# AI in Recruiting: Benefits, Limitations, and Legal Framework > AI in recruiting: where it helps, where it does not, what the AI Act, GDPR, and works councils require. With a tool selection checklist. Read the guide now. URL: https://techport.ai/en/hr-software/ki-recruiting --- 1. [HR Software](/en/hr-software)/ 2. AI in recruiting techport HR Suite # AI in recruiting: where it helps, where it does not, and the legal framework AI has arrived in recruiting. It drafts job advertisements, parses CVs, answers questions on the careers site, and evaluates applications. For HR leaders in mid-sized businesses, what matters is where AI genuinely saves work, where it causes harm, and what the KI-VO (EU AI Act), DSGVO (GDPR), and Betriebsverfassungsgesetz (BetrVG, Works Constitution Act) require. This guide contextualises these issues, addresses the question of AI-written applications, and concludes with a checklist for selecting a tool. ## Does this sound familiar? * Some roles attract so many applications that no one can read them all thoroughly, and many cover letters sound similar. * A vendor promises that their AI finds the right candidates, but cannot explain how. * The works council asks whether the new tool monitors performance or behaviour, and no one has an answer. * The hiring department debates whether AI-written cover letters should be rejected. ## Where AI is used in recruiting Use case What the AI does Classification under the AI Act Drafting job adverts Drafts from bullet points not high-risk, check text for AGG (General Equal Treatment Act) compliance Targeted ad delivery Select target audiences high-risk, Annex III no. 4 (a) explicitly names this Parsing CVs Structure data generally not high-risk, provided it does not influence the decision Evaluating, filtering, sorting applications Scores, rankings high-risk, Annex III no. 4 (a) Chatbot on the careers site Answer questions AI notice required (Art. 50, since 2 Aug 2026) Scheduling interviews Organisation generally not high-risk Inferring emotions from video, voice, or facial expressions Recognising feelings prohibited (Art. 5(1)(f), since 2 Feb 2025) The classification of CV parsing is based on the draft Commission guidelines from May 2026. The final version is still pending. ## Where AI helps * **Routine work.** Drafting text, transferring data, scheduling appointments, answering frequent questions. Here, AI saves time without making decisions about people. * **Consistent standards.** A well-built system applies the same approved criteria to every application, even the eightieth on a Friday afternoon. * **Overview.** With many applications, an assessment with citations shows where a requirement is evidenced in the CV. This accelerates reading, but does not replace it. * **Justifications.** When every assessment comes with evidence, a selection decision is easier to explain to the hiring department, works council, and candidates. ## Where AI does not help or causes harm * **The decision itself.** How someone works in a team, handles clients, or develops cannot be found in any CV. That requires interviews and work samples. * **Learning from past decisions.** A system that learns from historical hiring decisions adopts their biases. In 2018, Amazon abandoned an experimental rating system that penalised CVs from women. It had trained on ten years of predominantly male applications. * **Scores without justification.** A number that no one can explain cannot be audited by anyone—neither HR nor the works council nor the applicant. * **Emotions and facial expressions.** Emotion recognition in the workplace has been prohibited since February 2025. Under Commission guidelines, this also applies to recruitment processes. * **Fabricated information and manipulation.** Large language models sometimes hallucinate details not found in the CV. Hidden prompt injections in a CV can compromise an unprotected system. More on this in our guide to [CV parsing](/en/hr-software/cv-parsing). * **Blind trust.