AI in HR: job ads, onboarding and where the limits are
Written with AI and checked against the sources by a second AI. As of November 16, 2026. More in the notes on AI.
HR involves a lot of writing: job ads, invitations, onboarding plans, references. An AI can help with that. It shouldn’t judge people, and data from applications and personnel files doesn’t belong in an ordinary AI chat. Here is what works and where the limits are, as of November 2026.

Key points
- For job ads, interview guides, onboarding plans and templates, an AI is a good help.
- Have every ad checked for phrases that could exclude people because of age, gender or origin. You finish the check yourself.
- Applications, sick notes and personnel files don’t belong in a private AI chat.
- AI that screens applications or rates performance is generally high-risk under the EU AI Act.
- If there is a works council, talk to it before you introduce AI in HR.
In short: texts and plans yes, judgments about people no
An AI phrases, structures and builds templates well. It can’t judge whether an applicant fits into the team or whether a colleague deserves a formal warning (Abmahnung). And it sees everything you give it. That leads to a simple division:
| Well suited | Only with an approved service and rules | Don’t hand over |
|---|---|---|
| Job ads, invitations, rejection templates | Texts with real names and data of employees | Choosing between applicants |
| Interview guides, onboarding plans, checklists | Analyses with personnel data | Performance ratings, formal warnings, dismissals |
| Reference drafts from bullet points with placeholders | Summaries of internal documents | Decisions on pay or promotion |
All the examples in this article work with an ordinary AI chat, without data about real people. Which AI you use for work is up to your company.
In the trial lesson you put your first instruction together and get feedback. No account, in a few minutes. Start the trial lesson
Job ads: first the profile, then the text
A good ad doesn’t start with the text but with the question of what the person really needs to be able to do. Amara is an HR officer at a brewery with 120 employees. For a position in bottling, she first writes down what happens on a normal day and has the AI turn it into an ad:
Write a job ad for a skilled worker in the bottling department of a brewery, full time, two-shift work.
Tasks: set up and monitor the bottling line, fix malfunctions, take quality samples, document the shift handover.
Must have: completed vocational training as a machine and plant operator or similar, willingness to work shifts.
Nice to have: experience in the food or beverage industry, forklift license.
We offer: [enter your real benefits here].
At most 250 words, formal, clear sentences, no clichés.
Don’t invent any benefits or a salary. Where something is missing, write [missing].Then check for excluding phrases
Under the German General Equal Treatment Act (AGG), a position may not be advertised in a way that disadvantages people because of race or ethnic origin, gender, religion or belief, disability, age or sexual identity. Phrases meant harmlessly can do that too. So have the AI search for them:
Check this job ad for phrases that could disadvantage people because of age, gender, race or ethnic origin, religion or belief, disability or sexual identity.
Quote each passage word for word, explain the problem in one sentence and suggest a neutral phrasing.
Also check whether the job title addresses all genders.
[paste the ad here]- “Young, dynamic team” or “career starters” can exclude older applicants.
- “German native speaker” can disadvantage people because of their origin. Write which language level the work really requires.
- “Resilient and physically fit” says little. Better to describe what has to be carried or how long people stand.
The AI finds a lot, but not everything. Read the ad yourself at the end and ask yourself of every sentence: do we really need this for the work? Discriminatory ads can become expensive for the company.
Applications: the data stays out
CVs and cover letters contain a lot of personal details, sometimes also about health or origin. They don’t belong in a private AI chat. Even summarizing an application “just for an overview” passes the data on to the provider.
If an AI screens or filters applications or rates applicants, this is generally high-risk under the EU AI Act. From December 2, 2027, strict obligations apply to such systems, including for the company that uses them: human oversight, informing employees and their representatives. Whether you introduce something like this isn’t one person’s decision but the company’s, together with data protection and the works council. More on this in the guide The EU AI Act explained for working people.
What the AI does well without applicant data: draft an evaluation sheet you use to read every application by the same criteria. The criteria come from your profile, the rating is done by you.
