Recruiting in the AI era

We polled 1,428 US recruiters on what AI changed for them in the last 12 months. One answer came back louder than the rest. Finding people has got faster. Nearly three-quarters of recruiters, 72%, say AI increased the speed of finding candidates over the last 12 months.

But ask them whether they're any surer about those people they find, and it tells a different story. 59% say AI increased their confidence that they'd found the right person, but 18%, almost one in five, told us it’s actually made their confidence fall. Of the recruiters who say AI made finding candidates faster, around a third report no gain in certainty at all. Meanwhile, 35% say they have no clear picture of what “good” looks like in an AI role. 

So recruiters can now source faster. But are they any surer about who they find? 

In this report, we take a pulse check on recruiting in 2026. We will dig into this split, why it has occurred, and what can be done about it. 

When it comes to sourcing, nearly every tool competes on speed: more profiles, faster searches, automated outreach at scale. Many problems with sourcing that existed for years now seem largely solved.

However, the problem recruiters actually name as their hardest, certainty in the candidates they find, hasn't moved at all.

Let’s dig into why.

This stat, while eye-opening, isn’t entirely damning. A well-written resume has never really been proof of anything, and helping people present themselves clearly is a genuinely helpful thing for a tool to do. The problem is what happens to the claims inside.

Verifying that the skills on a resume are real was named by 57% as a top-three challenge, ahead of “a shortage of skilled candidates” and “culture alignment”. It ranked first when we asked in 2025 as well.

AI has only compounded the problem. 54% say that when a sourced candidate's profile claimed “AI skills”, those skills turned out weaker than the profile suggested, “often” or “almost always”. Nearly nine in ten say it happens at least “sometimes”. 

The claim doesn't fail at sourcing so much as survive it. We asked hiring leaders earlier this year on the state of hiring for AI fluency, and 59% said they'd made a bad AI hire in the previous 12 months: i.e., someone who sounded fluent in the interview and couldn't apply it on the job.

This is the pain point our own customer conversations surface most often, and the survey puts a hard statement on it: you can't trust what you see. 

The signal recruiters have always sourced on, the written profile, is now produced by the same category of technology that made searching those profiles fast. Recruiters got a better search engine and a worse index in the same 12 months. 

The effect concentrates exactly where hiring is hardest. It reaches 68% among recruiters whose work is mainly hard-to-fill specialist roles, and 63% among those sourcing AI and machine learning roles. The recruiters with the least room for error are the ones seeing claims fail most often. These are the searches where a wrong shortlist costs weeks, not days, and where there are fewest alternative candidates to fall back on.

The findings also reframe what "faster" delivered. If AI doubles the number of profiles you can assess in an hour, and a rising share of those profiles overstate what the person can do, the speed gain is mitigated by a lower-quality pool. 

Want to get more insight? We dived deeper into the trustworthiness of the AI-written resume

AI made searching faster for 72% of recruiters. Most candidate pools can be swam through in one afternoon. Back when a search took days, scarcity was hidden inside the work of looking. You didn't know whether the right people weren't there or whether you simply hadn't gotten to them yet. Now that the search can be done rapidly, that deep pool is suddenly starting to look pretty shallow. 

Nothing has necessarily changed about how many qualified people exist. What changed is that recruiters can now see how few there are.

Pool size and specialist role scarcity aren't independent problems. Faster search converts an abstract shortage into a concrete one, and it does it without adding a single qualified candidate to the market.

So in short, we have a paradox in that the recruiters who report being most confident in their candidate data are often the same ones who say that data let them down.

Two explanations for why this paradox is occurring:

Both explanations end up in the same place. Trust in tool-supplied candidate data isn't being calibrated by evidence. Nothing in a typical sourcing workflow tells a recruiter which claims were checked, how, or by whom. 

Can't trust what you see turns out to have a second layer: most recruiters can't see what they can and can't trust.

The outreach doesn't reliably land either. A third of recruiters, 33%, hear back from fewer than a quarter of the candidates they contact directly. Work through 26 profiles, contact the best of them, and most of that effort goes unanswered.

From this data, we see that that resistance to proving a candidate's skill level isn't the obstacle. If only 4% of recruiters would skip a check given a free one, the reason proof isn't being sought during sourcing is not that recruiters don't want it.

Identity, contact details, and employment history, for the most part. Sourcing platforms have invested heavily in confirming that a person is real, reachable, and worked where they say they worked. That work is useful, and it solved a real problem. But very few platforms can actually confirm that a person can do the thing their profile claims they can do, and fewer still will explain how their ranking arrived at the names it put in front of you. 

So the category verifies the facts easiest to check against a database, and leaves the one that requires the candidate to demonstrate something. Meanwhile, the profile itself got less reliable, with over half of recruiters now believing most are written by AI. The one problem the category didn't take on is the one that continues to get harder.

The two problems that more reach could solve were solved. But the two that require evidence rather than reach were not. Speed, scale, and coverage were the right answers to the questions the industry chose to ask, and none of them answer the question a recruiter is actually stuck on: can this person do the job?

Proof-first sourcing is the practice of treating evidence, rather than reach, as the measurement to limit a shortlist. It means criteria are agreed before candidates are scored, every name arrives with the reasoning attached, and a claim on a profile counts for less than a demonstration of the skill itself.

Recruiters got faster at finding people because the industry built for speed. They didn't get surer, because certainty was never what the industry was building. The signal sourcing runs on is now written by the same technology that made searching it quick; the pool is visibly thinner than it used to look, and the one check that would settle the question is still being done by hand, an hour at a time, by the recruiter.

Reach has stopped being a competitive argument, because reach is close to free. 900 million profiles is not a meaningfully harder thing to offer than 100 million. The argument worth having now is whether you can defend the names on your shortlist: why these people, on what evidence, against which criteria, and what happens when a hiring manager asks.

That's a different way to build a sourcing tool, and according to our survey it's the one recruiters are asking for.

On proof, you set the search criteria, AI improves the search, and every name arrives with a written reason for its ranking. Then it's over to Gorilla Assess to test the shortlist. Nobody is assessed automatically, and no candidate sits an assessment before a recruiter chooses them, giving you total control over your shortlist.