** Merely rubber-stamping suggestions means you are effectively no longer deciding for yourself. The AI Act explicitly identifies this risk as automation bias (Art. 14(4)(b)). ## The legal framework ### KI-VO (EU AI Act) AI systems that screen, filter, or evaluate applicants are high-risk AI systems (Annex III no. 4 (a)). The exemption under Art. 6(3) does not apply if the system performs profiling, i.e. evaluating the professional suitability of individuals in an automated manner. Following the amendment by Regulation (EU) 2026/1744, the obligations for such systems apply from 2 December 2027 instead of 2 August 2026. Employers deploying such a system are deployers (Art. 26). From the cut-off date, they must use it in accordance with the instructions for use, assign competent persons to oversee it, monitor operations, and retain logs for at least six months. Prior to deployment in the workplace, they must inform workers' representatives and affected employees (Art. 26(7)). Applicants must be informed that a high-risk system is being deployed (Art. 26(11)). The prohibitions under Art. 5 and the transparency obligations under Art. 50 already apply today: a chatbot on the careers page must be clearly identifiable as AI. The competent market surveillance authority in Germany is the Bundesnetzagentur (§ 2(1) KI-MIG, in force since 29 July 2026). For details, see the guide on the [AI Act in HR](/en/hr-software/eu-ai-act-hr). ### DSGVO (GDPR) In principle, no one may be subjected to a decision based solely on automated processing which significantly affects them (Art. 22 DSGVO). Following the CJEU's SCHUFA ruling (C-634/21, 7 Dec 2023), even a score can constitute such a decision if it plays a decisive role. For recruiting, this means: genuine human review, no automated rejections, no invisible pre-filtering. Data subjects also have the right to an understandable explanation of the procedure applied (CJEU C-203/22, 27 Feb 2025). For AI-assisted applicant scoring, a Data Protection Impact Assessment (DPIA) under Art. 35 DSGVO is generally required. ### AGG (General Equal Treatment Act) The AGG does not differentiate between discrimination caused by a human or by software. Criteria tied to age, ethnic origin, gender, or other protected characteristics are strictly off-limits in an AI configuration as well. Our guide to [evaluating CVs](/en/hr-software/bewerberauswahl) explains how to formulate criteria properly. ### Works council * **§ 87 Abs. 1 Nr. 6 BetrVG:** Co-determination regarding technical equipment designed to monitor employee performance or conduct. Potential capability is sufficient. An AI tool also records the work of recruiters and is therefore routinely subject to co-determination. * **§ 90 Abs. 1 Nr. 3 BetrVG:** Right to information during the planning stage, explicitly including the deployment of AI. * **§ 95 Abs. 2a BetrVG:** Selection guidelines require works council approval, even when AI is involved. * **§ 80 Abs. 3 BetrVG:** If the works council needs to assess AI, an external expert is deemed necessary. Learn more in the guide on the [works council and AI](/en/hr-software/betriebsrat-ki). ## How do I spot AI-written applications? The honest answer: you cannot do so reliably. * **Detectors make mistakes.** A 2023 study tested 14 tools for detecting AI-generated text. None performed accurately and reliably enough (Weber-Wulff et al., 2023). Another study found that such tools disproportionately misclassify texts written by non-native speakers as AI-generated (Liang et al., 2023). Filtering candidates on this basis risks unlawful discrimination on grounds of origin. * **Clues are not proof.** Generic phrasing without examples, a cover letter that does not match the CV, or near-identical texts across multiple applications stand out. However, this also happens when human applicants use templates. * **The better question: is the content accurate?** Using AI to assist writing is not prohibited. The Bundesagentur für Arbeit (Federal Employment Agency) presents it to applicants as a helpful tool. What matters is whether the statements are true and the person has the skills they claim. * **What helps.** Place less weight on cover letters, ask about concrete situations during interviews, use work samples and technical questions, and verify credentials. The situation is different with hidden prompts in a CV, such as white text. That is not a writing style, but an attempt at manipulation. Robust systems detect and flag this. ## Checklist: Selecting an AI tool for recruiting 1. **Purpose and classification.** What is the tool intended for, and how does the vendor classify it under the AI Act? For scoring or filtering, how are they preparing the conformity assessment by 2 December 2027? 2. **Criteria.** Who defines the evaluation criteria? Are they reviewed, approved, and versioned by humans? 3. **AGG.** Does the system warn against criteria referencing protected characteristics? Are name, photo, and date of birth redacted prior to evaluation? 4. **Explanation.** Does every assessment include a justification with evidence cited from the CV? 5. **Human decision.** Is it technically impossible for the system to reject, filter out, or change statuses autonomously? Can criteria be overridden individually? 6. **Audit trail.** Are AI prompts and human interventions logged and retained in an exportable format for at least six months? 