AI for HR Teams: Job Ads, Screening, Onboarding and a Usage Policy (2026)

Human resources is the department where AI saves the most time and, at the same time, the one where it must be used most carefully. On one side there is entirely text-based, recurring work such as job ads, interview question sets, onboarding documents and internal announcements. On the other side there is personal data such as CVs, salaries and performance. This guide separates the two: which task can be moved to AI safely today, and which only after a written rule.

The article follows the flow of the hands-on training I run with corporate teams. ChatGPT and Claude are used as the example tools, and the same logic applies to Copilot and Gemini. For the department-specific training format, see the AI training for HR teams page.

The limit first: where does candidate data stop?

HR's first contact with AI is usually a CV pasted into the chat window. This is the most common and the riskiest mistake. A CV is personal data, content entered in personal accounts is processed in the provider's systems and leaves the company's control. The rule is simple:

• A corporate account is essential: Business or Enterprise plans turn off data training and provide admin oversight. Candidate data is not processed with a personal account.
• The anonymisation reflex: No CV enters the tool before the name, phone number, email, national ID and photo are removed. A candidate number is enough for evaluation.
• The decision stays with a person: AI summarises, compares and generates questions. Screening, offer and performance decisions belong to the HR specialist.

Until these three rules are in writing, only the first group of flows below should be used. How the rule is written is explained in detail in the data security and privacy training content.

Tasks done safely today (no candidate data)

1. Job ads

The position's task list, the required competencies and the company's tone document are given to the tool, and three lengths of ad are requested: long for LinkedIn, medium for the careers site, short for social media. The best result comes when the team adds two of its ads that performed well before as examples. That way the tool imitates the company's language instead of producing a generic ad.

2. Interview question sets and scoring keys

For each position, competency-based questions, good, medium and weak answer examples for each question and a scoring table are requested. This reduces inconsistency between the managers who interview. The question set is produced once and only updated when the position opens again.

3. The onboarding pack

Scattered procedures, the organisation chart and the first-week plan are given to the tool, and a single "first 30 days" document is produced for the new starter. Variants by department are produced with a single request. Source-grounded tools such as NotebookLM also work well here for summarising long documents and Q&A.

4. Internal communication and announcements

For texts such as policy changes, benefit announcements and event invitations, the tone document and the main messages are given, and the email and intranet versions are requested together. For difficult announcements the text the tool produces is only a draft, and the final read stays with the HR manager.

Tasks done after the rule (with candidate data)

The common point in the table is this: AI prepares and organises in every row and decides in none. This distinction is fundamental both legally and for treating candidates fairly.

A representative week: an 8-person HR team

The scenario below is representative, not a measured client result. Suppose the HR team of a mid-sized company opens four new ads a week, holds around thirty interviews and takes on two onboarding groups a month. In the first week after the training the team does this:

• Monday: the tone document and two example ads are saved in the tool as an "HR writing guide". From then on every ad is produced with this guide.
• Tuesday: interview question sets and scoring keys are produced for the five most frequently opened positions and placed in the shared folder.
• Wednesday: the onboarding documents are reduced to a single pack, and department variants are produced.
• Thursday: the written usage rule draft is prepared and sent to legal and IT.
• Friday: the team notes how much time it saved on which task and picks three new flows for the second week.

The aim of this week is not speed but habit. By the end of the fourth week the team uses the tool not as a chat window but as a template bank for recurring work.

Three common mistakes

The generic prompt: The request "write a sales representative ad" produces the ad everyone has seen. Without the task list, the company tone and an example ad the result is unusable.

Making the tool the referee: The question "which of these two candidates is better" is not asked. The tool can produce biased patterns and the rationale for the decision does not stay in HR's hands.

Rolling out without a policy: If the team is trained but the rule is not written, three months later everyone uses their own method. The policy is one page, HR writes it, everyone signs it.

How to start

The shortest route is a department-specific one-day workshop: the team works with its own ads, its own onboarding documents and its own interview notes, and leaves at the end of the day with a template bank and a usage rule draft. For the program content and format, see the AI training for HR teams page, and for the full department catalogue the corporate AI training page. If the team is small or you want to learn yourself first, there is also the one-to-one session format.