AI in recruitment - What candidates see through, and what works in trades, construction and logistics

AI in recruitment went from experiment to standard in about two years. Job ads are generated, candidates are matched, screening questions are handled by a model. And at the same time something is happening that the product brochures leave out: candidates trust the hiring process less than they did a year ago. In the AI in Hiring study by Greenhouse, covering 4,136 job seekers, recruiters and hiring managers in the United States, the United Kingdom, Ireland and Germany, 70 percent of hiring managers said they trust AI to make faster and better decisions. Among job seekers, 8 percent thought AI makes hiring fairer.

That gap from 70 to 8 is the real story. Not because AI is bad, because AI is excellent at speed, variants and scale. But because the technology tends to get deployed at exactly the point where recruitment depends on trust. And in blue collar sectors, where scarcity bites hardest, trust is not a soft factor but the reason someone does or does not leave their details.

Where does AI in recruitment genuinely help?

Start with what is beyond argument. AI is strong at volume and variation. Fifteen ad variants per audience instead of three. Twenty headlines tested before you buy media. Translations for a Polish or Romanian audience that used to cost two weeks and an outside agency.

Behind the scenes the gain is just as real. Candidates hearing back within an hour instead of after nine days. The same Greenhouse study found that 34 percent of recruiters spend up to half their week filtering spam and unusable applications. Let AI loose on that.

Why can a candidate spot an AI written job ad?

Because a model has never stood on a shop floor. It writes the statistically most likely answer to the instruction to produce a job ad for a service engineer. And that most likely answer is the average of everything already out there. Dynamic team, challenging role, plenty of room for initiative, competitive salary.

An engineer with ten years of experience reads that and knows exactly as much as before. Nothing about the shifts, nothing about the van, nothing about who the supervisor is. Strictly speaking that is not an AI problem but an input problem: you gave the model nothing to work with. The effect is still that your text does not stand out, and in a market where everyone is fighting over the same fifty engineers that comes close to not existing at all.

Read our piece on AI in employer branding alongside this one, because that is exactly about the line between scale and substance.

What happens when candidates start using AI too?

You get an arms race nobody wins. The same Greenhouse study showed that 41 percent of American job seekers admitted using prompt injections to get past automated screening, and that 69 percent had come across fake job postings. In the United Kingdom, 45 percent of job seekers trust the hiring process less than a year earlier, and 40 percent of them point straight at AI.

On your side of the table the picture is no better. Only 21 percent of the recruiters surveyed were genuinely confident that their system is not rejecting qualified candidates. Twenty one percent. In a market short of applicants, that is an expensive risk.

Candidates also use AI to research you. Research by iHire among more than 1,600 American workers found in June 2025 that 11.6 percent had used AI tools to look into employers. Still a minority, but it is the direction of travel. Which means a model has to be able to find things about you that are actually true. If your recruitment website is mostly atmosphere, that is precisely what comes back out.

Does AI in recruitment work differently in trades, construction and logistics?

Yes, in two ways. The first is scarcity itself. The employer survey by UWV, the Dutch public employment agency, covering 3,551 employers in the autumn of 2025, found that 45 percent of vacancies were hard to fill. In construction that was 71 percent, in manufacturing 53 percent, in transport and storage 47 percent. Of employers with a vacancy, 87 percent named too few applicants as the cause. At those numbers, every candidate who drops out needlessly is a problem rather than statistical noise.

The second is search behaviour. A process operator or a site manager is not on LinkedIn all day. They look on a phone, during a break, through an ad or through a colleague who forwards something. AI helps enormously there with reach and with variants per region and per shift. It does not help with the question of why someone would leave their current boss. That answer has to come from people.

For Hanab, part of VolkerWessels, we therefore built an employer brand sourced on site rather than in the boardroom. For Picnic much of it turns on what people actually experience in that first year, from onboarding to moving up through the Picnic Academy. AI is the engine behind the rollout there, not the source of the substance.

What can a model not fake?

A real face with a real story. This is what we believe most firmly: AI for scale, people for substance. Why would you synthesise an employee you could simply film? An engineer explaining in thirty seconds what he repaired today and why he is still there after eight years beats any generated version of that. Not because it looks better, but because it is true and the viewer can feel it.

It sits in details a model cannot know. The nickname for the night shift. Why the schedule always runs over on Friday. Why the same people are still there after three years. Those are the things that make someone think: this is a real place.

How do you keep AI both a magnet and a filter?

By not aiming it only at attraction. An optimised funnel that doubles your applications is not a win if half of those people leave within three months. You have simply moved the problem to your supervisors and your onboarding.

An honest picture works as a filter. Say that the work is outdoors and that January is cold. The candidates who drop out because of that were not the ones you wanted. The ones who respond because of it tend to stay. AI helps you test that honest message in twenty variants, so you know which wording attracts the right people instead of the most people.

Is your problem really an AI problem?

Often not. Not every problem is a campaign, and certainly not every problem is a tool.

If your application process has seven steps and sits still for four days between step two and step three, no model repairs that. If your careers page takes eight seconds to load on a phone, that is your bottleneck. If first year turnover is high, the problem is not your intake but what happens after it. The UWV survey found that 52 percent of employers now more often hire people who still need training, because of the shortage. That is a sensible move, but it calls for guidance rather than a better ad.

So always start with the problem and agree in advance which number has to move. Applicants per vacancy, time to hire, first year turnover. Without a baseline you cannot tell afterwards what AI delivered, and then you are left with a feeling instead of a result. Our piece on recruiting technicians shows how that plays out in practice.

How does MonFri approach AI in recruitment?

We start with your problem. Are the vacancies not filling, or are they filling with the wrong people? Is it taking too long, or are people leaving again too quickly? That decides what we do, and therefore where AI does and does not belong.

Then we build an employee value proposition that holds up inside and out: what your own people say, what the outside world thinks of you and what the market around you offers. The substance comes from employees, with a camera on the shop floor. The scale comes from AI: variants, tests, languages, regions and shifts, speed. And we measure whether the number we agreed on actually moves.

More about how we work is on our services page. Rather spar straight away about where AI will and will not help you? Call 020 23 38 587 or email hi@mon-fri.com.