7. **Tamper protection.** Does the tool detect hidden text and prompt injections? 8. **Data flow.** Where does the model run? Are CVs sent to external AI services or used for training? Is there a Data Processing Agreement (DPA) and a list of sub-processors? 9. **Testing and audits.** How does the vendor test for accuracy and potential bias, and what does the instruction manual state regarding this? 10. **Documentation and inspection.** Are instructions for use, applicant notices, and documentation for DPIAs and works councils provided? Can the works council inspect criteria and logs? If you are planning the overall adoption of AI across HR, our consultancy services on [AI in HR](/en/hr-beratung/strategie-und-planung/ki-im-hr) and [AI strategy](/en/it-beratung/it-strategie-und-steuerung/ki-strategie) can support you. ## How techport HR Suite supports you In the [techport HR Suite](/en/hr-software), AI applicant scoring is a dedicated module alongside recruitment. It is designed to meet the requirements of the EU AI Act: * The AI proposes criteria based on the job advert, and a human reviews and approves them. Each approval creates an immutable version. * AGG-compliant criteria: The system warns against characteristics such as age, gender, or mother tongue; approval requires explicit confirmation. Personal characteristics such as name, photo, date of birth, and religion are redacted prior to evaluation. * Invisible text and hidden prompts are detected, removed, and flagged. * Every assessment comes with a score, calculation breakdown, and verbatim quotes verified against the CV. Same input, same result. * The AI suggests, a human decides. The system never sends automated rejections or modifies applicant statuses. Overrides are supported per criterion, requiring a mandatory justification. * Every AI query and every human action is logged; criteria and logs can be exported. Instructions for use and a template for 'Applicant Information' are included. * The Works Council Assistant provides the works council with read-only access to criteria, versions, and audit logs. The language model runs on the servers of the deployment; CVs are never sent to external AI services. In our assessment, AI applicant scoring constitutes a high-risk AI system under Annex III no. 4 of the AI Act. We are preparing the conformity assessment ahead of 2 December 2027. ## Frequently asked questions Ist KI im Recruiting erlaubt? Yes, AI in recruiting is permitted, but not without rules. For instance, emotion recognition has been prohibited since February 2025, including during recruitment processes. Systems that evaluate or filter applications are classified as high-risk AI and are subject to specific obligations from 2 December 2027. In addition, DSGVO (GDPR), AGG, and works council co-determination apply. Darf eine KI Bewerbern automatisch absagen? Generally, no. Art. 22 DSGVO strictly prohibits decisions based solely on automated processing that significantly affect individuals, and a rejection falls into this category. Under CJEU case law, even a score can constitute such a decision if it plays a decisive role. A human must therefore conduct a genuine review, rather than simply rubber-stamping. Wie erkenne ich KI-Bewerbungen? You cannot do so reliably. Studies show that AI text detection tools frequently make errors and disproportionately misclassify texts written by non-native speakers. Generic phrasing lacking concrete examples is an indicator, but not proof. It is far more sensible to examine the substance: ask about specific situations in interviews, request work samples, and verify references. Muss der Betriebsrat bei KI im Recruiting zustimmen? In most cases, yes. An AI tool in recruitment qualifies as technical equipment under § 87 Abs. 1 Nr. 6 BetrVG because it also tracks the work of recruiters. The works council must be informed during the planning stage under § 90 BetrVG, and selection guidelines require its approval under § 95 BetrVG. The council may also engage an external expert. Müssen Bewerber informiert werden, wenn KI eingesetzt wird? Yes. Under Art. 13 DSGVO, applicants must be informed about how their data is processed, including the logic involved in automated decision-making. From 2 December 2027, Art. 26(11) of the AI Act additionally requires applicants to be informed about the deployment of a high-risk AI system. Chatbots must already identify themselves as AI today. ## Sources * Verordnung (EU) 2024/1689 (KI-Verordnung), Art. 5, 6, 14, 26, 50, Anhang III: [https://eur-lex.europa.eu/eli/reg/2024/1689/oj](https://eur-lex.europa.eu/eli/reg/2024/1689/oj) (abgerufen am 23.10.2026) * Verordnung (EU) 2026/1744 zur