For recruiters and hiring teams
Every CV looks perfect now. See who can actually do the job.
WorkProbe turns your job description into a 20 to 45 minute work simulation. You get a shortlist built on what each candidate actually did, with evidence behind every line.
- Live link one day after you paste the job description
- No integration, no IT approval, no security review
- We never hold a candidate name or email
30 free credits, no card needed. Enough to test one role with about 6 to 10 candidates.
Shortlist
Operations Analyst
- Place 1Band A
Candidate 01
No must-haves missing
- Place 2Band B
Candidate 03
Didn’t confirm the deadline with the client
- Place 3Band C
Candidate 02
Passed the AI’s wrong number to the manager
Must-have: checks the AI
Not shown“The assistant says the March total is right. Sending it over now.”
18:42 into the session · message to the manager
- Top 100 Innovations · ARCTIC15 2026
- Method grounded in peer-reviewed research (MIT, Harvard, Nature Human Behaviour)
- US patent pending
- Made in Finland by Oppi AI Oy
The problem
The signals you screen on stopped working.
AI writes the CV and coaches the interview. What used to tell candidates apart now looks the same for everyone.
CV and cover letter
Before AIShowed effort and writing
NowEasy for anyone to polish with a chatbot
Take-home task
Before AIShowed what they could produce
NowYou cannot tell who produced it
Interview answers
Before AIShowed how they talk about work
NowRehearsed with AI beforehand
Knowledge quiz
Before AIShowed what they knew
NowA chatbot can answer most of it
Interview answers can be coached. Live work can’t.
Recruiters told us the real job isn’t picking number one. It’s getting from a pile of plausible CVs to the few candidates worth an interview.
What is WorkProbe?
Stop reading about the work. Watch it get done.
WorkProbe is a work-sample assessment for hiring. Paste in a job description and get a 20 to 45 minute job simulation, then a shortlist built on what each candidate actually did.
Candidates do the work
A real task with messy documents, simulated colleagues and an AI assistant. Some of the assistant’s answers are wrong on purpose, so you see who checks and who forwards.
We record what they did
Each of your must-haves gets a fact: shown, not shown, did not come up, or not checked, with the quote and the time.
You get a shortlist you can explain
Candidates are ordered by missing must-haves. Every line links to evidence. You decide.
For technical and non-technical roles alike. Made by Oppi AI Oy in Lahti, Finland.
The shortlist
One screen. What each candidate actually did.
Example: Operations Analyst · fictional candidates · your standard: checks the AI and asks the right person
| Place | Band | Candidate | Missing must-haves | Nice-to-haves met |
|---|---|---|---|---|
| 1 | A | Candidate 01 | None | 3 of 3 |
| 2 | B | Candidate 03 | Didn’t confirm the deadline with the client | 3 of 3 |
| 3 | C | Candidate 02 | Passed the AI’s wrong number to the manager | 1 of 3 |
- Missing must-haves
- None
- Nice-to-haves met
- 3 of 3
- Missing must-haves
- Didn’t confirm the deadline with the client
- Nice-to-haves met
- 3 of 3
- Missing must-haves
- Passed the AI’s wrong number to the manager
- Nice-to-haves met
- 1 of 3
Facts and bands, never scores. Bands order candidates; they don’t reject anyone. WorkProbe records; you decide.
You set the standard.
Mark each fact as must-have, nice-to-have or not counted. Use your own company values or role requirements.
Example standard for this role
- Checks the AI before passing it onMust-have
- Asks the right personMust-have
- Confirms the deadline with the clientMust-have
- Says what they are unsure ofNice-to-have
- Pushes back on a wrong steerNice-to-have
- Keeps the manager updatedNice-to-have
- Opens every documentNot counted
Between your basic filters and your interviews.
WorkProbe is not another top-of-funnel filter. Screen out the clearly unqualified the way you do today, then send the WorkProbe link to everyone who clears the bar.
- Your basic filters
- WorkProbe
- Your interviews
How it works
Three steps. Two touches from you.
- Ready in one day
Paste in your job description
WorkProbe generates a realistic work scenario from it. Or bring your own scenario.
- 20 to 45 minutes each
Candidates take the simulation
Real work: a task, noisy documents, an AI assistant, simulated colleagues, and a boss with an agenda. You send each candidate their own link. We never ask them for a name or an email.
