ChatGPT Training for HR Teams: Tasks, Prompts and Rules
ChatGPT training for HR teams works best when it is built around the documents HR already writes every week: job descriptions, interview questions, policy answers, onboarding plans and survey summaries. In a half-day session an HR team can learn to do these faster and more consistently, as long as they also learn the two rules that matter most in HR: keep personal data out of tools that are not approved for it, and never let AI make or rank decisions about people.
This guide is the outline I use when I train HR and people teams. It lists the tasks worth practising, prompts you can copy, the data and legal rules to agree before anyone starts, and a half-day agenda.
Why does HR need its own AI training?
Generic "intro to ChatGPT" sessions teach features. HR teams need something narrower. Their work is mostly writing and summarising, which AI does well, but it touches personal data and decisions about people, where mistakes are costly. A good HR session spends as much time on what not to do as on prompts.
It is also a legal point. Under the EU AI Act, AI systems used to recruit, select or evaluate people are classed as high-risk, and since February 2025 Article 4 expects companies to make sure staff who use AI have enough AI literacy. If your company hires in Europe, HR is the team that most needs to be trained, and most needs a record that it was. I wrote more about that in AI literacy training and Article 4.
Which HR tasks are worth practising first?
These eight tasks are frequent, mostly text-based, and safe to do with an approved company AI workspace when personal details are removed.
- Job descriptions from a short brief and your house template, in a consistent tone.
- Interview question banks tied to the competencies in the job description, with what a strong answer looks like.
- Policy FAQs: turning a long policy PDF into short answers employees can understand (the method is in this step-by-step guide).
- Onboarding plans: a first-30-days checklist for a new role, built from your own process documents.
- Internal announcements: drafts of policy changes, benefits updates and event notes in two languages.
- Survey comment summaries: grouping hundreds of anonymous comments into themes with example quotes.
- Training needs analysis: turning a skills matrix and manager notes into a short list of priorities.
- Replies to common employee questions using an approved knowledge base, so answers stay consistent across the team. A shared ChatGPT Project is the simplest way to do this (how to set one up).
What should the prompts look like?
Short prompts give generic results. HR prompts work when they include the role, the source document and the output format. Three examples from my sessions:
Job description
You are an HR partner at a [industry] company in [country]. Using the attached template and this brief: [brief], write a job description for [role]. Keep requirements to what the brief says, use inclusive language, avoid age or gender cues, and list 5 to 7 responsibilities. Ask me questions if the brief is missing anything.
Interview questions
From the job description below, list the 4 most important competencies. For each one, write 2 behavioural interview questions and describe what a strong, average and weak answer would include. Do not include questions about family, religion, age or health.
Survey comments
Here are anonymous comments from our engagement survey. Group them into at most 6 themes. For each theme give a one-line summary, how many comments mention it, and 2 short example quotes. Do not guess who wrote anything.
The pattern is the same each time: context, source, constraints, format, and an invitation for the model to ask questions.
What are the data and legal rules for HR?
Agree these before the session, ideally in writing:
- Use the company workspace, not personal accounts. Business and enterprise plans of ChatGPT, Claude and Copilot let the company control data retention and training settings. Personal accounts do not.
- Remove personal details (names, ID numbers, salaries, medical or disciplinary information) unless your data protection lead has approved the tool for that data.
- No AI decisions about people. AI can draft, summarise and suggest, but it must not screen, score or rank candidates or employees. That is where discrimination risk and high-risk rules apply.
- A human checks every output that goes to an employee or candidate.
- Keep a short record of who was trained and when. It is cheap, and it answers Article 4 questions later.
What does a half-day HR session look like?
| Time | Part | What the team does |
|---|---|---|
| 0:00 | Ground rules | Agree the data and decision rules above, using your own policy |
| 0:20 | Prompt basics | Rewrite a weak prompt into a strong one on a real HR task |
| 0:50 | Hands-on 1 | Job description and interview questions for an open role |
| 1:40 | Break | |
| 1:55 | Hands-on 2 | Policy FAQ and an employee announcement in two languages |
| 2:45 | Hands-on 3 | Survey comment summary on anonymised data |
| 3:20 | Shared setup | A team Project with shared instructions and approved files |
| 3:45 | Wrap-up | Each person picks one task to use AI on next week |
Every exercise uses the team's own documents, with personal data removed beforehand. That is what makes people keep using it after the session.
How do you measure whether it worked?
Pick two or three tasks from the list and measure them before and after: time to a first draft of a job description, time to answer a common policy question, or how long it takes to turn survey comments into a report. Check again after 30 days. The tasks people still use AI for after a month are the ones worth building into your templates.
Frequently asked questions
Can HR use ChatGPT to screen CVs?
I advise against it. Screening and ranking candidates is a decision about people, and AI systems used for it are high-risk under the EU AI Act and carry discrimination risk everywhere. Use AI to write the job description and the interview guide, and keep shortlisting with people.
Is the free version of ChatGPT safe for HR work?
Not for anything with personal data. Use a company plan your IT and data protection teams have approved, and remove personal details from anything you paste.
How long should HR AI training be?
A half day is enough to cover the rules and three or four real tasks. A multi-week programme, with practice between sessions, works better if you want AI built into your templates and processes.
Can the training run in Arabic and English?
Yes. I run sessions in Arabic, English or both, which suits HR teams in Egypt and the Gulf that write to staff in two languages.
I run this as a live half-day or multi-week session for HR and people teams, online or on site. See AI training for employees and executives or, for teams in the US, UK, Canada and Australia, remote AI training with public prices.