Fair and fast assessment for creative roles in games: art tests, design tasks and structured interviews

Summary

Our main points:

1. A well-run structured interview predicts job performance better than any other single method. Work samples help, though less than older figures suggest.

2. Short, scored, time-capped tests are fair. Long unpaid tests shut out good people.

3. Slow scoring and unclear briefs cost more time than the test itself.

1. What the research says

Sackett, Zhang, Berry and Lievens (Journal of Applied Psychology, 2022) rechecked decades of studies and corrected earlier over-estimates:

• Structured interview: Validity: about 0.42

• Job knowledge test: Validity: about 0.40

• Work sample: Validity: about 0.33

• Cognitive ability test: Validity: about 0.31

• Unstructured interview: Validity: about 0.19

The older figure for work samples, 0.54, is still widely quoted. Work samples remain useful, and they should sit next to a structured interview and not replace it.

Greenhouse, the hiring software company, looked at anonymous grading of take-home tests. Its own analysis found pass-through rates rose by 6.5 to 10% when graders could not see who did the work.

On length, our advice is two to three hours as the upper limit for an unpaid task, with a short live walkthrough afterwards. Longer tasks are hard for people with full time jobs or caregiving duties, and that includes most senior candidates.

2. What we see in games searches

From our 2026 searches and client feedback, anonymised:

• Tests set the bar. Clients held senior 3D art candidates to the level expected of a senior role, and were willing to wait for a second candidate's test before deciding.

• Senior roles need more than a test. Clients judged confidence, communication, leadership and attitude in the first call.

• Deadlines vary. Art tests with a one-week deadline were common, and extensions of a couple of days were granted when candidates asked early.

• A good timed test exists. One studio used a 30-minute timed motion design exercise. It was rigorous and quick for the candidate, and it needed someone to brief each person, hold the time window and chase submissions. We ran it for them across two months (see our case study "We Ran Their Assessment").

• NDA work gets in the way. Candidates on live projects often could not show recent work.

• Portfolio focus matters. Strong portfolios were turned down for focus, for example mostly characters for an environment role.

• AI questions are new. Candidates ask whether AI tools are allowed in tests, and some platforms do not allow AI in final assets.

3. A fair assessment kit for creative roles

Step 1. Decide what you are testing. Write three to five criteria from the job, such as style match, technical quality, speed, iteration on feedback and communication. Score each 1 to 5.

Step 2. Review portfolios against the same criteria. Two reviewers score separately before they talk.

Step 3. Run a structured interview of 45 minutes, with the same questions for all candidates at the same level. For example:

• Walk us through one portfolio piece from brief to final. What would you change now?

• Tell us about feedback you disagreed with. What happened?

• How do you work to a style that is not your own?

Step 4. Set a short test, only if the portfolio leaves a real question.

• 2D or concept artist: Test idea: Paint-over or thumbnail set from a short brief in the studio's style; Time cap: 2 to 3 hours

• 3D or environment artist: Test idea: Block-out or texture pass on a provided asset, with a poly or texture budget; Time cap: 3 hours, or paid if longer

• Animator: Test idea: Short cycle or action on a provided rig; Time cap: 2 to 3 hours

• Motion designer: Test idea: Timed edit from provided footage; Time cap: 30 to 60 minutes

• Game designer: Test idea: Critique and improve one system in a released game, on one page; Time cap: 2 hours

• UI or UX designer: Test idea: Redesign one screen flow with a short write up; Time cap: 2 to 3 hours

Rules for every test:

• Use the same brief for every candidate at that level.

• Never use live production work. Use a provided asset or an old, shipped one.

• Say in writing that the work is for assessment only and will not be used.

• Say whether AI tools are allowed, and how to label any AI use.

• Pay for anything over three hours.

Step 5. Hold a live walkthrough of 20 minutes, where the candidate explains their choices. This shows their thinking and settles any doubt about who did the work.

Step 6. Score blind if you can, by removing names from test files before review.

Step 7. Decide within five working days of the last submission, and give every candidate one or two lines of feedback.

4. NDA and AI lines to include in every brief

NDA line: "If you cannot share recent work because of an NDA, tell us. You can describe your role in writing without images, or take the short test instead."

AI line: "You may use AI tools for reference and ideas. Final pixels, models and animation must be your own. Label any AI-assisted step." Adjust this to your studio's and publisher's rules.

5. A timeline that keeps good candidates

• Portfolio decision: Target: 3 working days

• Interview to test: Target: Test sent the same day

• Test window: Target: 3 to 7 days, extension on request

• Scoring and decision: Target: 5 working days

• Offer: Target: Within a week of the final round

6. For candidates

• Ask for the scoring criteria and the time cap before you start.

• Ask whether AI is allowed.

• If the test looks like real production work or will take more than a weekend, ask for a shorter version.

• Keep your files, and mark them "submitted for evaluation only".

Sources

• Sackett, P. R., Zhang, C., Berry, C. M., and Lievens, F. (2022). Journal of Applied Psychology, 107(11), 2040 to 2068.

• Greenhouse, "How anonymous take-home tests mitigate bias and improve candidate odds".

• Playdawn data: anonymised client feedback and candidate calls, 2026; Playdawn case study "We Ran Their Assessment. Then They Asked What Else We Could Run."