The short answer
AI for HR teams in November 2026 is safe and useful for writing and admin: job ads, interview guides, policy drafts, onboarding plans and handbook questions, using a business plan of ChatGPT, Claude or Gemini Notebook where staff data is not used for training. Ranking, screening or evaluating people is different. The EU AI Act lists AI for recruitment, selection and worker evaluation as high-risk under Annex III, with those rules applying from 2 December 2027 after the AI Omnibus delay.
| Safe to hand over | Job ads, interview questions, policy drafts, onboarding plans, handbook answers, translations |
|---|---|
| Keep human | Screening, ranking, hiring, performance and dismissal decisions |
| EU AI Act: workplace emotion recognition | Banned since 2 February 2025 |
| EU AI Act: hiring and worker evaluation tools | High-risk (Annex III, point 4); rules apply from 2 December 2027 |
| AI Omnibus | Entered into force 27 July 2026 and moved the high-risk date |
| Plans to use | ChatGPT Business, Claude Team, Google Workspace with Gemini, Microsoft 365 Copilot |
How HR teams use AI in 2026, and where the line is
Most HR work is writing: job ads, offer letters, policies, onboarding plans, answers to the same twenty questions. General assistants such as ChatGPT from OpenAI, Claude from Anthropic and Microsoft Copilot are good at all of it. The trouble starts when AI moves from writing about roles to judging people. Ranking CVs, scoring interviews, predicting who will leave or flagging "low performers" are decisions about someone's livelihood, and the law in the European Union now treats tools that do this as high-risk.
So the rule on this page is simple. Hand AI the words and the admin. Keep every decision about a person with a person, and if you buy a dedicated tool that screens or evaluates people, treat it as a regulated system. Our best AI for HR and recruiting page compares the tools in more depth.
The 6 HR tasks worth handing to AI
1. Fair, clear job ads (ChatGPT or Claude)
Give the assistant the role's real responsibilities and must-have skills and ask for a job ad in plain language that avoids gendered wording and unnecessary requirements. Ask it to list any requirement that could exclude good candidates without being essential. The "Hiring in two languages" stack on /stacks adds DeepL for a natural translation.
2. Structured interview guides and scorecards (ChatGPT or Claude)
AI is good at turning a job description into competency-based questions with what a strong, adequate and weak answer looks like. That makes interviews fairer because every candidate gets the same questions. The scoring stays with the interviewers.
3. Policy and handbook drafts (Claude)
Claude is strong on long documents. Give it your current policy, the change you need and your tone, and ask for a redline with every change explained. A person with employment law knowledge signs it off, because AI does not know your jurisdiction's current law reliably.
4. A handbook people can ask questions (Gemini Notebook)
Add your handbook pages as sources in Gemini Notebook (formerly NotebookLM) and share the notebook. Its answers are grounded only in your sources, with citations. For work and school accounts Google says uploads, queries and responses are not reviewed by humans and not used to train AI models. This is the "New-hire onboarding that runs itself" stack.
5. Onboarding plans and welcome notes (Claude or ChatGPT)
Ask for a 30-60-90 day plan from the role description and team goals. The manager edits it and signs the welcome, not the AI.
6. Survey and exit-interview themes, anonymised (ChatGPT or Claude)
Remove names and identifying details first, then ask for themes with counts and representative quotes. Ask the assistant to say how many comments support each theme, so a single loud comment does not become a "trend".
What HR should keep human
- Screening and shortlisting. Do not paste CVs into a general chatbot and ask it to rank them. It can be biased in ways you cannot see, and in the EU this use is in the high-risk category.
- Interview scoring and hiring decisions. AI writes the guide. People score and decide.
- Performance, promotion and dismissal. These are listed in Annex III too. AI can help you write a clear review once you have decided what it says.
- Sensitive conversations. Grievances, health, investigations and redundancies need a person, and the data should not go into AI tools at all unless your legal and data protection advisers approve the setup.
The EU AI Act and HR: what applies and when
This is a summary of official EU sources, not legal advice. The European Commission's AI Act page (updated 19 October 2026) gives the timeline.
- Banned since 2 February 2025: the prohibited practices in Article 5, which include emotion recognition in workplaces. Do not use any tool that claims to read candidates' or employees' emotions from video, voice or text.
- AI literacy since 2 February 2025: the Commission lists AI literacy obligations alongside the prohibitions from that date, so train the people who use AI.
- High-risk from 2 December 2027: Annex III, point 4 covers AI used for recruitment or selection (placing targeted job ads, analysing and filtering applications, evaluating candidates) and AI used to make decisions affecting the terms of work relationships, such as promotion or termination, allocate tasks based on individual traits or behaviour, or monitor and evaluate workers. The AI Omnibus, which entered into force on 27 July 2026, moved these rules to 2 December 2027.
