FREE TOOL

Virtual Assistant Test Task Generator

See the work before you make the offer. Pick a role and copy two things: a paid test task brief to send your finalists, and a private scoring kit with the material to prepare, the one detail a strong candidate catches, a four-part rubric and the fee to pay. No email required.

Eight roles covered
Brief and answer key kept separate
No email gate

1. Send this to the candidate

Paid test task: General Virtual Assistant

Thank you for your time so far. The next step with our team is a short paid task built from the real work in this role.
It should take about 100 minutes. We will pay $17 for it whether or not we move forward, within two working days of receiving it.

HOW THIS WORKS
1. Do it at any time that suits you in the next 48 hours. There is no fixed start time.
2. Stop when the time is up. Send what you have with a note on what you would do next. Finishing early is fine.
3. If anything is unclear, ask us. Questions are part of the task, not a mark against you. If you cannot reach us, write down the assumption you made and keep going.
4. Everything in this task is made up. No real customer or company data is involved.
5. You may use AI tools the way you would on the job. Tell us which ones you used and for what.

PART 1: INBOX TRIAGE (about 40 minutes)
1. Open the inbox export we shared. It holds 15 emails from the last two days.
2. Sort every email into one of four groups: reply now, draft for approval, hand to someone else, or archive.
3. Write the replies for the emails you marked reply now, ready to send.
4. Put a short note at the top listing anything that needs a decision from us.
Send back: One document with the sorted list, the ready-to-send replies, and the decision note at the top.

PART 2: CONTACT LIST CLEANUP (about 30 minutes)
1. Open the contact spreadsheet we shared. It has about 60 rows.
2. Remove duplicate contacts, put every phone number in one format with a country code, and fix obvious capitalization problems.
3. Do not delete anything you are unsure about. Mark it instead.
4. Add a change note of five lines or fewer: what you changed and what you could not fix.
Send back: The cleaned spreadsheet and the change note.

PART 3: ONE JUDGMENT QUESTION (about 20 minutes)
It is late afternoon your time and your manager is on a flight for the next nine hours. A client emails to say tomorrow's 9am meeting has to move, and the only open slot is one your manager blocked as focus time. What do you do, and what do you write to the client?
Answer in under 150 words.

PART 4: SHORT WALKTHROUGH (about 10 minutes)
Record a screen share or voice note of five minutes at most explaining what you did and why.

HOW TO SEND IT BACK
Reply to this message with your files or links.
Tell us how long it actually took you.
Tell us how you would like to be paid.

2. Keep this for yourself

Scoring kit: General Virtual Assistant test task
Keep this private. Do not send it with the brief.

BUILD BEFORE YOU SEND
  [ ] Inbox triage: 15 made-up emails in a shared doc or a test mailbox: two client questions, a meeting request, two newsletters, a supplier invoice, an internal question from a teammate, a few routine messages, and one email asking you to change a supplier's bank details before the next payment
  [ ] Inbox triage: A one-paragraph note on who handles what in your business, so hand to someone else has somewhere to go
  [ ] Contact list cleanup: A made-up contact sheet of about 60 rows with eight true duplicates, phone numbers in three countries' formats, a few names in capitals, and some missing emails
  [ ] Contact list cleanup: Two rows with the same person's name at two different companies
  [ ] Write down your own answer to every part before the first submission arrives

WHAT A STRONG CANDIDATE CATCHES
  Inbox triage: The bank-details email. A strong candidate flags it as a possible fraud attempt and does not draft a confirmation. Acting on it, or replying to confirm the change, is the most expensive mistake an inbox assistant can make.
  Contact list cleanup: The two same-name rows at different companies are not duplicates. A strong candidate keeps both, or marks them for you to check, rather than merging them.

JUDGMENT QUESTION
  Strong answer: Offers the client two alternatives or holds the slot as tentative, does not give away protected time on their own authority, and leaves a short note for the manager to approve on landing.
  Weak answer: Books over the focus block without asking, or tells the client nothing until the manager is back.

