AI in recruitment

What does AI do well in recruitment?

AI performs where the work is repetitive, high in volume and describable in criteria. Four uses hold up today:

• Sorting and ranking applications against criteria you have set, when more arrive than anyone can read one by one on the same day.
• Producing a first draft: a job ad, an outreach message, a standard reply. A starting point to rewrite, never a text to publish as it stands.
• Finding profiles on professional networks, by widening a search to adjacent job titles nobody had thought of.
• Reading campaign data: which channel brings usable applicants, where people abandon the form, which message lands with which audience.

That last one is the most underrated. In recruitment marketing the gain is not that a decision gets taken for you, but that it can be prepared faster: spotting a signal in the numbers, testing message variants, stopping what produces nothing. Most employers already use AI without calling it that, in the search function of LinkedIn Recruiter, in the ranking their ATS applies, or in a text generator such as ChatGPT for a first draft of a vacancy.

What can AI not do?

A model recognises patterns in existing data; it does not understand a person. It cannot tell you whether someone will get on with the manager already in place, whether a career switch is solid, or why there are eight months between two jobs. Those judgements come out of a conversation, not out of a CV.

Two concrete risks follow from that.

• Bias. A model trained on who you hired in the past repeats the preferences behind those choices, including the ones you were trying to unlearn.
• The false rejection. A good profile drops out because it uses different words than the filter expects, and nobody ever finds out.

The countermeasure is simple and rarely applied: take a sample of what was rejected, put it back on the table and read it by hand. Do that a few times and you see straight away whether the filter is doing its job. Usually the criteria need changing rather than the tool. The terms you will run into are explained in our glossary.

Does an AI recruiter replace a real recruiter?

No. What gets sold as an AI recruiter is a software category: tools that find candidates, send messages and book interviews. They take over tasks, not a role.

The difference is persuasion. Finding someone is not the same as convincing them. The people you need usually have a job already and are not looking. They do not move because of an automated message that also reads like one. They respond to a concrete story from someone who knows their trade and can answer the question that comes next: how is that team put together, why is this seat open, what happens when things go wrong.

That is why active sourcing stays human work, and why employer branding decides the outcome before any tool gets involved. Software can make both faster. It cannot stand in for them.

What do the GDPR and the EU AI Act require?

Using AI in recruitment is allowed, but not without conditions. Two sets of rules meet here and it helps to keep them apart.

The GDPR covers your candidates’ data: what you collect, why, and how long you keep it. It also carries a principle that bites here. A person should not be subject to a decision with significant effects based solely on automated processing, apart from a few defined exceptions.

The EU AI Act covers the systems themselves. Systems used to recruit or select people, meaning targeted job advertising, filtering applications and evaluating candidates, fall into the high risk category. That does not ban them; it attaches obligations: risk management, quality of training data, documentation, transparency towards the people involved, and human oversight that actually means something. In short, a model may prepare a decision, but a person has to be able to understand it, set it aside and answer for it. You also need to be able to tell candidates that such a system is in use.

Application dates and the exact duties are still moving, so check them in the regulation itself or with your data protection authority before you deploy a selection tool. One question stays useful in the meantime, including towards any recruitment agency that hands you a shortlist: what filtered this list of candidates, and who read it back?

How does SocialFind use AI?

We use AI as an aid in preparation: setting up a search, producing text variants, reading campaign figures. No decision about a candidate is made by a tool. A recruiter reads, calls, listens and decides, and that person answers for it, to you and to the candidate.

So here is what we do not do. We do not sell matching technology of our own, we do not let a model score or reject candidates, and we promise no hours or percentages that AI would save you. Our work is convincing people who are not looking, under your name and with your arguments. That is contact work, and automating it removes the very thing that makes it work. AI buys us time on the preparation, not on the conversation.

If you want one party to take on recruitment as a whole, that is RPO: a team that recruits under your name for a fixed monthly fee rather than per placement. If you do not yet know where things stall, start with the recruitment diagnosis, which looks at your open roles and your market rather than at your software. And if your problem is the number of applicants rather than the selection, no tool solves that: reach has to come first.