Why Unsupervised Online Tests Tell You Nothing (And What to Do Instead)

A pre-hire test taken alone at home produces a number nobody can verify. Here is why unsupervised assessments fail as evidence, what the alternatives cost, and how a ten-minute supervised screen fixes the problem.

Most online hiring assessments share a design: the candidate gets a link, takes the test at home when convenient, and the employer gets a score. It is efficient. It is also, as evidence about the candidate, close to worthless, and the reason is not that candidates are dishonest. It is that the employer has no way to know, and so the score cannot be relied on either way.

This page explains what goes wrong with unsupervised testing, why the usual countermeasures fail, and how to run a short screen whose result you can actually trust.

What you do not know about an unsupervised result

• Who took it. A link and a name are not an identity. The candidate's more capable friend, partner or sibling can take the test, and sometimes does.
• Whether they had help. A second person in the room, a group chat, a phone.
• Whether they looked things up. Any question with a findable answer can be searched. Even "trick" questions circulate on forums within weeks of a test going live.
• Whether the test was seen before. Fixed forms leak. If the same questions are shown to every candidate, a previous candidate can describe them to the next one, and in competitive labour markets that happens routinely.
• Whether the conditions were reasonable. A timed test taken on a phone, on a bus, with a child in the room, measures the bus.

The honest candidate and the dishonest one produce the same kind of number. The employer cannot distinguish them, which means the number cannot be used to distinguish candidates. That is what "tells you nothing" means.

The countermeasures and why they fall short

Webcam snapshots and "AI proctoring". Some platforms photograph the candidate periodically or run software that flags "suspicious" behaviour. These raise legal problems of their own (biometric-privacy laws, automated-decision rules, and a growing record of false accusations against candidates with darker skin or disabilities), and they do not solve the core problem: a flag is not evidence, and an absence of flags is not proof.

Randomised question banks. Genuinely helpful against leaks, and a good screen should have them. But a randomised test taken alone still has the who-took-it and who-helped problems.

Longer tests. Making the test forty minutes does not make it verifiable. It makes it more tiring.

Honour statements. Worth having. Not worth relying on.

What verification actually requires

A result is evidence when a person you trust saw it produced. That is the whole principle, and it is why a work sample should be taken on a live video call with someone from your team present. Specifically:

1. The code is handed over on the call. The candidate cannot start without a single-use code, and the supervisor reads it out only once the candidate is visibly alone, camera on, entire screen shared. Nothing can be done in advance; the link alone is inert.
2. The supervisor watches. Ten minutes. Not to catch anyone, but because a present observer changes behaviour, and because if something is wrong the session can be stopped.
3. The session is recorded. Camera, microphone and screen, from Start to Finish, stored with the result. The supervisor watches live; the recording is the evidence afterwards, and it also keeps the supervisor honest.
4. The content is fresh. Every value on every attempt is drawn at random. There is no form to leak and no answer key to share.
5. The result is one input. A score and a plain band, read by a person who decides. The software never rejects anyone, which is both the fair design and the legally defensible one.

The Admin Skills Screen is built exactly this way. The cost of verification is ten minutes of a supervisor's time per candidate. For a role you are about to pay a salary to, that is not a cost; it is the cheapest due diligence available.

How fixed forms leak, in practice

Employers sometimes doubt that a test can leak. Here is how it happens, every time, in any competitive market.

A test platform has three fixed forms. A candidate takes Form B at home and does not get the job. She tells a friend, who is applying for a similar role next week, what the questions were like; she might even remember several exactly. The friend takes Form B too, because the platform rotates forms by date, and does well. Within a month there is a thread on a jobseekers' forum describing all three forms. Within three months, agency recruiters are coaching candidates on them, because a candidate who scores well is a placement fee. None of the people involved think of themselves as cheating. They are sharing information about a hurdle, the way people always have.

The only defence is to have no fixed form. When every value in every item is drawn fresh, a previous candidate can describe the format ("there is a reorder item, there is a list to remember") and it helps the next candidate not at all, because the content is new. Supervision then guarantees that the person taking it is the person being hired, and that they are taking it alone.

