Project-Based Hiring in 2026: Do Work Samples Beat Interviews?

What is project-based hiring?

Project-based hiring means you judge a candidate on a piece of real work from the job, done before or alongside the interview, instead of judging them mostly on a CV and a conversation. Industrial psychologists call that piece of work a work sample test. A copywriter rewrites a landing page. A junior analyst cleans a messy export and writes up what they found. A support hire answers five real, anonymised tickets.

The idea is old. What's new is how many employers say they want it. NACE's Job Outlook 2026 survey found 70 percent of employers using skills-based hiring for entry-level hires, up from 65 percent a year earlier, while the share screening candidates by GPA fell from 73 percent in 2019 to 42 percent. So employers want proof of skill. Most of them still try to collect that proof in a conversation, which is the weakest place to find it.

Related reading: The Hidden Workers ATS Problem in 2026: Qualified Talent Your Filters Reject · AI Role-Play Pre-Onboarding in 2026: How to Prepare New Hires Before Day One · CSR Employability Programs in 2026: How Employers Fund Programs That Work

Work samples vs unstructured interviews: the research before and after 2022

Work samples predict job performance clearly better than unstructured interviews, though both predict less than HR teams were taught for a quarter of a century. The old benchmark is Frank Schmidt and John Hunter's 1998 meta-analysis in Psychological Bulletin, which pulled together 85 years of studies. It gave work sample tests an operational validity of .54, the highest of any single method, against .51 for structured interviews and .38 for unstructured ones. By 2023 that paper had been cited more than 6,500 times, according to Sackett and colleagues, so you've almost certainly seen its numbers on a training slide.

In 2022 Sackett, Zhang, Berry and Lievens published a revision in the Journal of Applied Psychology. The argument is simple once you see it. Most validity studies are concurrent, meaning they test people already doing the job, and those groups usually aren't much narrower than the applicant pool. Earlier meta-analyses applied large statistical corrections for "range restriction" anyway, borrowed from predictive studies where the restriction really is big. That pushed many estimates up.

After the fix, unstructured interviews dropped to .19, half their old value, and structured interviews took the top spot at .42. Work samples fell from .54 to .33 for a different reason. Schmidt and Hunter's figure rested on a 1974 narrative review, so Sackett's team used Roth, Bobko and McFarland's 2005 meta-analysis in Personnel Psychology instead, which pooled 54 studies and landed on .33 with no range restriction correction at all.

Two details in Sackett's 2022 estimates get less attention than they deserve. Work samples varied less from study to study than structured interviews did (a standard deviation of .09 against .19), so you're less likely to end up with a dud version. And years of job experience, the thing CV screens lean on hardest, fell from .18 to .07. If you're filtering on "3+ years" before anyone has seen a line of work, you're filtering on very little.

For the record, a validity of .33 is a corrected correlation between scores on the method and job performance. It won't pick your best hire for you. It shifts the odds.

Assessment methods compared

Structured interviews, job knowledge tests and work samples sit near the top for validity, while unstructured interviews and years of experience sit near the bottom. The table below uses the side-by-side comparison Sackett and colleagues published in Industrial and Organizational Psychology in 2023, which puts their 2022 estimates next to Schmidt and Hunter's. The fourth column is the standardised Black-White score gap (Cohen's d) reported for each method. Bigger numbers mean more risk of adverse impact.

Source: Sackett et al. (2023), Table 1, Industrial and Organizational Psychology. The same table lists two later updates: assessment centres rise to .33, tying work samples, and cognitive ability falls to .23 using 21st-century data.

Look hard at the work sample row. The validity is solid, but the subgroup gap of .67 is nearly three times the .23 for structured interviews. Work samples look job-related, so it's tempting to call them the fair option. This data doesn't support that by default. Fairness depends on what the task demands and on who has had the chance to practise it.

Work samples have limits, and the research says where

Work samples work best when candidates already have the skill being sampled, which is why they suit experienced hires better than people looking for a first job. Sackett and colleagues say this plainly in their 2023 follow-up: work samples and job knowledge tests generally don't fit entry-level hiring where people pick up the skills after they're hired, through training or on the job.

That's a real problem for graduate programmes and university placement teams. Ask a final-year economics graduate to build a financial model in your house template and you're mostly testing whether they did an internship at a bank. You aren't testing whether they'd be good at the job after six weeks of training. Run it for a few cycles and you'll hire from the same handful of campuses.

The same paper points to the way out. It describes work samples as more open to improvement through study and practice than ability or personality measures, which means you can tell people what to prepare and expect them to get better. For entry-level roles that suggests a few changes:

• sample the trainable basics of the job, such as a clear reply to an annoyed customer or finding the error in a small dataset, rather than firm-specific craft
• publish the brief, the scoring criteria and a worked example in advance, so practice counts for more than insider knowledge
• pair the project with a structured interview scored on the same competencies, since the interview has the highest average validity and the smallest subgroup gap in the table

Our view: the strongest project-based hiring setup in 2026 is a short work sample plus a structured interview, both scored on one rubric. The unstructured "tell me about yourself" chat belongs in the part of the process where you sell the job.

