Published: October 2, 2026
Design backward from the decision
Name the job behavior the sample must reveal: categorizing requests, identifying an escalation, drafting a bounded reply, or documenting a handoff. Then list the minimum facts needed to perform that behavior. Real names, addresses, account numbers, private correspondence, and complete histories rarely improve the hiring signal. If a detail does not change the candidate’s decision or allow scoring, remove it.
Build a synthetic case with realistic friction
Synthetic does not have to mean simplistic. Create fictional records that include an ambiguous priority, a missing approval, an outdated instruction, and a privacy-sensitive field the candidate should avoid copying. Make the relationships internally consistent and label the package as fictional. Have a role expert solve it before launch so the rubric reflects the information actually supplied rather than knowledge candidates cannot possess.
Control distribution and return
Deliver the exercise through an approved channel with a unique link or candidate workspace when practical. State who may access it, whether outside tools are permitted, the deadline, and what happens to submitted material. Avoid asking candidates to install unapproved software or use a personal account to reach company systems. When the exercise closes, revoke access and follow the recruiting retention rule for both the prompt and response.
Score boundary judgment
Include a criterion for recognizing when the available information or authority is insufficient. A polished answer that invents a customer fact or exposes unnecessary data should not outrank a response that pauses and escalates. Anchors can distinguish safe preparation, justified assumptions, unauthorized action, and clear handoff. Tell candidates the permitted resources so retrieval skill is not mistaken for cheating.
Pay attention to candidate-created information
A response may contain the candidate’s original writing, personal contact details, accessibility information, or examples from prior work. Instruct candidates not to submit confidential material from another employer or client. Limit reviewer access and do not repurpose strong submissions as unpaid production work. If the organization compensates candidates or obtains rights to materials, explain the arrangement in advance through the authorized process.
Run a preflight before every campaign
Confirm that the prompt contains no live personal data, secrets, active credentials, or hidden document history. Check every link from an external account. Verify that the scoring guide matches the current role and that an accommodation route is available. After the campaign, inspect access revocation and deletion exceptions. A safe work sample produces job evidence without turning candidates into temporary insiders or exposing people who never consented to the exercise.
Build a sample that cannot harm a real person
Create a fictional customer, invented company names, and synthetic dates that cannot be mistaken for a live queue. Remove real email addresses, account numbers, tokens, and hidden spreadsheet tabs. Ask someone outside recruiting to inspect the package for copied comments, revision history, file metadata, and links to internal systems. A clean-looking worksheet can carry sensitive material where the candidate never sees it.
The task should measure the role outcome without production access. A scheduling exercise can use a fabricated calendar. Research can rely on public sources. A support exercise can use a sealed set of fictional messages. If the skill can be observed safely this way, convenience is not a reason to invite a candidate into a customer workspace.
State the boundary before timing begins
Tell candidates which sources and tools are permitted, whether generative tools may be used, what must be disclosed, how long the task should take, and where to ask questions. Make clear that they should not use current or former employer information. A candidate who declines to share proprietary examples should not be penalized for respecting a boundary the hiring team ought to value.
Include an accommodation route that does not disclose the request to unnecessary reviewers. If timing changes for an approved reason, scoring should still use the published job criteria rather than assumptions about the accommodation.
Score decisions, not polish alone
Suppose the sample asks someone to paste a contact list into an unapproved tool. A strong response may stop, identify the issue, and propose a safe next step even if the final layout is less polished. Build the scoring guide around that judgment. Reviewers should cite the work and separate formatting preference from role requirement.
Use the same core task, instructions, time expectation, and anchors for comparable candidates. If a technical failure changes conditions, record it and choose a consistent remedy. Do not quietly give a favored candidate extra context through an informal call.
Track who received the package, where responses land, and when access expires. After the decision, apply the retention rule to submissions and reviewer downloads. Shared drives, email attachments, and exported PDFs all need attention. A deleted portal record does not remove a copy from a desktop.
NIST's Privacy Framework offers a risk-management structure for privacy decisions. It does not certify a work-sample platform or define a lawful basis for one employer's processing. Authorized privacy, legal, security, and recruiting owners set the controls. Teams can use OnboardingEmployees' candidate screening support for a cleaner evidence path.
Inspect the candidate's return path
A safe package can still fail if the response channel is unclear. Test upload permissions from an external account, file-size limits, confirmation messages, and who can download the submission. The candidate should not need to place work in a public link or personal shared drive because the recruiting portal rejects a file. Provide a support route that can solve access without asking for the work over ordinary chat. After receipt, confirm the submission is associated with the correct candidate and only assigned reviewers can open it. This practical check protects both privacy and fairness because technical friction should not decide whose evidence reaches the panel.
Separate evaluation from free work
The prompt should produce evidence, not a deliverable the organization intends to sell or deploy. Keep scope proportionate, state the expected time, and avoid live customer problems. If the team later wants to use an idea from the submission, follow the organization's authorized terms and review route. A hiring exercise becomes ethically and operationally weak when candidates unknowingly perform production work without the protections given to workers.
Continue building the workflow
Related support: virtual assistant role brief, virtual assistant onboarding checklist. The U.S. Equal Employment Opportunity Commission explains that employment selection procedures should be job related and consistent with business necessity.
Related reading

Run a First-Day Readiness Check for a Virtual Assistant
A practical virtual assistant first day readiness check for making evidence, ownership, decision boundaries, and handoffs visible.

Plan a Virtual Assistant's First Week Around Real Work
A practical virtual assistant first week manager plan for making evidence, ownership, decision boundaries, and handoffs visible.

Use a 30-Day Review to Strengthen Virtual Assistant Retention
A practical virtual assistant 30-day retention review for making evidence, ownership, decision boundaries, and handoffs visible.