Who the unsupervised test is actually unfair to

The usual framing is that unsupervised tests are unfair to employers, who get a number they cannot trust. They are at least as unfair to honest candidates. The candidate who takes the test alone, in good faith, competes against candidates who did not, and loses to them on a number the employer treats as objective. A supervised screen is the only way the honest candidate's result counts for what it is. That is the argument we make to candidates who ask why they are being watched: it is the thing that makes their own effort mean something.

Setting up supervision without friction

Employers who run supervised screens for the first time tend to worry about logistics. In practice the setup is light:

• One supervisor, one login. Anyone on your team you trust. Many firms have their most reliable assistant supervise candidates, after someone has supervised her first.
• A video call the candidate can join from a browser. Google Meet in Chrome works; the candidate keeps the call in one tab and the exercise in another, and both can use the camera.
• Codes created on the dashboard, read out on the call, never sent. The link goes out in advance and is inert.
• Ten minutes present, muted. The supervisor does not help and does not need to; the exercise runs itself.
• The result on the dashboard the moment the candidate finishes, with the recordings linked beside it.

Fifteen minutes end to end, per candidate you are seriously considering. Set against the cost of the hire, that is the best fifteen minutes in the process.

"But supervision does not scale"

It scales to the number of people you are actually considering hiring, which is the number that matters. Unsupervised tests are built to screen hundreds of applicants cheaply. That is useful if you are a large employer sorting a thousand applications. For a small business hiring one assistant, it is the wrong tool: you do not need to sort a thousand people, you need to be confident about two or three. Ten minutes each, supervised, is trivially affordable at that scale, and the result is something you can trust.

If you do need to sort a large pool first, do it on the things an unsupervised step can legitimately measure, availability, basic requirements, written communication, and reserve the supervised screen for the shortlist.

The recording question

Recording a candidate raises a reasonable question: is it fair, and is it legal? It is fair when the candidate is told clearly what is recorded, for how long, and who will see it, and gives consent before anything starts; and when the recording is used only to review the attempt, never analysed by software, and kept privately. It is legal, in the jurisdictions we are aware of, on the same conditions, with the employer responsible for any additional notice its own jurisdiction requires. The screen collects explicit consent and a typed signature before the camera is requested, stores the exact consent text and version with the attempt, and keeps recordings in private storage readable only by the employer's staff. Candidates who are not comfortable can decline, and the employer decides how to proceed, as with any step in a hiring process.

What to do instead, in one list

• Interview first, as you do now.
• For the two or three candidates you would hire, run a ten-minute supervised, recorded work sample on a video call.
• Hand over the single-use code on the call, never in advance.
• Read the item-by-item result, not just the number.
• Follow up on anything surprising; offer extended time to anyone who needs it.
• Decide with everything in front of you. Let no score decide for you.

That replaces a number you cannot trust with ten minutes of observation you can.

A note for candidates who are reading this

Some of the people reading this page will be candidates wondering what they are about to be asked to do, so here it is plainly. The exercise is about ten minutes of everyday admin tasks, with nothing to study. You take it on a video call with someone from the employer's team present, on a laptop in Chrome, with your camera on and your screen shared, and it is recorded. You will be told all of that before you start, and asked to consent and sign. The recording is seen only by the employer's staff and is not analysed by any software. If you need extra time because of a disability, say so beforehand and it will be arranged. The supervision is not about catching anyone; it is what makes your own result mean something, because it proves it was yours.

Frequently asked questions

Can the supervisor be anyone? Anyone on your team you trust, with a login. Many employers have one experienced assistant supervise the others, with someone else supervising her first.

Does the candidate need special software? A laptop or desktop, Chrome, and a Google Meet link in another tab. The screen records from the browser; nothing is installed.

What if a candidate's internet drops mid-test? The attempt is single-use, so it cannot be restarted. The supervisor can create a fresh code, and the recording of the first attempt shows what happened. In practice this is rare and easily handled on the call.

Is the recording ever shared with the candidate? Not automatically. Candidates see their score and item breakdown immediately; the recording belongs to the employer's account. A candidate can ask the employer for a copy, and the privacy policy explains how.

How long are recordings kept? The employer's account keeps them for the retention period in the privacy policy, currently twelve months, after which they are deleted. Results and consent records are kept longer for record-keeping.