Short answer: Candidate perception of AI in hiring is cautious rather than hostile. Independent surveys in the US and the EU find majorities negative about AI selecting applicants or making the final call, and experiments show algorithm only evaluation feels less fair than a recruiter's. Employers control what moves trust: disclosure, a clear reason and a person who decides.
Key takeaways
- The objection is specific. In a Pew Research Center survey of US adults in December 2022, 71% opposed AI making the final hiring decision and 7% were in favour; 41% opposed AI reviewing applications.
- Across the EU in 2024, 57% of respondents view the use of digital technologies, including AI, to select job applicants negatively, and 74% say prohibiting fully automated decision making at work would be important (Special Eurobarometer 554).
- Controlled studies point the same way: when people are told an algorithm rather than a recruiter evaluated them, they rate the process as less fair, and that makes the employer and the job less attractive to them.
- More information is not automatically better. One experiment found that explaining why a step exists worked better than describing how it works, which could even make reactions worse.
- We found no independent survey of Dutch frontline applicants on this question. Treat every headline figure as a prompt to measure your own process.
On this page: Do candidates trust AI in hiring? · Why do candidate surveys disagree so much? · What does research say about how candidates react to AI evaluation? · Does telling candidates more about the AI build trust? · What does this mean for frontline hiring in the Netherlands? · What should employers do about candidate distrust of AI? · What does the law require you to tell candidates about AI? · What are the limits of the evidence on candidate perception of AI in hiring? · How Radius helps · Frequently asked questions
Do candidates trust AI in hiring?
Mostly not when AI decides; opposition is lower when it assists.
The most detailed independent US data comes from the Pew Research Center, which surveyed 11,004 US adults from 12 to 18 December 2022 and published the results on 20 April 2023. Three findings matter for employers:
- The final decision is the line. 71% opposed AI making the final hiring decision; 7% favoured it.
- Assistance is less contested. A smaller plurality, 41%, opposed AI reviewing job applications.
- It changes behaviour, at least in intention. 66% said they would not want to apply for a job with an employer that uses AI to help make hiring decisions; 32% said they would.
US adults against AI in hiring, by question
Pew Research Center survey of 11,004 US adults, 12 to 18 December 2022
People also see an upside. 47% thought AI would do better than humans at treating all job applicants in the same way, against 15% who thought it would do worse, although more expected AI to do worse than better at seeing potential in applicants who do not perfectly fit the job description.
European data tells a similar story. Special Eurobarometer 554, which the Commission published in February 2025, interviewed 26,415 people across the 27 member states between 25 April and 22 May 2024. On using digital technologies, including AI, to select applicants for a job, 36% were positive and 57% negative. 74% said prohibiting fully automated decision making at work would be important, 75% said the same of more transparency in digital HR decision making, and 77% of involving workers and their representatives in adopting new technologies. On transparency, the Netherlands gave the highest share in the EU: 87%.
What 26,415 people across the EU said, April to May 2024
Read together, these surveys describe a public that accepts help from software but wants a person to own the decision and wants to be told what is going on. If you are introducing an AI interview, see how the AI Interviewer handles disclosure and review on a frontline role.
Why do candidate surveys disagree so much?
Because they ask different people different questions about different kinds of AI.
Headlines point both ways, and both can come from honest data. Check five things before a figure shapes your process.
| Check | Why it changes the answer | What the independent surveys above did |
|---|---|---|
| Who was asked? | The general public, recent applicants and people who have just finished an AI interview answer differently. People who completed a step are self selected; those who dropped out are missing. | Pew and Eurobarometer asked the adult population, not applicants: broad, but not specific to candidates. |
| What did the AI do in the question? | "Make the final decision", "review applications" and "schedule interviews" are different things. Pew's results move from 71% to 41% opposition depending on the task. | Both name the task. Neither asked about a structured interview with a human decision. |
| Who paid for it? | A survey run or commissioned by a company that sells hiring software has a commercial interest in the result. | Pew is an independent research centre; Eurobarometer is commissioned by the European Commission and fielded by a research agency. |
| Is the method published? | Without dates, sample, weighting and question wording, a percentage cannot be checked or compared. | Both publish methodology, fieldwork dates and questionnaires. |
| Where and when? | Attitudes differ by country and shift as people meet AI in daily life. | Pew: US, December 2022. Eurobarometer: EU27, April to May 2024. |
Use the table on any survey that lands on your desk, including ours if we ever publish one, and treat a figure that fails two or more checks as a talking point.