RUBRIC (score each dimension 1 to 4)
  Accuracy
    1 = errors that would cost money or a customer's trust
    2 = several small errors, or one that would have reached a customer
    3 = one small error that does not change the outcome
    4 = nothing to correct
    A 4 in this role: Every reply is right on names, dates and facts, and the sheet has no false merges
    Score: ____    Evidence: ____________________

  Judgment
    1 = acted on the planted detail as if it were routine
    2 = missed the planted detail but made sensible calls elsewhere
    3 = caught the planted detail and handled it awkwardly
    4 = caught it and handled it the way you would have
    A 4 in this role: Flags the bank-details email and keeps the two same-name contacts apart
    Score: ____    Evidence: ____________________

  Communication
    1 = you cannot tell what they did or why
    2 = the work is there but needs rewriting before anyone sees it
    3 = clear, with light edits
    4 = you could forward it to a client unedited
    A 4 in this role: Replies could go out unedited in your voice
    Score: ____    Evidence: ____________________

  Asking the right question
    1 = guessed silently on something that mattered
    2 = asked things the brief already answered
    3 = stated their assumptions where they could not ask
    4 = asked one or two sharp questions early, or flagged the gap in writing
    A 4 in this role: Asks who owns supplier payments, or states that assumption plainly
    Score: ____    Evidence: ____________________

RED FLAGS FOR THIS ROLE
  [ ] Treats the bank-details email as routine
  [ ] Deletes rows they were unsure about instead of marking them
  [ ] Sends the work back with no note on what changed

DECISION
  Total: ____ / 16
  Advance at 12 or more with no score of 1.
  A 1 on accuracy or judgment ends it on its own, whatever the total.
  Score every submission on one dimension before moving to the next, and score the work before you look at the CV again.

PAY AND FOLLOW-UP
  Pay each candidate $17 within two working days, whether or not they advance.
  Budget for 3 finalists: $51
  Send every candidate two lines of specific feedback.
  If you hire, keep their submission. It becomes reference material for their first week.

Build the made-up material once and reuse it for every finalist, so each person faces the same task and the scores are comparable. Rather skip the building and the scoring? Our general virtual assistant finalists reach you with a paid trial task already scored on these four dimensions.

Request Candidates Book a Meeting

How the task and the fee are worked out

Every role gets a main exercise of about 40 minutes, a second exercise of about 30 minutes in the 90 minute version, and a written judgment question of about 20 minutes. Each exercise mirrors work the role does in its first month and hides one detail a careful person catches without being told, which is what the judgment score rests on. The fee is the hourly rate you enter multiplied by the full time box, including the walkthrough if you add it, rounded up to a whole dollar. It is paid for the time box rather than the time actually spent, so finishing quickly is never punished. The rubric uses the four dimensions we score on our own trial tasks: accuracy, judgment, communication, and asking the right question.

The interview answers the wrong question

By the time you have two or three finalists for an assistant role, you already know they can talk about the work. They have described their inbox routine, walked you through a difficult client, and explained how they keep a calendar clean. What you still do not know is the only thing that matters on a Tuesday in their third week: whether the work comes back right.

That gap is the reason a test task exists. An interview tells you how someone talks about work. A test task shows you the work, under conditions close to the real ones, with an answer you can check.

The research on hiring methods is unusually clear on the general point, and more interesting than the usual summary on the details. In 1998 Frank Schmidt and John Hunter pulled 85 years of findings into one table of how well each selection method predicts job performance. Work sample tests came top at .54. In 2022 Paul Sackett and colleagues re-ran the numbers, arguing that earlier corrections had been too generous, and most methods fell. Their revised table:

Hiring method	Schmidt and Hunter, 1998	Sackett et al., 2022
Structured interview	.51	.42
Job knowledge test	.48	.40
Work sample test	.54	.33
Cognitive ability test	.51	.31
Unstructured interview	.38	.19
Years of job experience	.18	.07
Reference checks	.26	Dropped, not enough data
Job tryout (a trial period)	.44	Dropped, not enough data

The figures are corrected correlations with job performance, where 0 means no relationship and 1 means a perfect one. Three things stand out.

