AI in Recruitment: Where It Helps and Where It Bites
Short answer: AI in recruitment works best on text and logistics: drafting job ads, answering candidate questions, scheduling interviews and summarising CVs. Once a tool scores, ranks or rejects people, legal exposure jumps. NYC requires bias audits, GDPR Article 22 restricts fully automated rejections, and the EU AI Act treats screening tools as high-risk from December 2027.
Back in 2018, Reuters reported that Amazon had scrapped an experimental tool built to rate CVs. It had learned to downgrade applications containing the word “women’s”, because it was trained on a decade of hiring decisions that skewed heavily male. The story still circulates for a good reason. It captures the core problem in one sentence: a model trained on past decisions learns the bad ones too.
That is no argument for keeping AI out of hiring altogether. The question is where in the process it sits.
How widely is AI used in hiring?
Less than vendor websites suggest, at least in representative data. One of the few large surveys with a clean method comes from Germany. Bitkom Research asked 602 companies in spring 2026, and 12% said they used AI anywhere in the application process. Individual uses were smaller still: 9% used it to write job ads, 6% for a first pass over incoming applications, and 2% had AI conducting interviews. In the previous survey, carried out in autumn 2024, only 1% screened applications with AI.
Source: Bitkom Research, survey of 602 companies, published 28 September 2026
Germany is a cautious market, so the US and UK numbers are probably higher. But the shape of the chart tends to hold: AI is used most where the legal stakes are lowest.
Where AI genuinely saves time
The most useful applications share one feature. They prepare work. They do not decide.
Drafting job ads
Give a language model bullet points about the role and you get a first draft in two minutes. It is the most common use and legally the easiest, provided nobody pastes candidate data into a public tool. Someone still has to read the result. Models are fond of “digital natives” and “young, dynamic teams”, both of which can look like age bias. What to check is in how to write a job ad.
Answering candidate questions
Put a chatbot on the careers page and it can explain the hiring timeline, the hybrid-work policy or the visa situation without touching anyone’s personal data. Since August 2026, EU rules require such a bot to say it is an AI unless that is obvious.
Scheduling
Coordinating three calendars for one interview is work nobody misses. Plain automation handles it well, no AI needed.
Summarising CVs
Extracting jobs, degrees and languages from a four-page CV saves reading time. The line is crossed when the summary includes a judgement or orders candidates by fit.
Where it gets legally risky
Scoring people triggers several rulebooks at once, and they differ by jurisdiction.
United States
Federal anti-discrimination law covers outcomes, not methods. If a screening tool rejects older applicants or women at a higher rate, the employer is exposed, whoever wrote the code. New York City went further. Local Law 144, enforced since 5 July 2023, requires an independent bias audit of automated employment decision tools within the year before use, a public summary of the results and notice to candidates at least ten business days in advance. Illinois has required notice and consent before AI analyses video interviews since 2020.
United Kingdom
UK GDPR restricts decisions based solely on automated processing that significantly affect people, and a rejection qualifies. The Data (Use and Access) Act 2025 loosened those rules somewhat, but kept safeguards such as the right to human intervention. The ICO published the results of audits of AI recruitment tool providers in November 2024 and asked for clearer information to candidates and better bias testing.
European Union
Article 22 GDPR applies, and the Court of Justice took a broad view in its SCHUFA judgment of 7 December 2023 (C-634/21): a score can itself count as an automated decision if the human downstream mostly just follows it. Translated to hiring, a ranked list that a recruiter works through top to bottom is not much of a human decision. On top of that, the AI Act classes systems that filter applications or evaluate candidates as high-risk. Employers using them face obligations from 2 December 2027. The details are in our article on the EU AI Act in recruitment.
| Criterion | Candidate data involved | EU AI Act high-risk | NYC bias audit |
|---|---|---|---|
| Drafting job ads | |||
| FAQ chatbot without login | |||
| Summarising CVs | partly | partly | |
| Ranking or filtering applicants | |||
| Emotion analysis in video interviews | banned |
“Partly” for CV summaries is deliberate. The AI Act exempts systems that only perform a narrow procedural task, but never those that profile people, and NYC’s rule depends on whether the output substantially assists the hiring decision. Where a summary nudges decisions, assume the stricter reading.
Video interviews and the emotion ban
Asynchronous video interviews, where candidates record answers, are a legitimate format and often easier for shift workers than a weekday morning slot. What the EU banned in February 2025 is inferring emotions from faces or voices at work. HireVue, one of the best-known vendors, dropped facial analysis from its product back in 2021. Still check your settings and contract for any sentiment or personality analysis. For prohibited practices, fines under the AI Act reach €35 million or 7% of worldwide turnover, whichever is higher.
How to start safely
Begin where no candidate data flows. Piloting AI drafts for job ads costs a few hours and shows quickly whether the team likes working that way.
If you then decide AI should touch candidate data, go in this order:
- Run a data protection impact assessment if you hire in the EU or UK. For automated candidate scoring it is almost always required.
- In Germany, involve the works council before signing anything. See works councils and IT systems.
- For NYC roles, commission the bias audit and publish the summary before go-live.
- Get the vendor to confirm in writing how they will meet EU high-risk obligations by December 2027.
- Decide who reviews AI suggestions, with what authority, and how reviews are logged.
- Update the candidate privacy notice.
Applications should sit in a proper system rather than in inboxes. Our guide to choosing an applicant tracking system for small business covers what to look for.
My take
For most companies handling fewer than a few hundred applications a year, AI screening isn’t worth it. The time saved is modest, the review burden is real, and you carry the risk in any discrimination claim. High-volume hiring for standardised roles, such as warehouse or retail staff, is different. There, carefully tested pre-sorting can make sense if a human genuinely owns every rejection.
If you want to connect an AI component to existing HR systems instead of buying a full suite, a studio such as OnlineBuilders can help with architecture and build.
Your next step: take your next open role, have an AI draft the ad and compare it with your current version. You will see quickly where the machine helps and where you need to sharpen it. You will find related topics in the recruiting section.
Questions people ask
Can an AI legally reject job applicants?
Under GDPR and UK GDPR, a rejection based solely on automated processing is heavily restricted and requires safeguards such as meaningful human review. In the US there is no general ban, but discrimination law applies to the outcome regardless of whether a person or a tool made the call.
Is it fine to use ChatGPT to write job descriptions?
Yes, as long as you do not paste candidate data into it. Read the draft before publishing, because language models like phrases such as 'young, energetic team' that can signal age bias.
Do I have to tell candidates I use AI?
Often, yes. GDPR requires you to explain automated decision-making in your privacy notice, NYC Local Law 144 requires notice at least ten business days before an automated tool is used, and Illinois requires notice and consent before AI analyses video interviews.
Are emotion-reading video interviews allowed?
Not in the EU. Since 2 February 2025 the AI Act bans emotion recognition in the workplace, with narrow medical and safety exceptions. That covers tools that infer feelings from facial expressions or tone of voice in interviews.
