Automating job applications: what you may and may not do
Almost every recruitment AI tool promises the same thing: screen CVs and rank candidates. That is precisely the part you usually may not automate. And it is not where you lose your time either.
TerenceAutomating job applications is on a lot of owners' lists, and that makes sense. You post a vacancy, thirty replies come in, and nobody has structural time for them. The tools you find all promise the same thing: screen CVs, score candidates, put the best five on top. But that is exactly the part you usually may not leave to software. And it is not where your time goes either.
This article covers three things: what you are allowed to automate in your hiring process, where the line has been drawn since this summer, and where the real gain sits for a company with five to fifty people. No legal waffle, but with the rules included, because they changed again in July 2026.
Why automating job applications is on everyone's list right now
The Dutch labour market is still tight. At the end of the second quarter of 2026 there were 375 thousand open vacancies according to Statistics Netherlands, and for every 100 unemployed people there are 95 vacancies. In the province of Overijssel it is 128. So if you are hiring, you are not only competing with other employers for the candidate, you are competing with time.
open vacancies in the Netherlands (Statistics Netherlands, Q2 2026)
And there is the real problem. Research by HR platform Tellent among jobseekers in the Benelux shows that on average 41.5 percent of applicants drop out at the very first step of the application process. Not because they do not want the job, but because the form is too long, because they have to take an assessment first, because they have to upload a CV and then type all of it in again, or because it is completely unclear what happens next.
of applicants drop out at the first step (Tellent research, Benelux)
Look carefully at where that group falls away: not at the assessment, but at the form and at the silence that follows. That is a process problem, not a judgement problem. And it is exactly the part you are allowed to automate.
What AI is allowed to do with your job applications
There is a large, boring part of your hiring process that contains no judgement at all. You can automate everything you like there, and that is also where most of the manual work sits.
- Send an immediate confirmation that actually says something: who will look at it, when they will hear back, what the next step is
- Answer candidates' questions outside office hours, about the shift pattern, the travel distance, the contract or the training on offer
- Request missing documents and chase them politely: diplomas, background checks, references, availability
- Schedule interviews without six rounds of email, including a reminder the day before
- Send a status update the moment something changes, so nobody is left in the dark for three weeks
- Turn your interview notes into a written summary you can still find six months later
- Rewrite the job ad in plain language and publish it on the channels you already use
- Get everything ready for the first working day as soon as someone signs: clothing, tools, logins, an introduction schedule
What these tasks have in common: nobody is being judged by them and nobody is being rejected by them. The system moves information around and keeps the pace up. The decision stays with you. It is boring work, and it is by far the biggest part of the process.
Where the line is: automatic rejection is almost never allowed
In March 2026 the Dutch Data Protection Authority published rules for online and game-based assessments in recruitment and selection. Those go further than games: they cover any test instrument that automatically assigns a score to an applicant. According to the authority there is an automated decision when a score is produced entirely without human involvement and that score significantly affects the applicant. In hiring that is nearly always the case, because the score determines whether someone moves on.
Such decisions are only permitted on the basis of an exception in Article 22 of the GDPR. In practice only one comes into play: that the decision is necessary in order to enter into an employment contract. Asking the applicant for consent usually does not work, because in that relationship consent is rarely freely given. And if no exception applies, using such a test instrument is, in the regulator's words, simply prohibited.
A human signature under the outcome does not automatically save you. The authority requires human involvement to be meaningful: the assessor must be competent and must have the authority to deviate from the outcome. A team leader who takes over a tool's top five without ever seeing the other candidates does not meet that bar. On top of that you have to tell candidates in advance that an automated decision is being made, explain in plain terms how that score is produced, and give them the chance to object and ask for a fresh assessment.
We do not build systems that rank or reject candidates. Not because it is technically impossible, but because you are the one who has to explain it when someone asks why they were dropped. If the answer is 'the system did that', you have a problem that no amount of saved time is worth.
As a small company you are allowed less, and that surprises most people
Here is the point almost nobody mentions. Whether an automated decision is necessary depends, according to the authority, among other things on the number of applications you expect and receive and on the size of your organisation. And there is a second test: if a reasonable alternative exists that is less intrusive for the applicant and works just as well, the automated decision is probably not necessary and therefore not allowed.
