
Outlier states an onboarding of 30 to 90 minutes with skill screenings, and Alignerr states Persona ID verification and, for some jobs, a "Zara interview or a skills assessment" before a calibration phase.
Three of the eight programs compared here state no screening at all, and none of them publishes a pass rate. This page lists what each states, what none publishes and what two court complaints allege about unpaid screening time, as the contributor side of AI training data licensing.
DataPacts Editorial Team: how we source, date and check every figure.
What an AI Trainer Assessment Tests, and Why Platforms Screen
An assessment tests whether an applicant can follow written guidelines and judge AI output the way a lab needs, because the platform sells the result of that judgment as training data.
The stages the platforms state are a short list: an account, an identity check, a test or an interview, and for some programs a calibration phase before the first paid task. Each platform combines them differently, and the next section sets the stated combinations side by side.
Platforms screen because labeler judgment is the product and it differs between people. The InstructGPT paper (arXiv 2203.02155, March 2022) states that the team "hired a team of about 40 contractors on Upwork and through ScaleAI", and it reports inter-labeler agreement of 72.6 ± 1.5 % between training labelers and 77.3 ± 1.3 % for held-out labelers.
Those figures describe one 2022 study, not the method of any platform today, but they show why platforms test and calibrate applicants before relying on their rankings. The ranking work behind RLHF data is what those labelers produced.
Screening Steps Each Platform States, Dated
Eight contributor programs state different steps before the first paid task, and three of them publish none, as of November 9, 2026. The table compares the stated steps, the test or interview and the identity and location checks, alphabetical by program, with each cell a claim of the named operator.
| Program (operator) | Stated steps before paid work | Test or interview | Identity and location | Source |
|---|---|---|---|---|
| Alignerr (ALIGNERR LLC, "powered by Labelbox") | Google sign-in, resume upload, Persona ID verification, apply; calibration phase; composite quality and speed score | for some jobs a "Zara interview or a skills assessment" | Persona ID verification; FAQ visa answer: only US identity documents accepted | alignerr.com; alignerr.com/en/faqs |
| DataAnnotation (Surge Labs, Inc., dba DataAnnotation) | Starter Assessment about an hour, once; specialized assessments 1 to 2 hours; result within a few days; no minimum hours | assessment | none on record | dataannotation.tech/faq |
| Handshake AI Fellowship (Stryder Corp., Handshake AI Solutions, LLC) | none on record | none on record | US-based; CPT and OPT may qualify, STEM OPT not supported | joinhandshake.com/ai |
| Mercor (Mercor.io Corporation) | none on record | none on record | none on record | talent.docs.mercor.com |
| micro1 | none on record for contributors; an AI interview is on record only in its company data program | none on record for contributors | none on record | micro1.ai |
| Mindrift video projects (Toloka AI BV) | no assessment; footage reviewed and accepted or rejected; setup time unpaid | none | residence in a listed country per project; VPN use leads to bans | mindrift.ai blog, 2026-09-17 |
| Outlier (Scale AI, Inc. and Smart Ecosystem, Inc.) | onboarding 30 to 90 minutes with skill screenings; Aether onboarding 15 to 60 minutes | Outlier Interview of about 20 minutes (Aether) | ID and location verification; valid ID and mobile phone from country of residence; VPN prohibited | outlier.ai; Flexible Working Guidelines, 2026-07-15 |
| Shipd (Datacurve AI, Inc.) | contributors compete on quests; top contributors are invited to projects | quests | not stated | shipd.ai |
Each step is a claim of the named operator on its check date of 2026-11-09, except where the source column gives another date. "None on record" means that no step appears on the pages checked, not that the program has none.
DataAnnotation's Starter and Specialized Assessments
DataAnnotation, the contributor brand of Surge Labs, Inc., screens applicants with a Starter Assessment that it says takes about an hour and can be taken only once, and with specialized assessments of 1 to 2 hours, checked November 9, 2026.
It states no minimum hours, and its pages list general AI response evaluation, coding, STEM, medical, legal and finance work and 43 languages. Public terms of service were not found on its site.
The pages checked do not state a pass rate, feedback after a failed test or whether assessment time is paid.
Third-party guides describe a "Core Assessment" with task counts, hours and waiting times that DataAnnotation itself does not state, so this page does not repeat them. Pay and reputation stay with the profile linked in the table.
