Why talent sourcing is broken

Why is talent sourcing so hard now?

ManpowerGroup's 2026 Talent Shortage Survey asked 39,063 employers across 41 countries how much difficulty they have filling roles. Seventy-two percent said they struggle, down modestly from 74% the year before. In the US, 69% report the same problem.

For the first time, AI skills lead the global list of hardest capabilities to find. The report mentions that AI model and application development requirements are both ahead of engineering, while traditional IT and data skills have fallen to seventh. TestGorilla’s  State of Hiring for AI Fluency Report, found that managers would rather have an employee with AI skills than deep domain expertise. 

Simply put, the skill employers now find hardest to source is the one almost every profile claims to have.

Is this just LinkedIn's fault?

Partly, and blaming it is comfortable, because a tool problem implies a tool fix. Sarah Doughty, a technical recruitment expert, told us: 

"[LinkedIn has] slowly changed the search functionality to a point where really even the basic Boolean searches are not functioning the way they should. It's very difficult these days to run a search on LinkedIn and get less than a thousand results."

So yes, LinkedIn seems to have changed. But swap it for any other sourcing tool, and you'll notice something. They all read the same field. Every one of them ranks self-described profiles. Change the vendor, and the candidate input will still be the thing it’s operating on.

Why self-reported profiles are now a broken signal

The pool of self-reported profile data that every sourcing tool reads and ranks has stopped separating candidates when writing it got automated on one side and reading it got automated on the other.

While profile data was always self-reported. What has changed now is what it costs to produce that profile. A well-optimized profile on LinkedIn, a thoughtful application - used to take real effort, and effort worked as a rough proxy for interest and competence. But with AI, the cost is close to zero, and the same models draft the profile and score it.

The result is a field where everything looks like the job description, because much of it was written from the job description.

None of that makes candidates the problem. If a filter rewards the right words, using the right words isn't cheating. It's literacy. People are responding rationally to a system that told them exactly what it measures. The failure sits with the measurement.

Why do sourced candidates miss the mark?

When a search returns a thousand near-identical profiles, the rational move is to try to work more of them. So teams do, and the hit rate drops. The problem, George Fironov, Co-Founder and CEO of Talmatic, says is that “a high volume of applications tends to mean poor targeting rather than recruiting success.”

Here's what a profile can and can't carry:

Every column on the left is a proxy. Sourcing on proxies works while the proxies stay expensive to fake. 

Two things usually follow this method. One, sourcers burn out working lists that don't convert, which is a workload problem created upstream of them. Two, the sheer amount of candidates means good people fall out of the pool for the wrong reasons. 

Chris Edelen, Client Development Director at Culver Careers, names the mechanism: rigid job descriptions with overly specific credentials like an exact title match or niche software experience can filter out adaptable, high-potential talent. 

Career changers, people returning to work, and candidates with adjacent skills get hit hardest, because their value doesn't show up as the right keyword. 

That’s why we recommend a skills-first sourcing approach. 

What is skills-first sourcing?

Skills-first sourcing means you decide who to approach based on evidence that they can do the work, collected before the first message, rather than on how they describe themselves.

It isn't a faster way to read profiles. It's an entirely different thing to read. Instead of asking which profiles contain the words you searched for, you ask which people have already demonstrated the capability, when they did it, and against what standard through skills-based assessments. 

That single change moves the decision off self-report and onto something you can check.

Is this just faster keyword matching?

Both a keyword tool and a human Boolean search read self-reported claims and rank them. Same field, same reliability, so speed is genuinely the only variable. Faster access to an unreliable field isn't progress. It's the same wrong shortlist. 

Verified-signal sourcing changes the field, not the pace. The question stops being "does this profile say Python" and becomes "did this person complete a Python task, what did they score, when, and measured against what benchmark?" That data doesn't exist on the profile. No search string reaches it, however well written, because it was never written down by the candidate at all.

The difference shows up in what comes back. Point both approaches at the same pool. The keyword approach returns everyone who used the word. The evidence-based approach returns a shorter list you can defend to a hiring manager without adding an interview round to check.

One honest limit. This only works where the evidence exists. It won't retrofit onto a profile that has none, and it doesn't tell you whether someone wants the job or would thrive on the team. Verified skills narrow the field to people who can do the work. The conversation still decides the rest.