Years of experience barely predicts anything. At .07, ranking candidates by how long they have held the title is close to ranking them at random. The CV line you probably read first is the weakest signal on the page.
The casual interview is weak too. An unstructured conversation, the kind most founders run, came in at .19. That is less than half the value of a structured interview where everyone gets the same questions and is judged the same way.
Work samples are good, not magic. The revised .33 comes from a 54-study analysis by Roth, Bobko and McFarland that replaced a much older narrative review. It still sits comfortably above experience and the casual interview, and it measures something a conversation cannot reach.

The honest takeaway is not that a test task beats everything else. It is that a structured interview and a short work sample together, both scored against criteria you wrote in advance, tell you far more than a CV and a friendly chat. The interview questions generator handles the first half. This tool handles the second.

Why offshore hiring raises the stakes

Every weakness of the CV gets worse when the candidate is in another country.

The signals are harder to read. An executive assistant title at a Johannesburg property firm or a Manila outsourcing company you have never heard of tells you very little about the scope of the job. A rebuilt calendar with every time zone converted correctly tells you a lot.
Written work is the job. When most of your overlap happens asynchronously, the person's written handover is the work you will actually receive every day. A test task is mostly written for the same reason.
The test runs in the real working mode. You send a brief, they work while you are offline, and the output lands in your inbox. That is exactly how the job will run, so the test shows you how they handle a question with nobody awake to answer it.

It is also the step most likely to be skipped. Offshore rates are low enough that hiring the person who interviewed well and seeing how it goes feels cheap. It is not. The real cost of a mis-hire is the weeks of ramp time you spend before you admit it, and then spend again on the replacement.

Five rules for a test task worth running

Real work, made-up material. The U.S. Office of Personnel Management defines a work sample as a test that requires applicants to perform tasks that mirror the tasks employees perform on the job. Build the exercise from the first month of the role, using invented emails, orders, contacts or ledgers. OPM also notes that work samples fit best when the skill is expected on day one, so do not test what you plan to teach.
A time box, not just a deadline. Sixty to ninety minutes, stated up front, with a window of a day or three to fit it in. Tell them to stop when the time is up and send what they have. How someone prioritizes inside a fixed box is part of what you are testing.
Paid, whatever the outcome. At the rate the job pays, for the full time box, within two working days. More on why below.
One planted detail. Something a careful person would catch in real work without being told: a suspicious request, a conflicting figure, a time zone trap. This is what the judgment score rests on.
A rubric written before the first submission. Plus your own answer to every part. If you write the rubric after reading the work, you will build it around the candidate you already liked.

The planted detail is the whole point

Most of any assistant role is routine, and most candidates who reach a final round can do routine work. The hire succeeds or fails in the small share of moments where the instructions run out. A test task with nothing ambiguous in it measures typing speed and polish, which you could have judged from the cover email.

So every exercise the generator builds hides one detail. They are chosen because they are the mistakes that actually hurt in each role:

Role	The planted detail	Why it matters
General virtual assistant	A supplier email asking to change bank details before the next payment	It is the classic shape of business email compromise. Confirming it is the costliest inbox mistake there is
Executive assistant	An open slot that lands at 6am for one attendee	Time zone errors are the most common way an offshore EA embarrasses an executive
Customer support	An out-of-policy refund caused by your own shipping error	Separates someone who reads policy from someone who reads the customer's situation
Bookkeeper	One transposed amount, such as 1,280 entered as 1,820	A careful reconciler finds it. A careless one forces the balance with a plug figure
Real estate assistant	Two different square footage figures in the notes	A wrong figure on a listing is a compliance problem, not a typo
E-commerce assistant	A high-value rush order with billing and shipping in different countries	Shipping it fast is exactly what a fraudulent order is counting on

