HR & Recruiting

By 2026, AI in HR sits in the tension zone between massive efficiency levers and strict compliance overlays. This hub page shows which HR and recruiting tasks really benefit from AI, which ones are sensitive, and which regulatory duties under the EU AI Act, GDPR, BetrVG and US equivalents you must understand. The examples come from European SaaS recruiting and tech onboarding — deliberately without full-automation fantasies, because legal limits forbid them.

Where does AI pay off in HR & recruiting?

Job descriptions and postings are the least sensitive lever. ChatGPT or Claude turn a bullet list of requirements into a complete job posting in 1–2 minutes — including inclusive language, clear scope and candidate-oriented voice. Realistic time saving: 70% versus manual writing. Side effect: cross-departmental consistency, because a shared brand template runs as system prompt.

Candidate outreach and sourcing is the second lever. Personalised LinkedIn messages, re-engagement emails to old pipelines and follow-ups — text-heavy work, well suited to LLMs. Important: outreach feels individual only when the AI gets account-specific hooks (current role, recent posts, shared interests).

Resume screening (with great care!) is the third — and most sensitive — lever. The EU AI Act classifies fully AI-driven candidate screening as a high-risk application. What’s allowed: tools that parse résumés, structure them and sort by clearly defined criteria — final decisions must be human, candidates have a right to explanation, and regular bias audits are mandatory. In the US, EEOC, NYC Local Law 144 and Illinois AIVIA layer additional audit and disclosure requirements. Sloppy implementation invites fines up to EUR 35M under the AI Act, plus class actions in the US.

Onboarding materials is the fourth, often-underrated area. AI generates a structured per-role onboarding guide, ramp-up checklists and 90-day FAQs from internal documentation. Time-to-productivity for new hires typically drops 15–25%, because standard questions are answered immediately.

Internal knowledge-base bots are the fifth lever. A custom GPT or Claude project with access to the employee knowledge base (HR policies, IT setup, expense rules) handles 60–70% of recurring employee questions. The HR team is measurably relieved.

Performance review drafts is the sixth area. AI structures manager-supplied bullet points into consistent, fair review texts — with the clear convention that final ratings must be human. Bias risk persists, so sample reviews of AI drafts for consistency across demographic clusters.

Employee feedback analysis is the seventh lever. Anonymised engagement surveys or exit interviews are thematically clustered and prioritised by an LLM. HR sees the most important themes instead of reading 200 free-text answers manually. Prerequisite: anonymisation before the LLM call, otherwise re-identification risk emerges in small teams.

Practical examples

Both setups show the same pattern: AI is deployed in low-risk areas (postings, outreach, onboarding); candidate screening and performance evaluation stay human or run with a compliance layer plus human final decision. This split is not over-caution — it’s a direct read-out of AI Act and anti-discrimination law (AGG in Germany, Title VII / EEOC in the US).

Munich SaaS recruiting team (35 open roles per quarter). ChatGPT Team drives the job-posting and outreach setup. Workflow: hiring manager supplies a bullet list with requirements; a custom GPT with the brand voice as system prompt produces the full job post in German and English. LinkedIn outreach messages are generated from the candidate profile plus the role profile; the recruiter personalises in 30 seconds instead of 5 minutes. Effect: time-to-hire down 18%, candidate pipeline per recruiter up 40%. Important: résumé screening still runs manually — the team deliberately uses AI only at the top of the funnel, because legal risk in screening is too high to absorb.

Berlin tech startup (60 employees, 8 onboardings per quarter). Claude for onboarding guides and an internal Q&A bot over a Notion knowledge base. Workflow: at onboarding start, every new joiner gets access to a custom Claude project trained on HR policies, IT setup guides and team knowledge base. Standard questions (“How do I book PTO?”, “Where’s the VPN setup?”) go directly to the AI; complex questions escalate to HR. Effect: HR-team load during onboarding waves estimated 40% reduced, time-to-productivity for new hires shortened by 1.5 weeks on average. The works council was involved early; the co-determination agreement documents data flows and the right to manual escalation.

Risks & compliance — the most important section for HR

HR is the most regulator-sensitive domain for AI use. These four pillars are not optional.

EU AI Act high-risk classification: AI-driven candidate screening, performance evaluation and salary recommendations fall under Annex III of the AI Act and count as high-risk. Mandatory: conformity assessment before deployment, risk-management system, decision logging, human oversight over AI suggestions. Fines for breaches: up to EUR 35M or 7% of global turnover.

GDPR for candidate data: résumés, cover letters and interview transcripts contain especially sensitive data. Right to explanation (Art. 22 GDPR), right to erasure, clear purpose limitation. Rejection logs must be deleted after the application closes — including AI logs. US-based teams face state laws (CPRA, etc.) plus EEOC record-keeping rules.

Anti-discrimination: AI learns from historical data that often carries historical bias (e.g. gender, age, ethnicity). German AGG, US Title VII / EEOC and similar frameworks ban discrimination, regular bias audits across clusters are mandatory. Vendors like HireVue and Pymetrics provide their own audit reports; in-house builds need their own audit pipeline. Practical test: send identical applications with swapped demographic markers through the system and check AI scoring for consistency.

BetrVG and works-council co-determination: § 87 (1) Nr. 6 BetrVG kicks in for AI tools that can monitor employee behaviour or performance — that includes internal knowledge-base bots if they collect usage data. Settle the co-determination agreement before rollout, otherwise the works council can block deployment. Austria has analogous provisions under ArbVG, Switzerland under federal Mitwirkungsgesetz, and US employers should expect equivalent friction with HR analytics policies even without a formal works council.

Related topics

Foundations: What is AI? explains the concepts behind language models, RAG and high-risk classification. The comparison ChatGPT vs. Claude shows which all-rounder fits text-heavy HR tasks (job postings, outreach, onboarding drafts) better. Related use cases: Public sector & Legal for the compliance kinship, Customer Support & Service for internal service desks and onboarding bots, plus Everyday & Productivity for the office tools every HR team runs alongside.

Recruiting is classified as high-risk under the EU AI Act. Our AI Risks guide explains what that means in practice. Required reading for any HR team using AI: Bias & Fairness — covering the Amazon recruiting case, the Mobley v. Workday lawsuit, NYC LL 144 audit obligations and mitigation methodology. Anti-bias constraints in prompt design (e.g. “Score exclusively based on the listed skills”) are only one building block — patterns in the Prompt Engineering guide.