When the high-risk rules apply, Article 26 sets the employer's duties as a deployer: use the system according to the provider's instructions, give human oversight to people with the competence, training and authority to do it, make sure input data you control is relevant and representative, monitor how it works, keep the logs it generates for at least six months, inform workers' representatives and affected workers before using it at the workplace, and tell people when a system makes or assists decisions about them. If you are buying a screening or assessment tool now, ask the vendor how it will meet these rules by December 2027. Read more in our explainer on the EU AI Act.
A starter AI stack for HR teams
| Job | Tool | Plan | Why |
|---|---|---|---|
| Job ads, guides, letters | ChatGPT or Claude | ChatGPT Business or Claude Team | Not used for training by default on these plans |
| Handbook questions | Gemini Notebook | Google Workspace account | Answers from your own sources; no human review or training on work accounts |
| Inside Microsoft 365 | Microsoft 365 Copilot | Work account with a licence | Enterprise data protection; prompts not used to train foundation models |
| Handbook and pipeline home | Notion | Business for Notion Agent | Notion says customer data is not used to train models by default |
| Translation | DeepL | Paid plan | The free version may use your texts for training |
A 30-day AI rollout plan for an HR team
- Days 1 to 5: rules before tools. Write a one-page policy: which tools and plans are approved, that no personal data goes into personal accounts, and that AI never ranks, screens or scores people. Check the plan of every tool in use; personal ChatGPT and Claude plans train unless you switch it off.
- Days 6 to 12: writing tasks. Use AI for every new job ad and interview guide. Have a second person check each ad for wording that excludes people.
- Days 13 to 19: the handbook notebook. Load the handbook into Gemini Notebook on a work account and test it with the twenty most common questions. Fix the pages behind wrong answers.
- Days 20 to 25: AI literacy. Run a one-hour session for HR and hiring managers on what AI is allowed to do here and what it gets wrong.
- Days 26 to 30: audit vendors. List every HR system with an AI feature. For any that filters, ranks or evaluates people, ask the vendor for its EU AI Act plan and note who provides human oversight.
The risks for HR teams, and how to manage them
- Bias in screening. The biggest risk, and the reason the EU classes these tools as high-risk. Keep AI out of shortlisting unless you use a system built and documented for it.
- Personal data in the wrong plan. Employee and candidate data is personal data under GDPR in the EU and UK. Use business plans, remove names where you can, and involve your data protection officer.
- Wrong legal answers. AI gets employment law details wrong, especially local rules and recent changes. Every policy goes past someone qualified.
- Over-promising confidentiality. If you use AI on survey or exit-interview answers, tell people who will see them and how.
Copy-ready prompts are in prompts for HR and prompts for job interviews.
Prompts for the 6 HR tasks
None of these prompts asks AI to rank, screen or judge a person. Replace the square brackets.
Prompts to copy
1. A fair job ad
Write a job ad for a [role] at [company] from these notes: [responsibilities, must-have skills, nice-to-have skills, location, salary range]. Plain language, under 400 words. Avoid gendered wording and jargon. Then list any requirement that might exclude good candidates without being essential, and suggest how to reword it.
You get: A clear ad and a list of requirements to reconsider
2. A structured interview guide
From this job description: [paste], write 8 competency-based interview questions, two per competency: [list four]. For each, describe what a strong, adequate and weak answer would include. Do not score any candidate. Format as a table we can print.
You get: The same fair questions for every candidate
3. Redline a policy
Here is our current [policy name]: [paste]. We need to change it so that [change]. Rewrite it in plain language for employees, keep everything else the same, and list every change you made with the reason. Flag anything where the law in [country] may affect the wording, so our adviser can check it.
You get: A clean draft and a change list for legal review
4. Handbook answers with the page (Gemini Notebook)
Answer only from these handbook sources and name the page you used. If the answer is not covered, say so and tell me which team to ask. Question: [question].
You get: Answers people can check against the handbook
5. A 30-60-90 day plan
Write a warm welcome note and a 30-60-90 day plan for a [role] joining [team]. The team goals are [goals]. Plain language, realistic goals, one learning goal and one relationship goal per phase. The manager will edit and sign it.
You get: A plan the manager edits rather than writes
6. Themes from anonymised survey comments
Here are anonymised survey comments with names and identifying details removed: [paste]. Group them into themes. For each theme, give the number of comments, two representative quotes and one suggested action. Do not guess who wrote anything.
You get: Themes with counts, so loud comments do not become trends