Review Time, Results and Silence
DataAnnotation states that applicants hear the result within a few days, and it publishes no longer deadline after which silence means a no. The "thanks for taking the assessment" stage is a confirmation of receipt, not a result, and no on-screen wording is quoted here.
Waiting times longer than a few days appear only in third-party guides, which DataAnnotation does not confirm.
Interviews and Calibration at Outlier and Alignerr
Outlier and Alignerr add an interview to the test: Outlier states onboarding of 30 to 90 minutes with skill screenings, and for Aether 15 to 60 minutes plus an Outlier Interview of about 20 minutes, while Alignerr states that some jobs need a "Zara interview or a skills assessment" followed by a calibration phase.
Both statements were checked on November 9, 2026, and the Outlier profile holds its pay and reputation detail.
Alignerr states a Google sign-in, a resume upload and Persona ID verification before an applicant applies, and a composite quality and speed score once work begins.
Its Terms of Use also give it a right to use interview material, covered under unpaid screening time below, and its FAQ lists Hubstaff time tracking with periodic screenshots after onboarding.
Pass Rates and Acceptance: What No Platform Publishes
No platform in this table publishes a pass rate, an acceptance rate or the number of applicants it turns away, so any figure on how hard it is to get in is a guess.
The table lists four programs by whether a rate or feedback on a failed test is published and by the network size each states, as of November 9, 2026, with the size a claim of the named operator.
| Program | Pass or acceptance rate published | Feedback on a failed test published | Network size it states (claim) | Source |
|---|---|---|---|---|
| Alignerr | no | no | Alignerr: 100,000+ experts; Labelbox: 2.6M+ "skill-tested experts" (home page) and more than 4 million (Series E post) for its expert network | alignerr.com; labelbox.com |
| DataAnnotation | no | no | none on the pages checked | dataannotation.tech/faq |
| Meridial (Invisible Technologies) | no | no | 24,000+ people | invisibletech.ai |
| Outlier | no | no | 900K+ experts (about page) | outlier.ai/about-us |
A network size is not a pass rate: it says how many people a program lists, not how many applicants it turned away. The AI training jobs and platforms compared page puts pay, screening and reputation side by side and has the same gap.
The Alignerr and Labelbox counts differ by more than an order of magnitude, and they are shown as two claims, not merged or averaged.
Identity, Location and Background Checks Before the First Task
Identity checks run during onboarding: Alignerr states Persona ID verification, and Outlier requires a valid ID and a mobile phone from the country of residence, verifies ID and location and prohibits VPN use, checked November 9, 2026. Alignerr's FAQ accepts only US identity documents for visa questions while it states that applicants come from anywhere.
DataAnnotation's privacy notice, effective 2026-06-23, allows the collection of facial recognition data from contractors. No platform page checked describes a criminal or credit background check, and providers, retention and complaints are covered in contributor privacy on human data platforms.
Unpaid Screening Time: Platform Terms and Court Allegations
No platform on record states that its assessments are paid, and Alignerr's terms say it has no obligation to pay for the interview while taking a perpetual right to use the interview material.
DataAnnotation and Outlier are silent on whether assessment or onboarding time is paid, and Mindrift states that setup time for its video projects is unpaid, with no assessment.
Two court complaints allege that screening time was unpaid, and both are allegations, not findings. Cavalier v. Surge Labs, Inc. dba DataAnnotation, et al., filed on 2025-05-20 in the Superior Court of California, County of San Francisco, alleges misclassification and unpaid training and qualification exams, and the case was pending as of 2026-11-09. Boggs v.
Stryder Corp., No. 3:26-cv-07498-TSH in the Northern District of California, filed on 2026-07-20 against the operator of the Handshake AI Fellowship, alleges misclassification under the FLSA and Cal. Lab.
Code 2775 and unpaid assessments and onboarding; the order of 2026-10-22 (Dkt. 22) stays the case pending mediation, with a status report due on 2027-02-05. The full record is in contributor lawsuits against human data platforms.
Preparing for an Assessment Without Answer Keys
DataPacts publishes no assessment answers or question content, and the rules platforms state are the useful preparation: one attempt at DataAnnotation, location and VPN rules at Outlier and Mindrift, and an ID document ready before onboarding.
Forum "answers" files are not a source, and Outlier's confidentiality clause survives termination (Terms of Use s.5, effective 2026-05-11), though whether it covers assessment content is not stated.