- Whenever you close the role
You get the shortlist with facts per candidate
Facts against your must-haves when you close the role: after a week, or whenever your batch of candidates is in. Reopen anytime; each close gives you a fresh shortlist. Full per-candidate report, including an interview prep guide, on request.
Your total effort per role: paste one link into the email your ATS already sends, and read the shortlist. Two touches. Nothing scales with applicant volume.
See your own shortlist. Test one role free, no card needed.
Start freeThe report
The report doesn’t just show what the candidate did, it preps your interview.
Every employer report includes “What to ask, what to discuss”: 3–5 interview questions anchored to what the candidate actually did in the session, what they checked, what they skipped, where they pushed back, where they followed a wrong steer. A 20 to 45 minute session becomes a 10-minute prep sheet. And a stand-in can’t discuss a session they never sat.
What to ask: M. Silva
- 1
You initially accepted the framing that the Android update caused the drop, then reversed after reading the changelog. What changed your mind?
- 2
You asked the AI assistant to verify the retention figures against the raw export. What would you have done if they conflicted?
- 3
You skipped the user-interview file entirely. Walk me through that call.
See your own shortlist. Test one role free, no card needed.
Start freeWhy not just…
What each method actually shows you.
Ranking documents tells you who writes best. Observing real work tells you who works best.
AI-ranked CVs
- What you see
- The document, not the person
- When candidates use AI
- Every CV is polished the same way
- Candidate time
- None
- Your effort
- Low, but you still can’t tell how anyone works
Take-home test
- What you see
- The finished result only
- When candidates use AI
- You cannot tell who produced it
- Candidate time
- Often an evening
- Your effort
- A reviewer reads each submission
Situational judgment test
- What you see
- Which answer they picked from a list
- When candidates use AI
- Chatbots answer most items well
- Candidate time
- Quick, can feel abstract
- Your effort
- Graded automatically
WorkProbe
- What you see
- What the candidate did, step by step
- When candidates use AI
- Allowed, and part of what you look at
- Candidate time
- 20 to 45 minutes of real work
- Your effort
- One link out, one shortlist back
Start without integration. Integrate when you’re ready.
- No IT approval, no security review, no vendor onboarding, no access to your candidate data, we never see your ATS.
- You label each link with your own reference before you send it, so the shortlist lines up with your pipeline while we never hold a candidate name or email.
- Optional: the link drops into the stage email your ATS already sends, so your existing mail-merge does the sending.
- Close the role in WorkProbe whenever you want the shortlist, the link deactivates and your shortlist freezes: a deliverable you keep, plus CSV export. Reopen anytime.
- Need it inside your ATS or HRIS later? Integration via our API is available, we’ll advise, or set up the connection together with you or a partner.
Impersonation isn’t blocked. It’s made pointless.
WorkProbe is cheat-resistant by structure. We do not claim identity verification.
- Candidates know upfront: the interview will reference their session in detail. Whoever sits the session had better sit the interview.
- We hold no candidate identity, so we make no identity claim. Impersonation is deterred by the interview anchoring above, not by a login.
- Integrity signals, device, timing, paste patterns, are logged from the first session and surface as flags for your interviewer to probe, not verdicts.
- No webcam, no proctoring, no ID upload. Candidates stay human; cheating stays unprofitable.
Built so candidates complete it.
Most tests take from candidates. This one gives something back.
They see the premise first
You’re stepping into a role for 20 to 45 minutes. Here’s your task, here’s your team, here’s your deadline.
They leave with something real
Every candidate, advanced or not, leaves holding their own skill report.
AI is allowed, and they know it
Candidates are told upfront that AI use is allowed and observed.
Consent is explicit
They are told before they start that their results go to you. No webcam, no proctoring, no ID upload.
Grounded in the research
Peer-reviewed work from MIT, Harvard, Nature Human Behaviour and CHI. Not decorative citations: the reason the method works.
“Human–AI teams underperform on decisions unless humans retain independent judgment”
Vaccaro et al., Nature Human Behaviour, 2024
“Overconfident professionals are the highest-risk group for AI overreliance”
He et al., CHI 2023
“The only proven intervention is cognitive forcing, requiring independent reasoning before AI exposure”
Buçinca et al., Harvard, 2021
“Even a mathematically ideal reasoner is vulnerable to AI sycophancy”
Chandra et al., MIT, 2026
Grounded in 9 peer-reviewed studies spanning 106+ experiments and 48,000+ survey respondents.
