# Where should an HR team start with AI if nobody on the team has used it yet? > Start with three team rules on employee data, then one low-risk drafting task, then a policy answer bot. A plain order for HR teams, with free lessons. Source: https://learn.ynteractive.com/content/questions/how-to-start-using-ai-in-hr · Updated 2026-10-21 · Free from Stepthrough (https://learn.ynteractive.com) By [Robert Breen](https://learn.ynteractive.com/content/about) · Published October 21, 2026 · Updated October 21, 2026 · 4 min read An HR team that has never used AI should start with rules, not tools: agree in one meeting what employee data never goes into a [chatbot](https://learn.ynteractive.com/content/glossary/chatbot), which account everyone uses, and that a person reads every answer before an employee sees it. Then practice on one low-risk drafting task together, starting with [Use AI at Work Without Leaking Client Data](https://learn.ynteractive.com/content/safe-ai-at-work-what-not-to-paste), a free 7-minute lesson. HR is a harder place to start than marketing or sales, because almost everything HR touches is personal: medical notes, pay, performance problems, investigations. That is not a reason to wait. It is a reason to go in a fixed order, so the first mistake is a clumsy email draft, not a leaked file. ## What order should an HR team go in? 1. **Week 1: Agree the three rules.** Write them on one page. What never goes in (Social Security numbers, medical and leave certifications, investigation notes, bank details). Which account you use (a business plan your admin controls, not personal logins). Who checks the output (the person who sends it, every time). 2. **Week 2: One shared drafting task.** Pick something everyone writes and nobody loves: [offer letter](https://learn.ynteractive.com/content/glossary/offer-letter) cover notes, open enrollment reminders, holiday schedule memos. Each person drafts one with AI and the team compares [prompts](https://learn.ynteractive.com/content/glossary/prompt). 3. **Week 3: The careful replies.** Employee questions where the honest answer is "it depends": leave, pay timing, accommodations. Here AI helps you write a kind, accurate reply from the facts you give it. It does not decide anything. 4. **Week 4 and later: A policy answer bot.** Only once the team trusts its own checking habits, try an agent that answers handbook questions from your handbook and sends everything else to HR. Recruiting tools, performance review help and anything that scores people come last, and get their own review. Those are the uses where fairness rules and new state laws apply most directly. ## What does this look like in a real lesson? In [Answer an Employee's Leave Questions Without Overpromising](https://learn.ynteractive.com/content/hr-leave-policy-reply) you play Renée, the HR manager at a 38-person architecture firm. An employee, Kofi, asks three things at once: is his parental leave 12 weeks at full pay, can he add unpaid weeks, and is his December start date approved. You paste his email thread into ChatGPT along with the exact facts from the leave policy, get a first draft, then ask ChatGPT to make a table matching every promise to a fact. Three of the six claims hold up. The pay, the unpaid weeks and the approved dates do not, because the draft promised more than the policy says. You fix those three lines and send a warm reply from Gmail. That is the habit worth teaching a new team in week 3: facts in, draft out, then check every promise before it reaches an employee. Later, [Answer Employee Policy Questions Instantly using Airia](https://learn.ynteractive.com/content/airia-hr-policy-agent-employee-handbook-questions) shows the week 4 step: you upload a handbook, build an agent that answers only from it, and test a question the handbook does not cover to make sure it says it cannot find the answer. ## What mistakes do new HR teams make with AI? - **Letting everyone pick their own tool.** Five people on five personal accounts means five sets of data settings nobody has checked. Choose one approved tool first. - **Starting with the hardest case.** A termination letter or an investigation summary is the worst first task. The stakes are high and the facts are sensitive. - **Pasting the whole file.** A draft about a leave question does not need the employee's diagnosis or ID number. Use a role or a placeholder. - **Treating the draft as the decision.** AI can word a reply. Whether leave is approved, an accommodation is reasonable or a policy applies is still an HR call, and sometimes a call for counsel. - **Skipping the written rules.** If nobody wrote down what stays out, a new hire will learn it after the mistake. ## Which tools should an HR team try first? - **ChatGPT or Claude** on a business plan for drafting, rewriting and checking replies against your policy text. - **Airia** when you are ready for a no-code agent that answers handbook questions from your own files and can be shared with staff. - **Your HR system's own AI features**, if your HRIS already has them, since the data is already there under your company's contract. For the full path of lessons for one HR person, see the [HR generalist start guide](https://learn.ynteractive.com/content/start/hr-generalist). For the automations that come after the basics, see [AI automation for HR teams](https://learn.ynteractive.com/content/use-cases/ai-automation-for-hr-teams), and for a list of free courses, [the best free AI training for HR](https://learn.ynteractive.com/content/blog/best-free-ai-training-for-hr). ## Key takeaways - Rules come first: what never goes in, which account, who checks. - Start with low-risk drafting the whole team does, then move to careful employee replies. - AI drafts the words. People make every decision about pay, leave and employment. - Policy bots come after the team has strong checking habits. ## Frequently asked questions **Do we need an AI policy before HR uses AI?