What does research say about how candidates react to AI evaluation?
Controlled studies consistently find that AI only evaluation feels less fair.
Experiments are better at showing cause, at the price of being artificial. Four peer reviewed studies stand out.
- Automated interviews and stakes. Langer, König and Papathanasiou (2019) showed 123 participants either a highly automated interview or an equivalent videoconference interview, for selection or for training. Automated interviews lowered acceptance through lower social presence and fairness, and in the selection setting they also created more ambiguity and less sense of control.
- Algorithm or recruiter. In two studies with 172 and 276 participants, Oostrom and colleagues (2024), a team of Dutch university researchers, told people that either an algorithm or a recruiter had evaluated their video interview and personality questionnaire. The algorithm condition lowered perceived fairness and how well people thought the assessment predicts performance, and through lower fairness it reduced their attraction to the organisation and their intention to accept a job. A positive outcome did not cancel the effect.
- Algorithm only or algorithm assisted. Across four studies, Lavanchy and colleagues (2023) found that people rated algorithm driven hiring as less fair than a process run by humans alone or by humans assisted by an algorithm, whether or not the outcome was in their favour.
- The devil you know. In 33 interviews with professionals, Mirowska and Mesnet (2022) found that people acknowledged AI evaluation could be more objective but still wanted human elements kept, preferring familiar human judgement to an unfamiliar algorithm.
The pattern lines up with the surveys: the more the software acts as the judge, the less fair the process feels, especially when a job is at stake.
Does telling candidates more about the AI build trust?
Not automatically: the reason matters more than the mechanics.
In an experiment with 124 participants, Langer and colleagues (2021) gave people information about how an automated interview works, a justification for why it is used, both or neither. Justification worked better than process information, and process information could even make reactions worse. No version made the organisation more attractive than giving no information at all.
The practical reading: tell candidates plainly that a step is automated, why you use it and who decides. A short, honest reason ("so every applicant gets the same questions and a reply the same day") does more work than a paragraph on how speech is turned into text.
What does this mean for frontline hiring in the Netherlands?
The evidence gap is widest exactly where frontline employers and agencies hire.
None of the independent sources above isolate frontline applicants such as warehouse staff, drivers, cleaners, security officers or hospitality crews, and we did not find an independent survey of Dutch frontline applicants on AI in hiring. Do not assume a US or EU average describes the people who apply to your roles.
There are reasons to expect frontline candidates to weigh some things differently. These are reasoned expectations, not findings:
- Silence may be the benchmark. Where applicants apply on a phone and hear nothing back, a step that replies the same day may be judged against that silence rather than against a friendly recruiter.
- Language changes fairness. Candidates who are not native Dutch speakers may find an interview in a language they handle well fairer than a phone call they struggle to follow.
- Agencies add a layer. An applicant may not know whether the agency or the client decides. Distrust of AI can become distrust of an unclear process.
What should employers do about candidate distrust of AI?
Act on four things: who decides, disclosure, the reason and a person.
| What candidates object to | What the evidence says | What you can do | How to check it worked |
|---|---|---|---|
| AI making the decision | 71% oppose AI making the final hiring decision (Pew); algorithm only processes are judged less fair than algorithm assisted ones (Lavanchy and colleagues). | Keep the decision with a named role. Never let a tool reject a candidate on its own, and say so in the candidate message. | Audit a sample of rejections each month: who made each one, and on what evidence? |
| Not knowing AI is used | Three quarters of EU respondents rate transparency in digital HR decisions as important; 87% in the Netherlands (Eurobarometer). | Tell candidates at the start of the step, in plain words, that it is automated and what happens to their answers. | Compare how many candidates start and finish the step before and after you change the wording. |
| Not knowing why | Justifications improved reactions; process detail alone did not (Langer and colleagues, 2021). | Give one sentence of purpose: what the step is for and what it means for the candidate. | Read the questions candidates send. Fewer "what is this?" questions is a good sign. |
| No person to turn to | Interviewees wanted human elements kept in evaluation (Mirowska and Mesnet); 74% of EU respondents rate prohibiting fully automated decisions at work as important. | Name a contact route for questions and for a review of the outcome. | Record review requests and how long a reply takes. |
Underneath the table sit two habits: evaluate only what the job needs, and tell candidates what happens next and when. Distrust grows in the gap between steps.
