AI in recruitment: how do you hire when everyone writes perfectly?

When AI tools started showing up in recruitment, I expected the change to be mostly about speed: faster screening, less admin, more time for conversations. Some of that has happened. But the most interesting changes I've seen have been somewhere else entirely. AI has made us question what our selection tools are actually telling us and where human time is most valuable in a hiring process.

I should say upfront that I don't have many answers here. A large share of the candidates we speak to are passive. They are people who are not actively looking and whom we approach directly. What follows is what we've observed in that context.

When everyone writes the same way

A big part of any application process relies on written content. CVs, cover letters, application forms, open questions and take-home assignments. Today, almost every candidate uses AI to help with some or all of it, and I don't see that as a problem in itself. Recruitment is time-consuming for candidates too, and if a tool helps them save time, that's understandable.

What I do notice is that most people use these tools quite superficially. The result is a striking homogenisation in how candidates express themselves in writing. Answers follow the same structure, use the same vocabulary and land on the same polished but empty conclusions. 

It has become very hard to get a sense of who the person actually is, and very hard for a candidate to stand out. More and more, I find myself preferring a short, slightly rough answer that clearly came from a person over three perfect paragraphs that could have come from anyone.

Written tests have lost their predictive value

My background is in work and organisational psychology, so I tend to look at this through the lens of predictive validity. Before AI, written material was one of the tools we used to make fast and informed decisions before speaking to anyone. How someone approached an assignment said something real about how they think.

That link has weakened. AI hasn't made candidates worse, it has made the instrument less reliable. We now see both false positives (candidates whose written work is excellent but who struggle to explain it) and false negatives (strong candidates whose words look less impressive next to AI-polished answers). At times, filtering on written material alone feels close to arbitrary, and I don't think that's a fair way to select anyone.

Bringing people in earlier

One of the concrete changes we've made at Alts is to increase the time we spend in live conversations with candidates and to do them much earlier in the process. Written stages still exist, but they no longer carry the weight they used to. Speaking with someone early has become the most reliable way to understand whether there's substance behind what we've read.

The end of the form-filling interview

For a long time, many interviews looked more like administrative forms than conversations: rate yourself from 1 to 5, do you have experience with this tool? Those interviews were never very good, and they are even less useful now, since the answers are easy to prepare and easy to fake. Telling a good candidate apart from a great one now requires the interviewer to listen properly and dig into what the person says: How did you actually do it? Why that approach? How would you evaluate the result?

That kind of questioning is much harder to outsource to a tool, because it depends on lived experience and personal reasoning. In a way, AI is pushing recruitment processes to become better, as it’s demanding more from interviewers than it used to.

Where AI does help us

None of this means we avoid AI. We use it to automate the parts of recruitment that sit around the human contact: organising sourcing and outreach, keeping our pipeline visible, transcribing and summarising interviews and pulling together reporting, etc. These uses give us back time we can then invest in conversations.

Where we draw the line is the contact itself. When most of the people we want to hire are passive candidates, automating touchpoints is almost the same as deciding not to have candidates at all. Very few people who aren't looking for a job will engage when their first interaction with a company is a bot, and I can't say I blame them.

Final thoughts

I'm still not sure where this is heading. AI hasn't simply made recruitment faster or slower; it has shifted where the value sits. Human interaction seems to carry far more weight in recruitment processes today than it did a few years ago, which inevitably raises questions about scalability.

It's also worth saying that every company's recruitment context is different. At Alts, much of our hiring relies on approaching passive candidates, so the problems we're trying to solve may not be the ones other companies are facing.