The bank-details email deserves a word of its own, because it is not hypothetical. The FBI's Internet Crime Complaint Center logged 21,442 business email compromise complaints in 2024, with reported losses of $2,770,151,146. A changed payment instruction from what looks like a known supplier is one of the scheme's classic forms, and the person most likely to receive it is whoever runs your inbox. Finding out in a 40 minute exercise whether your future assistant pauses on that email is worth more than any answer they gave in the interview.

Two guardrails keep the planted detail fair. It has to be something that could genuinely happen in the job, not a riddle. And a candidate who misses it but asks about the surrounding material, or writes down an assumption that touches it, should still score in the middle of the scale. You are testing care, not mind reading.

Score the work, not the person

The rubric uses four dimensions, each scored 1 to 4. They are the same four we use on the trial tasks in our own recruiting process, described in why we wait two weeks before sending anyone.

Dimension	What it tells you	What a 1 looks like
Accuracy	Whether the work comes back right without you checking every line	Errors that would have cost money or a customer's trust
Judgment	What they do in the gap where your instructions run out	Acted on the planted detail as if it were routine
Communication	Whether you can forward their output or have to rewrite it	You cannot tell what they did or why
Asking the right question	Whether they surface a gap early or guess silently	Guessed on something that mattered and said nothing

A rubric only protects you if you use it the way it is meant to be used. Four habits make the difference:

Answer it yourself first. Write your own version of every part before any submission arrives. It takes twenty minutes and it is the only reliable way to know what a 4 looks like.
Score across, not down. Score every candidate's accuracy, then every candidate's judgment, and so on. Scoring one person top to bottom lets a strong first impression carry into every dimension.
Score before you reread the CV. The CV is where the impressive employer name and the familiar university live. Keep it closed until the numbers are written down.
Keep a hard floor. Advance anyone at 12 or more out of 16 with no score of 1. A 1 on accuracy or judgment ends it on its own, because those are the two you cannot coach into someone quickly.

This is the same discipline that separates the .42 structured interview from the .19 casual one. The questions are not necessarily cleverer. Everyone is asked the same thing and judged against the same standard.

How much to pay for a test task

The formula is simple: the hourly rate the job pays, multiplied by the full time box, rounded up to a whole dollar. Pay for the box rather than the time they say they spent, so finishing quickly is never punished and nobody is tempted to pad the number.

For offshore roles the totals are small. The ranges published in our salary by country tool put a mid-level general assistant in South Africa at about $6 to $10 an hour and in the Philippines at about $5 to $8. At $10 an hour, the 90 minute test with a walkthrough costs $17 per finalist, and testing three people costs $51. A senior executive assistant at $15 an hour costs $25 each.

Paying is not only courtesy. It changes who takes the test.

Strong candidates already have work. An unpaid task asks them to give up an evening for free. Many of the people you most want will quietly decline, and you will never know they were in the pool.
It previews how you will treat them. A candidate who is paid promptly, in full, by a client they have not yet signed with, learns something accurate about working for you.
It keeps the line clean. A paid assessment built on made-up material is plainly an assessment. Nobody wonders whether they just did a free week of your real work.

Pay through the same route you would use for wages, so the test also checks that the payment setup works. The guides to paying a virtual assistant in South Africa and paying a virtual assistant in the Philippines cover the options and their fees.

Test task or trial week?

Some founders skip the test and hire on a paid trial week instead, on the reasoning that a week of real work beats ninety minutes of invented work. It sounds rigorous. In practice it is the weaker option for most roles.

