AI for HR (2026): What a 10 to 200 Person Company Can Hand It, What Stays Human, and Where Each HR Record Goes

How should a small HR team use AI?

At a 10 to 200 person company, HR should use AI for the writing and keep the decisions. Hand it policies, job posts, interview questions, onboarding checklists and employee FAQ answers, all drafted from your own handbook. Keep it away from anything that decides something about a person. And sort every HR record before anyone pastes it anywhere: most of HR's paperwork is safe in a company AI account, and a short list of records never is. As of 2026, that sorting is the part most small companies skip.

At this size, "HR" is often one person, or an office manager who also does HR, or a founder doing it at night. That's who this page is for. The enterprise HR-tech conversation (predictive attrition models, AI talent marketplaces) is real and it isn't your problem yet.

What AI does well in HR, and what stays human

The split is simple to say and easy to forget under deadline: AI drafts documents, people make decisions about people.

• Drafts well: handbook sections and policy updates, job descriptions, interview question banks tied to the role, offer letter templates, onboarding checklists, scheduling and rejection emails, answers to "how many PTO days do I have" questions with the handbook section cited, and the first version of training material.
• Stays human: who gets hired, who gets promoted, compensation, performance ratings, discipline, terminations, accommodation requests, and anything to do with a complaint or investigation. AI can format a document about any of these after a person has decided. It doesn't get a vote.

That second list is also the one where a confident wrong answer costs the most. A bad job post gets edited. A bad termination letter gets a lawyer involved.

The HR data map: which records can go where

My one-page AI acceptable use policy sorts company data into three classes: green (public), yellow (internal, approved company AI account only) and red (no AI tool without a written, specific approval). That policy puts "employee HR and payroll records" in red as a block. Here's the same rule, record by record, so an HR lead can apply it without guessing:

Two things make this work. First, the yellow rows assume a company business account, not a personal one. If your team doesn't have one yet, pick a plan before you write the rule (the Claude Team plan breakdown is one place to start comparing what a team seat includes). Second, the carve-outs are written down. "Use your judgment" is how a resume ends up in someone's free account.

Hiring: where AI drafts, and where it must not decide

Recruiting is where the most AI products are being sold to HR, and where a small company should move slowest. The drafting side is easy to say yes to: a job description from the hiring manager's notes, interview questions mapped to the three skills the role actually needs, scheduling emails, and a respectful rejection template a person still sends.

The deciding side is different. Some states and cities now regulate automated tools that screen, score or rank candidates, with requirements that vary by place and keep changing. I'm deliberately not summarizing those rules here, because a summary on a blog is not what you want to rely on for a hiring decision. If any tool you're considering filters or ranks applicants, have employment counsel review it for every location you hire in, before it touches a real candidate. Until then, a person reads every application.

Onboarding: build the first 30 days from your own handbook

Onboarding is where a small HR team gets the most back for the least risk, because almost everything involved is yellow at worst. The workflow:

1. Load the handbook, benefits summary and org chart into one saved project in your company's assistant.
2. Draft a role-specific 30-day checklist from the job description: accounts to set up, people to meet, documents to read, the first real piece of work. The hiring manager edits it.
3. Draft the new hire's FAQ (PTO, payday, expenses, who to ask about what). One rule: every answer cites the handbook section it came from. If the draft can't cite a section, the answer isn't in your handbook, and that's a handbook gap worth fixing.
4. Draft the day-one and week-one emails from the manager, in the manager's voice, for the manager to send.

The citation rule is the whole safety system here. An answer with a section number can be checked in ten seconds. An answer without one is a guess in HR's voice.

A worked example: what onboarding drafting is worth at your size

Do this on your own numbers, because nobody else's apply. Count your hires last year, then time how long it takes to assemble one onboarding packet (checklist, FAQ, first-week emails) by hand. Multiply them.

An example, so the arithmetic is clear (these are placeholders, not a measurement): a 60-person company that hires 12 people a year and spends 4 hours assembling each packet spends 12 × 4 = 48 hours a year on assembly. If an AI draft plus a careful review takes half as long, that's 24 hours back, about three working days a year. Small, honestly. The bigger payoff is the FAQ: a cited, current set of answers stops the same twenty questions from reaching HR every month, from every employee, not just new ones. Count those too.

AI tools for HR: start with what you already pay for

Before you look at a single HR AI product:

• Check your HR or payroll system. Many now include AI features on some plans, or sell them as an add-on. They have one big advantage: your data is already there, under a contract you've already signed. Read what your plan includes on the vendor's own pricing page.
• Use your company's approved assistant (ChatGPT, Claude, Copilot or Gemini on a business plan) for the drafting work above.
• Add a specialist only for a named job those two can't do, and if that job is screening candidates, go back to the hiring section first.

I haven't quoted vendor prices on this page. They change often enough that a number I didn't read today is a number I won't print.

Who owns the company's AI rules? Usually HR.

At a small company, AI policy tends to land on HR's desk whether or not anyone decided it should. That's not a bad fit: HR already owns the handbook, the acknowledgment forms and the onboarding that teaches people the rules. The pieces are the one-page policy (approved tool, data classes, one sign-off rule), a line in onboarding, and a short working session so managers can answer "can I paste this?" on the spot. The management side of that, deciding which of a team's tasks AI takes and how quality gets checked, is in AI for managers; HR's job is making sure every manager is working from the same rules.

When the rules are written and it's time to teach them, the four-week order and the staff-hours math for a 50-person company are in AI training for employees.

Who can help an HR team roll out AI?

I work with owners and leadership teams at organizations of about 10 to 200 people, and HR is usually where the rules for a rollout get written down. For an HR lead and the managers who'll apply the rules, the lightest option is The Shop's team seats, $397 for your seat plus $97 for each teammate, 2 to 25 people. An HR lead plus nine managers is $1,270. That's the core of my AI training for employees: people working on their own documents, with the data rules explained by someone who can answer questions. If you need the whole picture first (which tool, which data rules, which teams go first), that's the AI Readiness Assessment: $2,500 flat, a written roadmap in two weeks. Setups that put the tool on your real files start from $4,500 and take about two weeks. If what you actually need is a person to own HR decisions, that's a different hire, and I've written up when fractional HR makes sense. Not sure which? A free 30-minute strategy call sorts it out.

What I'd do this week

Print the data map above and walk it with whoever owns HR, line by line, changing anything that doesn't fit your company. Then load the handbook into your company's assistant and ask it the ten questions employees ask most often. Every answer that can't cite a section is a handbook gap. Fix two of them before Friday.

Sources

• Search demand: DataForSEO keyword overview and AI search volume, United States, English (read October 8, 2026): "ai for hr" 2,900 Google searches a month, keyword difficulty 20; "ai tools for hr" 260, difficulty 4.
• Data classes: the one-page AI acceptable use policy on this site (green, yellow, red; employee HR and payroll records listed as red), as published October 8, 2026.
• The Shop team seat, Assessment and setup prices: justinmckelvey.com/ai-training, /assessment and /ai-implementation, as published October 8, 2026.
• No vendor prices and no specific employment law are quoted on this page. Check vendor pricing pages on the day you buy, and ask employment counsel about hiring rules where you operate.