Illustrative example: a disclosure message for a frontline applicant
Adapt the wording to your process and check it against your privacy notice.
Measure trust in your own funnel
Four measures, per role and per channel, tell you most of what your applicants think:
- Start rate. Of the candidates invited to the AI step, how many start it?
- Completion rate. Of those who start, how many finish?
- Questions and objections. What do candidates ask or complain about, in their own words?
- Review requests. How many ask for a person to look at their outcome, and what happens next?
Change one thing at a time, such as the disclosure wording, and compare the same measures before and after. Two short questions after the step ("Was it clear what this step was for?" and "Do you know who decides?") add the candidate's own view.
What does the law require you to tell candidates about AI?
EU law already requires some disclosure, and more applies from 2 December 2027.
- AI that talks to people. Since 2 August 2026, Article 50(1) of the AI Act has required providers to design AI systems that interact directly with people so that those people are told they are dealing with AI, unless that is obvious from the context. For a candidate facing interviewer or chatbot, the safe practice is to say so clearly at the start.
- Information about data. Under Article 13 of the GDPR, candidates must be told who processes their data, why, on what legal basis, who receives it, how long it is kept and what their rights are. If solely automated decisions with significant effects are used, they must also get meaningful information about the logic involved and the likely consequences.
- Solely automated decisions. Article 22 of the GDPR gives candidates the right not to be subject to a decision based solely on automated processing that significantly affects them, such as an automatic rejection, unless an exception applies. Where an employer relies on the contract or explicit consent exception, candidates must at least be able to get human intervention, give their view and contest the decision.
- High risk hiring systems. From 2 December 2027, the date set by the Digital Omnibus on AI (Regulation (EU) 2026/1744), deployers of high risk AI systems listed in Annex III, which include systems used to filter applications and evaluate candidates, must tell the people they make or help make decisions about that such a system is being used (Article 26(11) AI Act).
- Emotion recognition. Since 2 February 2025 the AI Act has banned AI systems that infer people's emotions from biometric data, such as facial expressions or voice, in the workplace, except for medical or safety reasons. The European Commission's guidelines on prohibited practices, adopted in July 2025 and not binding, say this ban also covers candidates during recruitment.
The law sets a floor; the surveys suggest candidates expect more. Our overview of the EU AI Act for recruitment covers the deployer duties in more depth.
What are the limits of the evidence on candidate perception of AI in hiring?
The evidence is consistent in direction but thin for frontline candidates in Europe.
- Population. Pew asked US adults and Eurobarometer asked EU citizens aged 15 and over living in the member states. Neither asked applicants as such, and neither isolated frontline roles.
- Age of the data. Fieldwork dates from December 2022 (Pew) and spring 2024 (Eurobarometer).
- Intentions, not behaviour. We found no independent field study that measures how disclosure changes real application or completion rates in frontline hiring.
- Artificial settings. The experiments used videos, scenarios and study participants. They show direction, not the size of the effect you will see.
- Different AI. Most studies describe AI that evaluates candidates. Few separate AI that collects structured answers from AI that judges them.
How Radius helps
Radius Hire is the AI hiring platform for frontline roles. It is built around the conditions the evidence links to trust: candidates are told, a person decides, and a decision can be explained later.
- Candidates are told at the start that the AI Interviewer is automated and that a person reviews the result. It is a short, structured voice interview that candidates can take on their phone, whenever it suits them, in the language set for the role.
- For each requirement, Match shows whether it is backed by a document or a check, rests only on what the candidate said, or is still missing.
- Radius never rejects a candidate on its own. A recruiter reviews the evidence, can overrule anything Radius presents and makes every hiring decision, as we explain in why the tool should never be the one saying no. Each interview can be kept as a recording, a transcript and a summary.
- Candidate data is hosted in the EU and processed under a data processing agreement, and the employer sets how long it is kept. Details are on our trust page, and what candidates are told is on the candidate FAQ.