Job interviews and rejections
Interviews become fairer and easier to compare when all applicants get the same questions. The AI can build such a guide from your profile:
Create a guide for a 45-minute job interview for the bottling position (profile below).
Eight questions, each on one requirement from the profile. Where possible, ask about real situations: “Tell me about a malfunction you fixed.”
Under each question: how we recognize a good answer, in two bullet points.
No questions about family, plans to have children, pregnancy, age, religion, health or origin.
[paste the profile here]A template for rejections is worth having: friendly, short, without a reason that could relate to personal characteristics. The AI writes the template, you insert names in your applicant management software.
Onboarding that works on day one
Dario runs a scaffolding company with 18 people. New colleagues used to stand in the yard on their first day without knowing where to go. Now he has the AI build a plan for the first two weeks from his notes:
Turn my notes into an onboarding plan for new scaffolders for their first two weeks.
Format: table with Day, What happens, Who is responsible (roles only), Done.
On the first day: occupational safety briefing, hand out protective equipment, show the vehicle and the storage yard.
Use only what is in my notes. If you think something important for safety is missing, write it as a question below the table, not into the plan.
Notes:
[paste the notes here]The questions below the table are the most valuable part. Whether a safety briefing is complete is something Dario checks with his occupational safety specialist and the documents of the statutory accident insurance institution (Berufsgenossenschaft), not with the AI.
Employment references: draft yes, secret code no
Under the German Trade Regulation Act (Gewerbeordnung), an employment reference (Arbeitszeugnis) must be clear and understandable. Phrases that say something other than what is written are not allowed. This calls for care: AIs have read many references and like to adopt the typical formulas, whose meaning you don’t always know.
Rania is an HR administrator at a home care service. She writes the bullet points with the manager and gives them to the AI with placeholders, without name, date of birth and periods of sickness:
Write a draft of a detailed employment reference for [name], registered nurse, employed from [date] to [date].
Tasks and assessment from these bullet points:
[paste bullet points]
Rules: clear, unambiguous sentences. No coded reference formulas. Assess only what is in the bullet points.
Mark every sentence that contains an assessment with [A] so we can check each one.The manager reads every marked sentence and asks: is this true, and would I say it this way? An electronic reference is only allowed with the person’s consent and a qualified electronic signature.
Works council, data protection and rules in the team
If a company introduces a technical system suited to monitoring employees’ behavior or performance, a works council generally has a right of co-determination. With AI in HR this is often the case. So talk to each other early, not once everything is set up.
- Clarify which AI is approved for HR work and with which account.
- Write down which data may go into it and which never may, such as health data and applications.
- Decide who reads AI texts before they are sent.
- Record how you trained the team. Companies have to promote their people’s AI literacy.
Writing clear instructions, checking results and keeping data out is what you practice in the learning app “KI können”, in short lessons with examples from the office, and you apply it in your own AI. You can start with the trial lesson without an account.
Sources
- German General Equal Treatment Act (AGG), Section 11 Job advertisements (in German) (opens in a new tab)
- German General Equal Treatment Act (AGG), Section 1 Purpose of the act (in German) (opens in a new tab)
- German Trade Regulation Act (Gewerbeordnung), Section 109 Reference (in German) (opens in a new tab)
- German Works Constitution Act (Betriebsverfassungsgesetz), Section 87 Rights of co-determination (in German) (opens in a new tab)
- Regulation (EU) 2024/1689 on artificial intelligence (EUR-Lex) (opens in a new tab)
- Regulation (EU) 2026/1744, “Digital Omnibus on AI” (EUR-Lex) (opens in a new tab)
- Directive (EU) 2023/970 on pay transparency (EUR-Lex) (opens in a new tab)
- Federal Anti-Discrimination Agency (Antidiskriminierungsstelle des Bundes, in German) (opens in a new tab)
Information as of November 16, 2026. Laws and offers can change; this article is not legal advice.
Read more
Try it now
The trial lesson with examples for HR specialist. No account, a few minutes.
Start the trial lesson