AI screening

What the system contributes to screening

An AI tool can extract experience from a CV, summarise an application or estimate how closely it matches specified requirements. These outputs serve different purposes. A summary helps someone read; a score or ranking can influence who receives an interview. Reviewing CVs without a model is covered under resume screening.

For an illustrative example, a model might identify experience with a related programming language even when the CV does not use the wording in the job description. The reviewer still needs to check whether that experience meets the actual requirement. Similar wording, a plausible summary and demonstrated ability are different kinds of evidence.

A fixed rule can automate screening without AI

A knockout question applies an explicit condition to an answer. It does not need a model to infer suitability. Join’s documented question settings, for example, distinguish optional, required and knockout questions. A disqualifying knockout answer marks the application as rejected; the email follows the next day unless the team withdraws the rejection first.

That is a specific form of automation. It does not establish that a product also offers AI candidate rankings or explainable model scores. Check the behaviour of the particular feature being considered.

Check the evidence behind an output

Decide what task the tool should support before testing it. Inspect original applications alongside its output, including cases it scores poorly. Check for omitted experience, invented qualifications and inconsistent treatment of equivalent evidence. A fast review is not useful if the summary silently removes the reason to interview someone.

Ask the vendor what the tool infers as well as what it reads. When the UK Information Commissioner’s Office audited AI recruitment tools in 2024, it found tools inferring gender and ethnicity from a candidate’s name, and filters that let recruiters exclude people with certain protected characteristics. A person approving a score without examining its basis carries such an error into the decision.

Which rules apply depends on where you hire

European Union. Annex III of the EU AI Act lists AI systems intended to analyse and filter job applications and to evaluate candidates. They count as high-risk unless a narrow exception in Article 6(3) applies, and a system that profiles people is always high-risk. Regulation (EU) 2026/1744 moved the start of these high-risk obligations to 2 December 2027. From then, an employer using such a system must inform workers’ representatives and affected workers before using it at the workplace (Article 26(7)) and tell the people it assesses that it is being used (Article 26(11)). Under Article 111(2), a system already placed on the market or put into service before that date is covered only if its design changes significantly afterwards. The AI Act glossary entry covers the other deployer duties. GDPR obligations do not wait for that date, including Article 22 on solely automated decisions.

United Kingdom. Since 5 February 2026, the UK GDPR has defined a decision as solely automated when there is no meaningful human involvement (Article 22A). Such a decision with a legal or similarly significant effect needs safeguards: information about the decision, a way to make representations, human intervention and the right to contest it (Article 22C). The ICO’s March 2026 guidance adds that employers should tell candidates when automated decision-making is used and test regularly for biased outputs.

New York City. Local Law 144 bars employers and employment agencies from using an automated employment decision tool unless it has had a bias audit within the past year, information about that audit is public and candidates or employees have been notified. The city has enforced it since 5 July 2023. It is a city rule, not a US-wide one.