Änderung der KI-Verordnung: [https://eur-lex.europa.eu/eli/reg/2026/1744/oj](https://eur-lex.europa.eu/eli/reg/2026/1744/oj) (abgerufen am 23.10.2026) * Europäische Kommission, Leitlinien zu verbotenen KI-Praktiken (4.2.2025): [https://digital-strategy.ec.europa.eu/en/library/commission-publishes-guidelines-prohibited-artificial-intelligence-ai-practices-defined-ai-act](https://digital-strategy.ec.europa.eu/en/library/commission-publishes-guidelines-prohibited-artificial-intelligence-ai-practices-defined-ai-act) (abgerufen am 23.10.2026) * Europäische Kommission, Entwurf der Leitlinien zur Einstufung von Hochrisiko-KI-Systemen (19.5.2026): [https://digital-strategy.ec.europa.eu/en/library/draft-commission-guidelines-classification-high-risk-ai-systems](https://digital-strategy.ec.europa.eu/en/library/draft-commission-guidelines-classification-high-risk-ai-systems) (abgerufen am 23.10.2026) * KI-Marktüberwachungs- und Innovationsförderungsgesetz (KI-MIG), § 2: [https://www.gesetze-im-internet.de/ki-mig/BJNR0DF0B0026.html](https://www.gesetze-im-internet.de/ki-mig/BJNR0DF0B0026.html) (abgerufen am 23.10.2026) * Bundesnetzagentur, Pressemitteilung zur KI-Verordnung (29.7.2026): [https://www.bundesnetzagentur.de/SharedDocs/Pressemitteilungen/DE/2026/20260729\_KI\_VO.html](https://www.bundesnetzagentur.de/SharedDocs/Pressemitteilungen/DE/2026/20260729_KI_VO.html) (abgerufen am 23.10.2026) * Verordnung (EU) 2016/679 (DSGVO), Art. 13, 22, 35: [https://eur-lex.europa.eu/eli/reg/2016/679/oj](https://eur-lex.europa.eu/eli/reg/2016/679/oj) (abgerufen am 23.10.2026) * EuGH, Urteil vom 7.12.2023, C-634/21 (SCHUFA): [https://curia.europa.eu/juris/liste.jsf?language=de&num=C-634/21](https://curia.europa.eu/juris/liste.jsf?language=de&num=C-634/21) (abgerufen am 23.10.2026) * EuGH, Urteil vom 27.2.2025, C-203/22: [https://curia.europa.eu/juris/liste.jsf?language=de&num=C-203/22](https://curia.europa.eu/juris/liste.jsf?language=de&num=C-203/22) (abgerufen am 23.10.2026) * Betriebsverfassungsgesetz, §§ 80, 87, 90, 95: [https://www.gesetze-im-internet.de/betrvg/](https://www.gesetze-im-internet.de/betrvg/) (abgerufen am 23.10.2026) * Reuters, Bericht über ein eingestelltes KI-Bewertungssystem bei Amazon (10.10.2018): [https://www.reuters.com/article/us-amazon-com-jobs-automation-insight-idUSKCN1MK08G](https://www.reuters.com/article/us-amazon-com-jobs-automation-insight-idUSKCN1MK08G) (abgerufen am 23.10.2026) * Weber-Wulff u. a., Testing of detection tools for AI-generated text, International Journal for Educational Integrity 19 (2023): [https://doi.org/10.1007/s40979-023-00146-z](https://doi.org/10.1007/s40979-023-00146-z) (abgerufen am 23.10.2026) * Liang u. a., Studie zu Fehlklassifikationen von KI-Textdetektoren bei Nicht-Muttersprachlern, Patterns 4 (2023): [https://doi.org/10.1016/j.patter.2023.100779](https://doi.org/10.1016/j.patter.2023.100779) (abgerufen am 23.10.2026) * Bundesagentur für Arbeit, Bewerben mit KI: [https://www.arbeitsagentur.de/bildung/bewerbung/bewerben-mit-ki](https://www.arbeitsagentur.de/bildung/bewerbung/bewerben-mit-ki) (abgerufen am 23.10.2026) ## Further reading [techport HR Suite HR Software HR software with personnel administration, recruiting, AI candidate evaluation and company knowledge on one platform. Hosted in Germany. Request a demo now. ](/en/hr-software)[techport HR Suite Digital Personnel File Digital personnel file with master data, contracts and documents, role-based permissions and access logging. Part of the techport HR Suite. Request a demo now. ](/en/hr-software/digitale-personalakte)[techport HR Suite Leave Management Leave management with requests, remaining leave, team calendar, and sickness notifications. Managers approve in the browser, HR keeps track. Request a demo. ](/en/hr-software/abwesenheitsverwaltung)[techport HR Suite Onboarding and Offboarding Onboarding software starting right from offer acceptance: personnel files and tasks for HR, IT, and managers are created automatically. Including offboarding. Request demo. ](/en/hr-software/onboarding) More from techport.ai [ Software Custom process software for mid-sized companies. ](/en/loesungen)[ IT consulting Strategy, architecture, operations and security. ](/en/it-beratung)[ IT maturity check Ten minutes to a clear position. ](/en/it-beratung/reifegrad-check)[ HR maturity check 24 statements, a result per field. ](/en/hr-beratung/reifegrad-check)[ Funding BAFA grant plus more than 50 programmes for delivery. ](/en/foerderung)[ Process in practice How workflows become reliable software. ](/en/sop-praxis)[ Data and AI Analysis, forecasts and assistance systems. ](/en/daten-ki)[ About us The people behind techport.ai. ](/en/ueber-uns)