Quick answer: A work sample test asks candidates to do a small, realistic piece of the job, then scores it against criteria you set in advance. It’s one of the more predictive hiring methods. A 2022 meta-analysis in the Journal of Applied Psychology put its validity at about .33. Keep it short (60 to 90 minutes for most roles), use it late in the process, and pay for anything longer than about two hours.
What is a work sample test?
It’s a piece of the actual job, done by the candidate, scored by you.
Examples by role:
| Role | Work sample |
|---|---|
| Customer support | Reply to three real (anonymized) customer tickets |
| Software developer | Fix a small bug or extend a function in a sample repo |
| Marketing | Draft a one-page plan for a product launch with a fixed budget |
| Sales | Run a 15 minute discovery call with an interviewer playing the customer |
| Designer | Critique an existing screen and sketch one improvement |
| Operations | Clean up a messy spreadsheet and summarize what’s wrong |
| Writer | Edit a 400 word draft |
The best ones look like a normal Tuesday in the role, not a puzzle.
Do work sample tests actually predict performance?
Yes, reasonably well.
The most current large review is a 2022 meta-analysis by Paul Sackett, Charlene Zhang, Christopher Berry and Filip Lievens in the Journal of Applied Psychology. It re-estimated the validity of common selection methods and found most were lower than older estimates suggested. Work sample tests came out at about .33. Structured interviews came out on top at about .42. Unstructured interviews were around .19.
Two takeaways for a small team:
- A work sample is a strong signal, but not a perfect one.
- It works best combined with a structured interview. See what a structured interview is.
When should you use a work sample test?
Late in the process, for a short list.
- Not at application. Asking 200 applicants for two hours of work wastes everyone’s time and drives away people with options.
- After the first interview. When you’re down to three to five candidates who could plausibly do the job.
- Before or as part of the final round. So the final conversation can discuss the work.
How do you design a good work sample?
Five rules.
- Test a must-have skill. One of the three to five skills from your role definition. Not something nice to have.
- Make it realistic. Use your real context, simplified. Strip out confidential data.
- Keep it short. 60 to 90 minutes for most roles. Tell candidates the expected time up front.
- Write the scoring criteria first. What does a 5 look like? A 3? Score every submission against the same anchors.
- Give everyone the same task and the same information.
A good scoring sheet has three to four criteria, each scored 1 to 5. For a support ticket task:
| Criterion | 5 looks like |
|---|---|
| Accuracy | Solves the actual problem, no wrong information |
| Tone | Clear, warm, matches the customer’s situation |
| Efficiency | Short enough to read on a phone, nothing missing |
| Judgment | Escalates or asks a question where needed |
Should you pay for work sample tests?
Pay for anything that takes more than about two hours, or anything you could use.
Long unpaid take-homes have three problems:
- They filter for free time, not skill. Candidates with caring duties or a demanding current job drop out.
- Strong candidates with several options often refuse.
- If the task looks like real work you’ll use, it feels exploitative, and candidates talk.
If paying isn’t possible, run the exercise live instead. A 45 minute session with an interviewer is fairer and gives you the chance to see how someone thinks.
Take-home or live?
| Take-home | Live | |
|---|---|---|
| Candidate stress | Lower | Higher |
| Shows thinking process | Less | More |
| Risk of outside help or AI use | Higher | Lower |
| Scheduling | Flexible | One fixed slot |
| Interviewer time | Review only | Full session |
The AI point matters more each year. If you use a take-home, follow it with a short live discussion: “Walk me through why you did it this way.” Someone who did the work can explain it. For more on this, see how to spot AI-assisted cheating in hiring.
What mistakes should you avoid?
- Testing the wrong thing. An algorithm puzzle for a role that builds forms.
- No time limit. Candidates who spend 10 hours will beat the ones who respected your stated 90 minutes.
- No feedback. Anyone who did your task deserves at least a short note on how it went.
- Using submissions in your product without paying. Just don’t.
- Adding it on top of five interview rounds. Replace a round with the work sample instead.
Where does an ATS fit?
The test itself usually happens outside the applicant tracking system (ATS), in a document, a repo or a live call. The ATS keeps the result with the candidate.
In Canvider you can attach the submission to the candidate’s application files, record scores and notes in internal comments, and @mention the other reviewers. For technical roles where your team lacks the expertise to judge the work, Canvider also offers vetted experts for technical screens.
When you’re comparing finalists’ work side by side, see compare candidates without a spreadsheet.
Try Canvider free. No credit card required.
Frequently asked questions
What is a work sample test?
A work sample test is a hiring assessment where the candidate does a small, realistic piece of the actual job, like writing a customer reply, fixing a bug, or drafting a campaign brief. It's scored against criteria set in advance.
Are work sample tests effective?
Yes. A 2022 meta-analysis by Sackett, Zhang, Berry and Lievens in the Journal of Applied Psychology estimated the validity of work sample tests at about .33, among the stronger selection methods it reviewed, behind structured interviews at about .42.
Should you pay candidates for a work sample test?
Pay for anything that takes more than about two hours, or anything you could use in your business. Unpaid long assignments filter for free time rather than skill, and candidates with caring duties or second jobs drop out.
How long should a take-home assignment be?
As short as possible while still showing the skill. For most roles 60 to 90 minutes is enough. Tell candidates the expected time up front and design the task so it can be done in that time.