the blank page goes when AI handles the first eighty percent and you handle the part that matters. Then add the next safe task, and the next, while keeping every decision about a specific person firmly in human hands.

That discipline, AI drafts and a human decides, is the entire game in HR. Get it right and you reclaim hours a week without creating a single new liability. Get it wrong, usually by automating selection, and you trade a little time for a lot of risk. The professionals who win this moment are the ones who know exactly where that line sits and never cross it by accident.

Part of TLY's AI Workflows → workflow playbooks for senior professionals.

Frequently asked questions

What HR tasks can AI actually do well?

AI is strongest at drafting and summarizing: writing first drafts of job descriptions, condensing long policy and handbook documents, generating interview question sets, turning rough onboarding notes into structured plans, and shaping internal announcements. In every case a human reviews and owns the result. AI is weakest, and should not be used as the decider, on anything that judges a specific person, including candidate screening, performance ratings, and termination decisions.

Can AI screen resumes or rank job candidates?

Technically yes, but you should not let it do so as the basis for a decision without a documented bias audit and legal review. Automated hiring tools carry real bias and legal exposure, and a tool that passed a vendor's fairness check can still produce a discriminatory result in your specific applicant pool. Treat any AI use in candidate selection as a regulated decision, not a time-saver, and keep a qualified human accountable for every outcome.

Is it safe to put employee data into an AI tool?

Not into a general-purpose AI tool without an approved company policy. Salaries, health information, performance notes, investigation records, and anything tied to a named employee should stay out unless your organization has a vetted, compliant setup. When you need AI's help on a sensitive task, remove the names and numbers and work with the structure. When in doubt, leave the confidential details out.

What is the safest HR task to start with?

Drafting job descriptions. It is high volume, every HR team does it, the time savings are immediate, and a human reviewer easily catches anything off before it posts. It teaches the core habit, AI drafts and a human decides and verifies, on a task where a mistake is cheap to fix. Once that workflow is trusted, you can extend the same pattern to policy summaries, interview guides, and onboarding plans.

Does using AI in HR create legal risk?

It can, but the risk is concentrated in a few uses. Drafting documents that a human reviews is low risk. The exposure lives in letting AI make or heavily influence decisions about specific people, especially in hiring, and in feeding confidential employee data into tools that are not approved for it. Keep AI to drafting, keep decisions human, keep sensitive data out, and document your process, and you remove almost all of the legal risk while keeping the time savings.

Build the workflow, not just the opinion

Knowing that AI should draft while a human decides is the easy part. Building it into how your HR team actually works, with the prompts, the templates, and the guardrails that make it repeatable and defensible, is the skill. That is what we teach: a practical system for putting AI to work across the HR function without ever handing it a decision it should not make.

Go deeper with The Leveraged HR Professional course Join The Leverage Club for $49 and get the HR prompts, templates, and task-routing guides Not sure where to start? Take the 2-minute course finder

Sources: TLY hands-on use of AI on HR drafting tasks including job descriptions, policy summaries, interview guides, and onboarding plans (June 2026); general guidance on bias and legal exposure in automated employment decisions, confirmed against current regulatory guidance and counsel. Rules around AI in employment vary by jurisdiction and change quickly; verify against current guidance before relying on this page.