For several years now, artificial intelligence has dominated HR conferences and vendor pitches. The available data tells a more measured story. Adoption is genuinely progressing, but it remains limited and concentrated in a narrow band of tasks. At the same time, AI is already reshaping the work of HR teams in organisations that have never deployed it, through the effects it produces on the candidate side and through the legal framework taking shape around it.
A Persistent Gap Between Narrative and Practice
Within HR functions, 39% of professionals surveyed report having already deployed some form of AI, and a further 7% expect to do so before the end of 2026. More than half of organisations have done nothing and have no plans to. Company size remains the strongest determinant here, with 60% adoption among large organisations against a third in smaller structures (SHRM, 2026).
That gap produces a notable side effect. Around a third of HR functions already equipped believe they are lagging behind their peers, when they are in fact ahead of the 31% of organisations that have made no move at all. The misperception fuels defensive purchasing, driven by fear of falling behind rather than by an identified need.
Individual Time Savings, Limited Return
Where AI is used across HR varies sharply from one discipline to the next. Recruitment leads by some margin (27%), ahead of HR information systems (21%), learning and development (17%) and employee experience (14%), while inclusion and compliance register usage rates below 2% (SHRM, 2026).
In practice, adoption remains firmly operational and aimed at the time-consuming end of the workload, namely parsing information from applications, scheduling interviews and drafting job advertisements. More advanced use cases exist, but they remain marginal.
HR professionals using these tools report tangible effects, with 87% observing an improvement in their efficiency and 75% in the quality of their work, though half see no improvement in the quality of their decisions (SHRM, 2026).
AI is not yet redefining how organisations decide. It is absorbing the administrative load that precedes the decision.
At organisational level, the picture is markedly harsher: 88% of HR leaders report that they have yet to derive significant economic value from these tools. The paradox is plain. Employees may gain up to an hour and a half a day through AI, yet only 7% of organisations set any expectation for how that time should be reinvested (Gartner, 2025). Without clear guidance, individual productivity dissipates into the working day instead of converting into collective performance. That missing link is precisely where the distance lies between what the tools promise and what shows up in the accounts.
AI in Candidates' Hands: Fraud and the Return of the In-Person Interview
The most immediate effect of AI on HR comes not from the tools companies buy, but from those candidates use. By the end of 2024, roughly four candidates in ten reported turning to AI during their application, chiefly to produce a CV, a covering letter or written answers. Recruiters are consequently buried under applications that all look alike, and the more files arrive, the less each one tells them.
Identity has now joined competence as an open question, with 6% of candidates surveyed in mid-2025 admitting to having cheated in an interview, either impersonating someone else or having someone stand in for them. Gartner forecasts that by 2028, a quarter of candidates on the labour market will be fake profiles.
To counter this fraud, the in-person interview is returning as a verification mechanism, and candidates are not resisting it: 62% say they are more likely to apply when an employer builds face-to-face interviews into the process. Automating the front end of recruitment makes human validation more necessary than ever.
What AI Does to Assessment
Using artificial intelligence to screen applications carries a substantial risk of bias. A university study presented in late 2024 tested three large language models against more than 500 real CVs in which only the first names had been altered. The models favoured names associated with white individuals in roughly 85% of cases and female-associated names in 11% of cases, and never preferred a name associated with a Black man over its white equivalent (K. Wilson, A. Caliskan, 2024). These were general-purpose models rather than dedicated HR software, but the finding restates an obvious point: algorithmic neutrality is never a given.
Embedding AI in the selection process also alters candidate behaviour directly. When people know they are being assessed by a machine, they instinctively adjust their posture, foregrounding their analytical traits and playing down their intuitive ones, on the assumption that the machine rewards the former. These self-censoring biases fade once the actual workings of the tool are explained to them. Transparent communication about what the AI genuinely measures thus changes the quality of responses considerably (J. Goergen, E. De Bellis, 2025).
A Shifting Regulatory Timetable
The European regulatory framework is undergoing a recalibration of its timetable. By classifying AI systems applied to recruitment, promotion and employee evaluation as high-risk, the European AI Act set a compliance date of 2 August 2026. The simplification regulation that came into force on 27 July 2026 defers those obligations to 2 December 2027, although transparency requirements remain applicable from August 2026.
This shift gives employers operational breathing room, not an exemption. The substantive obligations are unchanged, and putting the governance in place (evaluation documentation, human oversight, traceability of criteria) takes long enough to justify using the transition period from the outset.
Decide Before You Buy
In more than half of organisations, the HR function is involved in the company's AI strategy neither directly nor through cross-functional working, despite being the function most exposed to the consequences of deployment.
Organisations that get results proceed differently, and their method is unremarkable. They start from an identified point of friction rather than from an available tool, define before deployment the indicator that will settle the question, make a deliberate call on how freed-up time is reallocated, and document their assessment criteria. The reason most often cited by those not using AI is neither cost nor mistrust, but a lack of awareness of what these tools actually do. That barrier is also the easiest to remove, and it argues for prioritising internal literacy and training ahead of any software purchase.
Key Takeaways
- Real adoption remains limited : Fewer than one HR function in two expects to use AI in 2026, and the gap between large and small organisations remains wide.
- Deployed use cases are transactional : Recruitment concentrates the bulk of applications, on screening, scheduling and drafting rather than on the decision itself.
- Time savings only become value when reallocated, and they can only be demonstrated if they are measured, which more than half of organisations do not do.
- Disclosing the use of AI changes how candidates behave : Disclosure is both required and desirable, but it calls for an assessment method that does not rest on written self-reporting alone.
- The dividing line is validation, not automation : A tool proven against the role in question beats managerial intuition, while the reverse holds for a tool deployed without controls.
- The European deferral is a preparation window : High-risk system obligations will apply from December 2027, and the expected documentation takes months to build.