- Stated length: DataAnnotation states about an hour for the Starter Assessment, and Outlier states 30 to 90 minutes of onboarding, checked 2026-11-09.
- One attempt: DataAnnotation states that the Starter Assessment can be taken only once.
- VPN rules: Outlier prohibits VPN use, and Mindrift states that VPN use leads to bans.
- ID documents: Outlier requires a valid ID, and Alignerr accepts only US identity documents for visa questions.
An offer that sells answers, training or certificates before any paid task fits the scam pattern, and how to check whether AI training work is legit has its own page.
Sitting an Unpaid Assessment: Benefits, Risks, Downsides and Alternatives
Sitting an assessment trades an hour or more of unpaid time and identity data for access to paid tasks that the platform does not guarantee, and each point below is a claim of the stating party as of 2026-11-09.
Benefits
- Paid work as stated: DataAnnotation states that general projects start at USD 25 to 50 per hour, Outlier states that it pays either hourly or per task, and Alignerr states that rates vary by project and skillset (checked 2026-11-09).
- A short stated length: DataAnnotation states about an hour, and Outlier states 30 to 90 minutes.
- Specialized assessments of 1 to 2 hours at DataAnnotation for specialized work.
Risks
- One attempt: DataAnnotation states that the Starter Assessment can be taken only once.
- Identity and biometric data: DataAnnotation's privacy notice of 2026-06-23 allows facial recognition data, and Alignerr states Persona ID verification.
- Complaints: noyb and partner groups filed five GDPR complaints against Scale AI on 2026-10-27, including over Persona biometric checks, as allegations with no authority decision as of 2026-11-09.
Downsides
- Unpaid time: no platform on record states that it pays for assessment time.
- No published pass rate, and no feedback on a failed test.
- Interview material: Alignerr takes a perpetual right to use it with no obligation to pay.
Alternatives, including not applying at all
- Projects that state no assessment, such as Mindrift's video projects, which still leave setup time unpaid and ban VPN use.
- Other programs in the table above, each with its own steps.
- Not applying, or waiting until a platform publishes its terms; DataAnnotation's public terms of service were not found.
Where Assessments Sit in Human Data Work
Assessments are the gate into human data work; the pages below cover what the work is, what it pays, how contributors are classified and how to check a platform. They sit within human data for AI, and the first three are the closest next reading.
- Pay for AI training work
- Expert contributor work by field
- How AI training contributors are classified
- Taxes for AI training contributors
- AI raters and evaluation data
- AI training work by language
Interviews at Mercor, micro1, Turing and Invisible
Mercor, micro1, Turing and Invisible Technologies recruit contributors too, but none of their contributor interview steps is on record here, and each company's own page covers what is.
The AI interview of micro1 is on record only for companies referred to its data program (checked 2026-11-09), and the micro1 data buyer profile describes it, while Mercor's referral terms say a referred person must be hired and start billable work.
Neither Turing's profile nor Invisible Technologies and its Meridial network, which states 24,000+ people, holds a screening step.
How This Page Is Compiled
Every step on this page comes from the platform's own pages or a court record, with the date it was checked; forum posts, videos and third-party guides are not used as sources. Platform pages were read on November 9, 2026, and a step that is not stated is written as none on record.
Frequently Asked Questions
How long does DataAnnotation take to review an assessment?
DataAnnotation states that results come within a few days, checked November 9, 2026. It publishes no longer deadline after which silence means a no, and waiting times longer than a few days come only from third-party guides that DataAnnotation does not confirm.
Can you retake the DataAnnotation Starter Assessment?
No: DataAnnotation states the Starter Assessment can be taken only once. The statement is a claim of the operator, Surge Labs, Inc., as of 2026-11-09, and its pages state no appeal or second attempt on the pages checked.
Are AI trainer assessments paid?
No platform on record states that it pays for assessment time, and Alignerr's terms say it has no obligation to pay for the interview. Cavalier and Boggs allege unpaid screening time; Cavalier was pending and Boggs was stayed pending mediation as of 2026-11-09.
Where can I find data annotation assessment answers?
Nowhere on this page: DataPacts publishes no answers or test content, and DataAnnotation allows one attempt, so the stated rules are the preparation. The preparation section lists the length, attempt, VPN and ID rules that platforms state.