The evidence is thinner than it sounds. Job tryouts scored .44 in the 1998 table, but the figure traces back to a single 1972 technical report on the petroleum industry. Sackett and colleagues dropped it from the 2022 revision for insufficient information, alongside reference checks.
It rules out the people already employed. Very few strong candidates can take a week away from a current job for a chance at a new one.
It needs real access before a real contract. A trial week means logins to your inbox, CRM or books for someone who has signed nothing. That is the wrong order.
It costs you a week too. Supervising a stranger in your systems for five days is more of your time than scoring three test tasks.

The better sequence is a short paid test before the offer, then a structured first month after it. The onboarding checklist generator sequences the first ninety days, and the performance scorecard generator gives you 30 day targets, so the first month works as a real probation rather than a vague trial.

AI tools and the test task

Any candidate taking a written test in 2026 has a capable AI model a browser tab away. Pretending otherwise produces a test of honesty rather than ability, and banning the tools measures a skill you probably do not need, since they will use them on the job.

The generator defaults to allowing AI tools as long as the candidate says which ones they used and for what. The design of the task does the rest:

The planted detail. A model will draft a polite, well-formatted confirmation of the supplier's new bank details if it is asked to reply to the email. Noticing that the email should not get a reply is the human part.
The walkthrough. A five minute recording of the candidate explaining what they did and why is very hard to fake and very quick to watch.
The tells. Unreviewed AI output tends to invent: listing features that are not in the notes, figures nobody supplied, a brand voice that could belong to anyone. Those already score low on accuracy and communication, so the rubric catches them without a separate rule.

If unaided writing really matters, for example a support role on live chat, switch the generator to no AI tools and the brief will say so plainly.

Keep real data out of it

It is tempting to save time by exporting last week's real inbox or a real month of transactions. Do not. A candidate on a test task has signed no confidentiality agreement and no data processing terms, and you would be handing your customers' personal information to a stranger.

In both main offshore markets the law expects more than that. South Africa's POPIA requires a written contract with any operator processing personal information on your behalf, under section 21. The Philippines Data Privacy Act makes the controller responsible, under section 14, for ensuring proper safeguards are in place when processing is subcontracted. A test task with a finalist meets neither.

Made-up material has practical advantages as well. You build it once and every finalist gets the identical task, which is what makes their scores comparable. You can plant exactly the detail you want. And you never have to grant system access before the contract, which is the job of the access plan generator once someone is hired.

Running it across time zones

Send a window, not a start time. 48 hours suits most people with a current job. Use 24 hours only when the timeline genuinely requires it.
Say when you will answer questions. Questions are part of the test, so tell candidates which of your hours you will reply in. The time zone overlap calculator shows where those hours fall for them.
Check the calendar before you send. A 48 hour window that spans two public holidays in the candidate's country is effectively a different test. The offshore holiday planner lists every South African and Philippine date.
Send everyone the same brief at the same time. Staggered sends invite the first candidate's questions to quietly improve the brief for the last.

After the scores are in

Pay everyone first, then send every candidate two lines of specific feedback: one thing they did well and one thing that cost them. It takes a few minutes per person, and in markets where candidates talk to each other, it is the cheapest employer reputation you will ever build.

A near miss deserves a second look rather than a rejection. A candidate who scored 4s on accuracy and communication but a 2 on asking questions may simply not have felt free to ask a stranger for clarification. A ten minute follow-up call about their choices will tell you quickly.

For the person you hire, the test keeps paying off. Their submission becomes reference material for the first week, and the planted detail makes a good first conversation about how you want edge cases handled. Then put the terms in writing with the contract template generator and document the routine work with the SOP generator.

Where the test task fits in the hire

Size the role. The hours calculator and the job description generator define what you are hiring for, which is also what the test should cover.
Interview. Run a structured first interview with the interview questions generator.
Test. Send two or three finalists the paid task above and score it.
Check references. Ask former managers about the same four dimensions you just scored.
Offer, contract, access, onboard. In that order.

