Skip to main content

Getting started

How to pass Outlier- and Mercor-style assessments

4 min read · Updated

Most AI training platforms screen new workers with a timed test, a writing sample, a practical task or an automated interview. Many people fail not because they lack ability but because they misread what is being measured. This guide explains the common formats and how to prepare. It contains general advice only: we do not publish or endorse leaked questions, and sharing them can get you banned.

What assessments are really measuring

Whatever the format, platforms are looking for four things: you can follow detailed instructions, your judgement is consistent with their quality bar, your written communication is clear, and you are who you say you are. Raw speed rarely matters. A careful, correct submission beats a fast, sloppy one.

That is why assessments often include a long guideline document. The test is partly whether you read it. Candidates who skim and rely on intuition tend to lose marks on edge cases the guideline explicitly covers.

Common assessment formats

  • Guideline comprehension quiz: multiple-choice or short-answer questions based on a style guide you must read first.
  • Ranking or rating task: you compare two or more model responses and pick the better one, usually with a written justification.
  • Writing or editing sample: write a prompt, rewrite a poor answer or summarise a passage to a stated standard.
  • Domain test: problems in coding, mathematics, medicine, law or finance, graded for correctness and reasoning.
  • AI-led or recorded interview: you answer questions by voice or text and an automated system produces a score that a human may review.

Preparation checklist

  1. Choose a quiet window of time. Many tests cannot be paused once started.
  2. Test your internet, microphone and browser beforehand, and close other tabs and apps.
  3. Read the full instructions twice. Take brief notes of rules about tone, formatting, safety and what to do when a request is ambiguous.
  4. Keep a dictionary or grammar checker allowed by the rules open if the test permits tools. Read the rules to confirm what is allowed.
  5. Warm up by reading well-written explanations and rewriting a few paragraphs concisely.

Doing well on ranking and rating tasks

The goal is to match the guideline, not your personal taste. For each pair of answers, decide against the stated criteria, such as helpfulness, accuracy, safety and tone. Verify factual claims, because a confident-sounding answer can be wrong.

  • Justify decisions in one or two clear sentences that cite evidence from the answers.
  • Notice subtle errors: wrong numbers, invented sources, ignored constraints, off-topic padding.
  • When both answers are poor, say which is less harmful or less wrong instead of inventing a middle ground.
  • Be consistent. If you treated a certain error as serious in one task, apply the same standard in the next.

Writing samples

Graders look for precise, natural language that follows the brief. Answer the question that was asked, use the requested length and format, and avoid filler. Proofread. A single spelling slip will not sink you, but a pattern of careless errors signals low reliability.

If you write for a specific audience, such as a teenager or an expert, hold that voice throughout. Mixing registers is a common reason for low scores.

AI interviews: what to expect

Some employers screen candidates with an AI-led interview. A system asks questions, listens or reads your answers and produces a score. Reputable employers tell you that AI is used, treat the score as an aid, and keep a human in charge of the final decision. On Chakri.ai the candidate notice says exactly that, and you can delete your interview data from your settings.

  • Answer in structured steps: situation, what you did, result.
  • Be specific. Names of tools, numbers and outcomes beat general claims.
  • It is fine to ask for a question to be repeated or to take a moment to think.
  • Speak at a normal pace in a quiet room with the microphone close.

If you fail

Failing a first assessment is common. Some platforms allow a retake after a waiting period; others do not. Do not create duplicate accounts, which is usually a bannable offence. Instead, review what you think went wrong, strengthen that area and apply to other platforms while you wait.

Keep a short log of each application: date, platform, format, result. It helps you spot patterns and avoid repeating mistakes.

After you pass

Passing opens the door but quality scores determine how much work you receive. Follow updates to guidelines, ask questions through official channels rather than guessing, and review feedback. Consistent high-quality work is the strongest route to better projects and rates.

Ready to look for work?

Browse live AI training, annotation and RLHF openings, each labeled with its source platform.

Browse jobs