** You need written rules, even if they fit on one page. A short list of what never goes in, which tools are approved and who reviews output covers most of the risk while a fuller company policy is written. **Should HR use a free ChatGPT account?** Not for anything about real employees. Use a business plan your company controls, with training on your data turned off, and keep personal details out of prompts even then. Read Is it safe to put employee records or performance reviews into ChatGPT? before you paste anything. **How long does it take an HR team to get comfortable?** A month of short weekly sessions is enough for the basics in this order. Comfort comes from using it on real, low-risk tasks, not from training alone. ## Learn it hands-on Free, interactive lessons: you do every step yourself in a practice copy of the real tool, it checks your work as you go, and nothing touches your real accounts. - [Use AI at Work Without Leaking Client Data using ChatGPT](https://learn.ynteractive.com/content/safe-ai-at-work-what-not-to-paste): Turn off model training, strip a tenant's bank details and address out of a prompt, then catch the two promises ChatGPT made up before the reply… (about 7 minutes) - [Answer an Employee's Leave Questions Without Overpromising using ChatGPT for Recruiting & HR](https://learn.ynteractive.com/content/hr-leave-policy-reply): An employee asks about pay, unpaid weeks and whether his leave dates are approved. Give ChatGPT the thread and your leave policy, catch the answers… (about 10 minutes) - [Answer Employee Policy Questions Instantly Using Airia](https://learn.ynteractive.com/content/airia-hr-policy-agent-employee-handbook-questions): HR answers the same time-off and benefits questions all day. Give Airia your handbook, build a no-code agent that answers from it, test it, then… (about 10 minutes) [Start the free lesson](https://learn.ynteractive.com/modules/safe-ai-at-work) ## AI tools in this answer Each one in the Stepthrough AI tools directory: what it does, who it is for and the free lesson where there is one. [ChatGPT](https://learn.ynteractive.com/content/ai-tools/tool/chatgpt) · [Claude](https://learn.ynteractive.com/content/ai-tools/tool/claude) · [Airia](https://learn.ynteractive.com/content/ai-tools/tool/airia) [The full AI tools directory](https://learn.ynteractive.com/content/ai-tools) ## Related terms Plain-English definitions, each with an example from a free lesson. [Chatbot](https://learn.ynteractive.com/content/glossary/chatbot) · [Offer letter](https://learn.ynteractive.com/content/glossary/offer-letter) · [Prompt](https://learn.ynteractive.com/content/glossary/prompt) · [AI pilot project](https://learn.ynteractive.com/content/glossary/ai-pilot-project) · [AI policy](https://learn.ynteractive.com/content/glossary/ai-policy) · [Deferred revenue](https://learn.ynteractive.com/content/glossary/deferred-revenue) [The full AI glossary](https://learn.ynteractive.com/content/glossary) ## Related prompts - [ChatGPT settings checklist prompt](https://learn.ynteractive.com/content/prompts/chatgpt-account-settings-checklist-prompt): A copy-ready prompt that turns your plan and data rules into a one-page checklist of ChatGPT settings for staff, with a column… - [Check reply promises prompt](https://learn.ynteractive.com/content/prompts/check-reply-promises-prompt): A ChatGPT prompt that checks a draft before you send it: a table of every promise, the fact behind each, and safer wording for… - [Email newsletter prompt](https://learn.ynteractive.com/content/prompts/email-newsletter-prompt): A ChatGPT prompt for a short customer newsletter built from your numbered updates: one section each, one main call to action, no… [All free prompt templates](https://learn.ynteractive.com/content/prompts) ## Related questions - [Can AI answer employees' handbook and benefits questions without giving wrong answers?](https://learn.ynteractive.com/content/questions/can-ai-answer-employee-handbook-questions): Yes, if it answers only from your handbook, cites the section and hands anything personal to HR. How to keep it accurate, with a… - [Can lawyers use ChatGPT without breaking their ethics rules?](https://learn.ynteractive.com/content/questions/can-lawyers-use-chatgpt-ethically): Yes, if you understand the tool, protect client confidences, supervise staff, verify every citation and bill fairly. Not legal… - [How do consultants use AI on client work without exposing client data?](https://learn.ynteractive.com/content/questions/how-consultants-use-ai-client-work): Consultants use AI for kickoff recaps, market scans and decks, on business plans that do not train on data, with client names… - [How should an independent insurance agency start using AI?](https://learn.ynteractive.com/content/questions/how-insurance-agencies-start-with-ai): Pick one approved tool, set data rules, and start with policy review recaps and claim-status replies that never promise what the… - [Where should a paralegal start with AI if the firm has no AI policy yet?](https://learn.ynteractive.com/content/questions/how-paralegals-start-with-ai): Ask a supervising attorney first, keep client confidences out, and practice on non-confidential tasks until the firm sets rules… - [How should a finance team start using AI without breaking controls?](https://learn.ynteractive.com/content/questions/how-to-start-using-ai-in-finance): Treat AI as a preparer, never an approver: written rules, an approved tool, a person who reviews and evidence you keep. A plan… [More questions: AI by department](https://learn.ynteractive.com/content/questions/by-department) · [All AI questions](https://learn.ynteractive.com/content/questions) · [All free tutorials](https://learn.ynteractive.com/content)