How do you source on tested skills?

1. Brief the role. Don't build the filters.

Write down what the person has to be able to do in the first ninety days. Not the title you'd like them to have held, not the years you'd like them to have served. Capabilities are testable. Credentials are proxies, and proxies are what broke. You brief it. You don't build it.

2. Decide what counts as proof before you search.

Pick the two or three capabilities that actually predict performance in the role, and agree with the hiring manager what proof looks like. Sort them as you go. Must-haves decide who makes the list. Nice-to-haves decide the order. Do this first, and you stop negotiating the shortlist after the fact

3. Work a pool where the evidence already exists.

Checking candidates yourself, one at a time, doesn't scale past a handful of roles. Sourcing from people who have already sat an assessment does. This is the part that changes the economics, and the market is already moving in the wrong half of it. Half of sourcing teams say they struggle to find qualified candidates for their open roles. Most teams are widening the pool. Fewer are asking whether the wider pool carries better evidence than the old one.

4. Start where the score already exists.

Most teams run one order. Source on keywords, message at volume, assess whoever replies. That order is right when you're reaching into the open market, and confirming interest before you ask someone to sit a test is fair to the candidate. What changes the math is starting with the people who already carry a score. For them, the evidence arrives before your first message, so you write to someone you already have a reason to want.

5. Put the evidence in the message.

If the proof is already on the profile, you can say precisely why you're reaching out, criterion by criterion. A work history suggests. Only an assessment shows. That beats a personalization treadmill, and it's the only outreach improvement here that doesn't cost you more time.

Do I still need LinkedIn Recruiter?

Most likely, yes. Reach and evidence are different axes, and you need both. LinkedIn is where a very large share of the working world is discoverable, and nothing in this argument suggests leaving it. The teams handling this well aren't cancelling their licenses. They're right-sizing seat counts and adding a second avenue that reads a different field, so the reach tool stops carrying a job it was never built for.

If you want the detail on where each tool actually helps, we've compared the sourcing tools recruiters use and what each one verifies.

Where to go next

Sourcing didn't break because recruiters got worse or tools got slower. It broke because the layer everyone sources on stopped carrying signal, and no amount of faster searching fixes a field that no longer discriminates.

If you're rebuilding your approach, start with the full picture of how candidate sourcing works end to end, or look at how to reach people who aren't actively job hunting.

Talent sourcing FAQs

Why is talent sourcing so hard right now?

Because the data you source on stopped separating candidates. ManpowerGroup found 72% of employers struggle to fill roles in 2026, barely changed after a decade of better search tools. Profiles are now cheap to optimize and increasingly written with AI, so keyword-based sourcing returns large, undifferentiated lists. More search effort produces the same hit rate.

Why do all candidate resumes look the same now?

Because most of them are written from the same job description, often with the same tools. Keyword-based filters told candidates exactly what they measure, and candidates responded sensibly. The result is a field of documents optimized for retrieval rather than description. That isn't dishonesty. It's a predictable response to a system that rewards phrasing over proof.

What is skills-first sourcing?

Skills-first sourcing means selecting who to approach based on evidence they can do the work, gathered before the first message, rather than on self-description. In practice you define the role by capability, agree what counts as proof, and source from a pool where that proof already exists. The input changes, not just the speed.

Is AI sourcing just faster keyword search?

Often, yes, and that criticism is fair. If a tool ranks self-reported profile text, it's doing what a Boolean search does, only quicker, and speed on unreliable data buys you nothing. The distinction worth paying for is what the tool reads. Scored evidence from a completed skills test is a different field, not a faster query.

How do you verify a candidate's skills before you reach out?

Two routes. Run a short skills assessment as part of your process, which works well for a handful of roles but doesn't scale. Or source from a pool of candidates who have already been tested, where the evidence exists before you arrive. The second changes the economics, because verification stops being work you add per candidate.

What is the difference between sourcing and screening?

Sourcing decides who to approach. Screening decides who to advance. They fail differently: sourcing fails when you reach the wrong people, screening fails when you can't tell good applicants apart. This article is about the first. If your problem is the applicant pile rather than the pipeline, that's a screening question.