Work that through for your own situation. A company of twenty people gets thirty-five replies to a vacancy. Reading thirty-five letters costs someone an afternoon. That is exactly such a reasonable, less intrusive alternative. Automated pre-selection then becomes hard to defend. For an organisation receiving four thousand applications a month, that balance looks completely different.
So the logic runs the opposite way to what most owners expect: the smaller you are, the less room you have to automate on the assessment side. On the communication and scheduling side nothing changes. There you may do everything, and that is where your gain is.
The AI Act has been postponed, except the part that affects you
Recruitment and selection is classified as high risk under the European AI Act. For a long time 2 August 2026 was the date on which heavy requirements would kick in: risk management, documentation, human oversight, demonstrably fair assessment. That has changed. On 24 July 2026 Regulation (EU) 2026/1744 appeared in the Official Journal, entered into force three days later, and pushes the obligations for this category to 2 December 2027.
new date for the high-risk requirements in recruitment (Regulation (EU) 2026/1744)
Be careful with what you read online: many articles on this subject were written before that regulation was published and still name 2 August 2026 as a hard deadline. Since the end of July that is no longer correct. Anyone telling you that you must comply with the high-risk requirements right now has not read the latest version.
One thing was not postponed, though. Since 2 August 2026, anyone communicating with an AI system must be able to know it. If you let a chat on your vacancy page or a phone assistant talk to candidates, that needs a clear notice. Not buried in the terms, but visible at the moment itself. And more importantly: the GDPR was never postponed. Article 22 has applied since 2018 and it is the stricter of the two here.
Almost everything in the AI Act has moved to December 2027. The one piece that touches your contact with candidates, telling them they are talking to AI, has not. That is good news: you have time to set this up properly rather than quickly.
Where you actually save time (and it is not in screening)
Do the sums on an ordinary vacancy. Thirty replies. Every reply gets a confirmation. Ten candidates ask a question by email or WhatsApp. You schedule interviews with eight of them, and that takes four to six messages back and forth each. Two cancel, one does not show up, and everyone who is dropped gets a proper answer. Add the reminders, the diary changes and the chasing of a background check. That is half to a full working week of admin per vacancy, spread across someone who is doing it on the side.
That is the work you can take away. And it pays back more than time alone, because in a tight market response speed is the real difference. Someone who applies at nine on a Tuesday evening and has a concrete appointment on their phone on Wednesday morning stops looking. Someone who has heard nothing after ten days is already working somewhere else. You do not win by screening better, you win by being the first to put an appointment on the table.
And consider what the other side costs: every week a technician, driver or service employee is not working, you pay in overtime, in temporary staff, in work you turn down and in colleagues absorbing it. You know that figure better than anyone. Put a number on it before you build anything, because that number decides whether this project is worth it.
When you are better off not starting
If you hire a couple of times a year, you do not need anything built. The job boards and staffing platforms you already use send confirmations themselves and have a scheduler built in. Switch those on, write a decent standard email and you are done. That costs you an hour and no money at all.
And if your problem is that nobody applies at all, automation will not help you either. In that case there is something wrong with your job ad, your salary or your visibility, and you do not fix that with a faster answer to zero applications. We would rather say that in the first conversation than afterwards.
The tipping point is not the number of employees, it is the rhythm. If you recruit all year round for the same roles, technicians, drivers, waiting staff, cleaners, production workers, you repeat the same steps dozens of times a year. That is where it pays. If you look for someone in a unique role twice a year, it does not.
How to approach it
- Write down exactly what happens now between 'application received' and 'interview scheduled', including who does it and where it stalls
- Measure two things for a month: how many hours pass between arrival and first response, and how many candidates drop out before the interview
- Automate only the communication and the scheduling. Leave the assessment alone, even if a supplier promises you it is allowed
- Put in your job ad and privacy statement what happens automatically with candidates' data. That is required anyway and it costs you half an hour
- Say so when a chat or phone assistant answers on your behalf. Since 2 August 2026 that is mandatory, and it is simply more decent
- If it measurably works better after one hiring round, extend it to the next role. If not, stop
It is not a spectacular project. No system is going to pick your candidates. What you get is something that answers every application within an hour, schedules interviews without fuss and forgets nobody. That sounds small, until you notice that the candidate you lost to silence last year simply turns up for an interview this year.
Ready to get started?
Request a free consultation. We look together at where you are losing time.
Schedule free call