How do you design a fair paid or short project?

A fair project is short, close to the real job, scored against a rubric written before anyone submits, and paid once it starts to look like real work. Start from a job analysis, even an informal one: list the tasks that fill a new hire's first 90 days and pick one a capable newcomer could do with only the information you hand them.

On pay, two well-known employers show what good looks like. PostHog's public hiring handbook (as of 2026) says it pays every candidate a flat US$1,000 for its SuperDay, a full day spent on a task like the real job, and donates the money to the Django Girls Foundation if a candidate can't accept payment. The task is designed to be too much for one person in a day, so the team can see how people prioritise. Automattic, the company behind WordPress.com, said in its 2024 design hiring guide that all trial projects are paid at a standard US$25 an hour.

The law matters too, and it varies by country. In Australia, the Fair Work Ombudsman's guidance says an unpaid trial is only lawful if it's needed to assess the person's skills for the vacant job, lasts only as long as it takes to show those skills, and happens under direct supervision. Anything beyond that must be paid at the minimum rate. It's a sensible default even where the law is vaguer.

The practical checklist we'd use:

• Time-box it. State the expected time and mean it. If your own team can't finish the brief in that time, cut it down.
• Write the rubric first, with three to five criteria and a short description of a weak, good and strong answer for each.
• Have two reviewers score each submission independently, with names and universities hidden, then compare notes.
• Never use the output. If it's real client work you plan to ship, it's a contract and should be priced like one.
• Offer an alternative, like a shorter task or a live version, for people with caring duties or a day job.
• Send feedback to everyone who submits, even if it's two lines.

Our rule of thumb is blunt. If the brief takes more than about two hours, pay for it.

Candidate experience is where most take-homes go wrong

Candidates like work samples as a method; what they resent is unpaid time that vanishes into silence. A 2025 study in the International Journal of Selection and Assessment by Zibarras, Castano and Cuppello asked 281 working adults to rate 17 selection methods, and work sample tests, knowledge tests and in-person interviews came out most positive. Asynchronous video interviews landed near the bottom. That fits an earlier 2010 meta-analysis by Anderson, Salgado and Hülsheger, which found that applicants in different countries react to selection methods in much the same way.

The anger shows up in a different place. Resume Genius surveyed 1,000 active US job seekers in March 2026, and 25 percent listed being asked to do unpaid assignments or tests among their hiring frustrations. Bigger complaints were hearing nothing after applying (55 percent) and getting no reply after interviews (44 percent). Greenhouse's 2024 State of Job Hunting report found 61 percent of job seekers had been ghosted after an interview.

Picture a graduate who spends a Saturday on your brief, submits on Sunday night and hears nothing for three weeks. They won't remember your well-designed task. They'll remember the silence, and so will their course group chat. A two-line note with one specific strength and one thing to work on takes a reviewer a few minutes, and it might be the only useful feedback that person gets all month.

Where project-based hiring fits for universities, CSR and L&D teams

Project-based hiring is most useful when you need evidence of skill from people without a long CV: graduates, career changers and talents from under-represented backgrounds. That's also where talk runs furthest ahead of practice. A February 2024 report from the Burning Glass Institute and Harvard Business School found that dropping degree requirements changed fewer than 1 in 700 hires in 2023, and about 45 percent of firms that announced the change showed no real difference in who they hired.

Vendor surveys are rosier. TestGorilla's State of Skills-Based Hiring 2024, which polled 1,019 employers in March 2024, found 81 percent using skills-based hiring and 90 percent saying it cut mis-hires. Those are self-reports in a test vendor's own survey, so read them as sentiment rather than outcome data. Employers believe in the idea. Few have built the plumbing to act on it.

For university placement teams, the practical move is to run employer-set projects during term, with the brief and rubric agreed in advance, so talents leave with scored work and not just a list of modules. For CSR and impact leads, funding paid projects is a direct way to make an employability programme count, because a paid task removes the cost barrier for people who can't afford to give away a weekend. And for L&D, the rubric you hire against can become the first week of onboarding, which is where pre-onboarding role-play picks up.

How Talenlio lets talents show the work first

Talenlio's HireOS lets an employer post a project with a real task instead of, or alongside, a job ad, and talents complete it so you see how they work before you interview. Talents can then step into AI role-play interviews and realistic scenarios, so hiring teams see judgement as well as output before an offer. Unified job posting, the verified talent pool and shortlisting are covered on the Talenlio for Employers page.

On the talent side, the Challenges agent sets daily and weekly skill challenges. A talent who has done a dozen short, scored challenges arrives at your brief used to the format, so you get a cleaner read on what they can do. Practice is the whole idea here. A work sample copies the job, so getting better at the sample usually means getting better at the work.

If you change one thing in your next hiring round, make it this. Pick a single role you fill every year, write a two-hour brief and a five-line rubric this week, and pay for anything longer. Then put the shortlist it produces next to the one your CV screen would have produced. We'd bet the two lists overlap less than you expect.