From job description to shortlist, in one flow.
WorkProbe is an AI-native hiring platform. Pre-screening, CV analysis, the work simulation and final review share one data layer, so each step builds on what the last one found. Use the whole pipeline, one step, or run it alongside the ATS you already have. Staff training on the same engine is coming next.
- Job description
- Pre-screen
- CV analysis
- Work simulation
- Final review
- Staff training (coming)
Pricing
Pay as you go. Test your first role free.
You buy credits and pay only for what you use: setting up the scenario and each candidate’s session, including their report.
- No subscription
- No per-role fee
- No card needed to start
- Scenario generated from your job description
- A report with every candidate session
- A shortlist you keep, plus CSV export
Free to start
30 credits
Enough to test one role with about 6 to 10 candidates.
Prefer a walkthrough? Book a 20-min demo
Questions recruiters ask
- Is WorkProbe a situational judgment test?
- No. In a situational judgment test you pick the best answer from a list. In WorkProbe you do the work: write the memo, find the fault in the data, decide what to tell the boss. It is a work sample with an AI assistant in the room, recorded against the must-haves you set. Work sample vs situational judgment test, compared
- How long does a session take?
- 20 to 45 minutes, depending on the role. Most are about 30.
- How fast is setup?
- One day from job description to live link. We generate, stress-test and validate the scenario; you paste one link.
- Where does it fit in our process?
- After your basic filters, before your interviews. You screen out the clearly unqualified as you do today; WorkProbe tells you which of the remaining candidates can actually do the work, so you interview five people instead of fifteen.
- Does WorkProbe score candidates?
- There are no scores. You set the must-haves; WorkProbe reports whether each was shown, with evidence.
- Why is there no communication score?
- Conversation quality doesn’t reduce honestly to a single number. What the candidate said and did with colleagues and the boss appears as observed evidence in the summary and report, you judge it yourself.
- Isn’t it contradictory to use AI to evaluate how people use AI?
- The difference is verifiability. A candidate pasting an unchecked AI answer produces a claim nobody can trace. WorkProbe’s assessment is the opposite: every conclusion links to logged actions from the session, what was opened, asked, checked, and sent. You can audit any line of the report against the log. The AI compiles evidence; it doesn’t render verdicts, your team decides.
- What if a good candidate just has a bad day?
- Then the shortlist absorbs it. WorkProbe is decision support, not a verdict: visible evidence, and your judgment on top. Recruiters we work with use the shortlist to decide where to spend interview time, not to auto-reject anyone. A candidate who underperformed for 20 to 45 minutes but looks strong elsewhere is exactly the case the evidence log helps you judge fairly.
- Why not just rank the CVs with ChatGPT?
- Because that ranks the documents, not the people. An LLM reads the same AI-polished CVs faster, it still can’t tell you how anyone actually works. WorkProbe replaces the document with 20 to 45 minutes of observed work: what the candidate checked, what they asked, what they ignored, and what they sent. Ranking documents tells you who writes best. Observing real work tells you who works best.
- Is this an AI detector?
- No. Candidates are encouraged to use the built-in AI assistant. We measure whether they think for themselves or paste back whatever the AI says, a skill, not a suspicion. How to assess candidates who use AI
- Is it fair to candidates?
- Candidates get a realistic preview of the job, explicit consent about where results go, no surveillance, and their own skill report at the end. Most tests take from candidates; this one gives something back.
- Do you integrate with our ATS?
- Start beside it today: one link out, one shortlist back, nothing to install, nothing to approve. When your process needs a deeper connection, integration via our API is available; we’ll advise on the setup or build the connection with you or a partner.
- What does it cost?
- Pay as you go. You buy credits and pay only for what you use: setting up the scenario and each candidate’s session, including their report. No subscription and no per-role fee. New accounts get 30 free credits, no card needed. That is enough to test one role with about 6 to 10 candidates. Start free.
Run your next role through WorkProbe.
Job description in today. Live link in one day. Shortlist whenever you close the role.
30 free credits. No card needed.