Radius is new, so this article explains how the product works rather than claiming results.
Frequently asked questions
Do candidates who have done an AI interview feel differently from those who have not?
Possibly, but the independent evidence is thin. Surveys of people who completed an AI step leave out everyone who dropped out, so they can overstate approval.
Do younger candidates trust AI in hiring more?
Somewhat, in the European data. Special Eurobarometer 554 (2024) found positive views of digital technologies at work most widespread among younger respondents for every activity tested, including selecting applicants.
Should candidates be able to reach a person during an AI step?
Yes, as good practice, because it answers the main concern in the research. Where an employer relies on the contract or explicit consent exception for a solely automated decision with significant effects, the GDPR requires that candidates can at least get human intervention and contest the decision.
Can AI read a candidate's emotions from their voice or face?
Not lawfully, on the European Commission's reading of the AI Act. Since 2 February 2025 the AI Act has banned AI systems that infer emotions from biometric data in the workplace, except for medical or safety reasons, and the Commission's guidelines, which are not binding, say this covers candidates.
How can we find out what our own applicants think about AI?
Measure your own funnel: start and completion rates for the AI step, the questions candidates send, review requests and two short questions after the step.
Conclusion
The independent evidence on candidate perception of AI in hiring is narrower than the headlines: people object to software that decides alone, in silence, for reasons nobody explains. Those are design choices. Tell candidates when a step is automated, give the reason in one sentence, keep a named person responsible for every decision and give candidates someone to ask. Then measure whether your own applicants start, finish and trust the step, because no survey of US adults or EU citizens can tell you that.
General information, not legal advice. Position as of 22 October 2026.
Sources
- Pew Research Center (20 April 2023). AI in Hiring and Evaluating Workers: What Americans Think. Survey of 11,004 US adults, 12 to 18 December 2022. Accessed 22 October 2026.
- European Commission, Directorate General for Employment, Social Affairs and Inclusion (2025). Special Eurobarometer 554: Artificial Intelligence and the future of work. Fieldwork 25 April to 22 May 2024, 26,415 respondents, EU27. Accessed 22 October 2026.
- European Commission (13 February 2025). Commission survey shows most Europeans support use of artificial intelligence in the workplace. Accessed 22 October 2026.
- Langer, M., König, C. J. and Papathanasiou, M. (2019). Highly automated job interviews: Acceptance under the influence of stakes. International Journal of Selection and Assessment, 27(3), 217 to 234.
- Oostrom, J. K., Holtrop, D., Koutsoumpis, A., van Breda, W., Ghassemi, S. and de Vries, R. E. (2024). Applicant reactions to algorithm- versus recruiter-based evaluations of an asynchronous video interview and a personality inventory. Journal of Occupational and Organizational Psychology, 97(1), 160 to 189. Published online 10 September 2023.
- Lavanchy, M., Reichert, P., Narayanan, J. and Savani, K. (2023). Applicants' Fairness Perceptions of Algorithm-Driven Hiring Procedures. Journal of Business Ethics, 188(1), 125 to 150.
- Mirowska, A. and Mesnet, L. (2022). Preferring the devil you know: Potential applicant reactions to artificial intelligence evaluation of interviews. Human Resource Management Journal, 32(2), 364 to 383.
- Langer, M., Baum, K., König, C. J., Hähne, V., Oster, D. and Speith, T. (2021). Spare me the details: How the type of information about automated interviews influences applicant reactions. International Journal of Selection and Assessment, 29(2), 154 to 169.
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), OJ L, 12 July 2024: Articles 5(1)(f), 26(11), 50(1) and Annex III point 4. Text checked 22 October 2026.
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), OJ L, 24 July 2026: high risk obligations for Annex III systems apply from 2 December 2027. Text checked 22 October 2026.
- Regulation (EU) 2016/679 (General Data Protection Regulation): Articles 13 and 22. Accessed 22 October 2026.
- European Commission (29 July 2025). Guidelines on prohibited artificial intelligence practices established by Regulation (EU) 2024/1689, C(2025) 5052 final. Content approved 4 February 2025; library page updated 31 July 2025. Not binding. Accessed 22 October 2026.