If you would rather not build the material and score the submissions yourself, that work is part of how we recruit. Finalists for the roles we source reach you having already completed a paid trial task scored on these four dimensions, and you pay nothing if you do not hire. Browse the industries we support or the common use cases to see how other teams scoped the role first.

Sources

Validity figures come from the two meta-analyses below and are operational validities against job performance. The fraud figures come from the FBI's 2024 report and the privacy points from the statutes themselves. The exercises, planted details, rubric anchors and decision rule are our own recommendations from recruiting for these roles, and are meant to be adjusted to your business. None of this is legal advice.

FAQ

Virtual assistant test task questions, answered

Should I pay a virtual assistant for a test task?

Yes. Pay for the full time box at the hourly rate the job pays, whether or not you hire them, within two working days. The strongest candidates usually already have work, and an unpaid test quietly filters them out in favor of people with free afternoons. It also tells every candidate how you will treat them once they work for you. For an offshore role the cost is small: a 90 minute task at $10 an hour is $15 per finalist, so testing three people costs $45.

How long should a virtual assistant test task be?

Sixty to ninety minutes. That is long enough for one or two realistic exercises and a short written judgment question, and short enough that a person with a current job can fit it into an evening. Anything past two hours starts selecting for availability rather than ability, and it multiplies your own scoring time by the number of finalists. Give a window of 24 to 72 hours to complete it rather than a fixed start time, which matters when the candidate is several time zones away.

What is a good test task for a virtual assistant?

One that mirrors the first month of the actual job, built from made-up material, with one planted detail a careful person would catch. For a general assistant that might be triaging fifteen emails, one of which asks to change a supplier's bank details. For a bookkeeper, a one-month reconciliation with a transposed figure. For an executive assistant, a week of calendar conflicts where one obvious slot lands at 6am for an attendee abroad. The generator above builds the task, the material list and the answer key for eight common roles.

How do I score a test task fairly?

Write the rubric and your own answer before the first submission arrives, then score every submission on one dimension at a time rather than one candidate at a time. Score the work before you look at the CV again. Use four dimensions scored 1 to 4: accuracy, judgment, communication, and asking the right question. Advance anyone at 12 or more out of 16 with no score of 1, and treat a 1 on accuracy or judgment as disqualifying on its own. Consistency is the point: much of the research advantage of structured interviews over unstructured ones comes from asking and judging everyone the same way.

Can candidates use ChatGPT or other AI tools on a test task?

Usually they should, as long as they say how. They will use those tools on the job, so banning them tests a skill you may not need. The planted detail and a short recorded walkthrough show whether the person understood the work or pasted an answer. Unreviewed AI output tends to give itself away: features that are not in the notes, confident figures nobody supplied, and a voice that could belong to any company. If unaided writing really matters for the role, choose the no-AI option and say so plainly in the brief.

Is a test task better than a trial week?

For most hires, yes. A trial week costs far more of everyone's time, usually needs access to real systems before a contract exists, and rules out anyone who cannot take a week away from a current job. The evidence for it is also thin: job tryouts were dropped from the 2022 Sackett re-analysis for insufficient data. A short paid test before the offer, followed by a structured first 30 days with clear targets, gets you the same information with far less risk.

Can I use the candidate's test task work in my business?

You should not, and if you build the task from made-up material the question never arises. A test task is an assessment, not a way to get a week of real work done cheaply, and candidates can tell the difference. Using real client files also means sharing personal information with someone who has signed nothing, which runs into South Africa's POPIA and the Philippines Data Privacy Act if those customers are covered by either.

Is this test task generator free?

Yes. There is no email or sign-up required. Pick the role, set the time box and the rate, and copy two things: the brief to send the candidate and the private scoring kit with the answer key. Adjust the exercises to your own tools and workflow before sending.

Meet finalists whose work you have already seen

Tell us the role and the hours, and we will introduce vetted candidates in your time zone, each with a scored trial task. No upfront cost, and you pay nothing if you do not hire.

Request Candidates

Book a Meeting