Get paid for AI review work. One test. No CV.
What the work is
AI models are checked by people. Subcontractors read AI answers, find the errors, sort content into categories, and rate how accurate and clear each answer is. These are small, fast judgements that models cannot make reliably about themselves.
This is how AI is built. Every major model is trained and checked by people doing this exact work. The easy questions are done, so the demand now is for the hard ones, where models are most likely to be wrong. That is the work Jobpeak qualifies you for.
What the test really measures.
We do not test your credentials. We test what predicts good AI review work: what you know, and how you work.
The behavioural section is scored too. One example: 'You notice your accuracy dropping after an hour of work. What do you do?' It measures how you work: staying steady, knowing your limits, and judging well when the work is repetitive. That is what separates a good reviewer from someone with a degree.
Subject knowledge
50 questions from the specialty you choose. A mix of easy, medium and hard, including a section where you review AI answers in your subject, because that is what the work actually is.
Everyday judgement
30 behavioural questions from six trait groups: judging your own confidence, following instructions, handling unclear tasks, staying accurate when tired, spotting bias, and writing clearly. The traits that separate a good reviewer from a credentialed one.
Self-awareness
Spotting bias, explaining yourself clearly, and being willing to say when you are not sure.
Four steps to paid work
Practise free
10 questions in your specialty, from the same banks as the real test. Take it as often as you like, no card, a few minutes.
Unlock the Cognition Test
One payment of $25 at launch. Each payment buys one attempt. If you do not pass, you can pay again and sit the test again.
Pass at 78%
80 questions: 50 in your subject, 30 behavioural. No time limit.
Work, paid weekly
Qualified. Get Paid. Weekly.
Built for the way independent work really pays.
You are paid every week. Your earnings are calculated in USD and paid via Stripe, PayPal, direct bank transfer if you have an IBAN, or a supported local payout partner in your country. Open worldwide. We take no commission.
What you can earn
Rates by tier.
Rates rise with the difficulty of the work and the test behind it. Every rate is published up front.
BASIC
Up to $30/hr
Generalist work: reviewing AI answers on everyday subjects, sorting content into categories, and basic labelling. The entry route into the platform.
STEM
Up to $40/hr
Specialist work in mathematics, physics, chemistry and biology. You pass the Cognition Test in the science you choose.
CODING
Up to $100/hr
Reviewing code, fixing bugs in code written by AI, and judging model answers across programming languages. The highest live rate ceiling on the platform.
PROFESSIONAL
Up to $50/hr
Finance and accounting are live now. Law and medicine are still in development. Register your interest and we will tell you when they open.
After you pass
You pass
Your qualification is confirmed straight away, with no reviewer to wait for.
Within 24 hours
Your first tasks appear in your dashboard.
End of your first week
Add your details. Get your first payout!
Ongoing
New tasks appear as client demand comes in, and you work at your tier rate.
The tasks you'll be doing.
The work is the questions AI gets wrong. Models are fast and sound certain, even when they invent the answer. Subcontractors are the people who catch it. Two real examples from the question bank:
Spot the bug the model just wrote.
A model writes 'if x = 5:' in Python. One equals sign sets a value, two equals signs compare values, so the line should read 'if x == 5:'. Good subcontractors flag this in seconds. Most tasks are small, fast judgements like this.
Check the working, not just the answer.
A model gives a confident calculus solution with the wrong final answer. The mistake is rarely in the last line. It is usually a small algebra slip a few steps back, or a lost minus sign in a substitution. Often the final answer is right while the working is invented. Both count as failures.
See for free if you are ready.
10 questions, a few minutes, and a readiness report that tells you if the $25 test is worth taking.