How to get started with Mercor, step by step

Create your profile

Go to work.mercor.com and make an account. Fill in a clear headline, the roles you want, your skills, your work history, portfolio links if they apply, and a resume saved as a text-based PDF.

Mercor uses the profile to decide which assessments and projects to show you, so be exact about what you have done. A line like "AI evaluation experience" tells a reviewer very little. Something like this does more: "Reviewed AI responses for accuracy, instruction following, clarity and completeness, scored them against a project rubric, and wrote down my reasoning for each rating."

Keep the resume honest. Cut any skill you could not talk about for two minutes, because the AI interview may ask you about anything on the page.

Apply to listings that fit

With the profile done, browse the listings. You can sort by fit, newest, trending or pay. An application may ask for a resume, proof of work authorization, an AI interview and a written assessment.

Mercor says these steps no longer have to happen in a fixed order. You can do them in whatever order suits you, and a finished step carries over to your other applications, so most of them only need doing once.

For a generalist background, these areas are worth a look:

• AI response evaluation
• Search quality and webpage evaluation
• Data annotation
• Technical research
• Content review
• Writing and editing
• Coding and prompt writing, if you have the skills

Take the AI interview

An AI interviewer asks questions tied to the role, records a transcript, and scores how well you fit the listing. It runs on camera and usually lasts 15 to 20 minutes, though the length changes from role to role.

Mercor offers a free practice interview first. It is not linked to any application and does not count against you later, so use it to get used to the format.

You can retake the official interview up to three times, and Mercor looks at your latest attempt. If you do not feel ready, there is no reason to burn the first try.

A simple shape works for most answers:

• Give your position in the first sentence.
• Explain why you hold it.
• Add one concrete detail, such as a result, a tool, a timeframe or a number.
• Stop when you are done instead of drifting.

The interviewer tends to reward specific answers in the first person. "I reviewed 200 AI responses against a rubric and flagged the claims that had no support" lands better than "We did AI evaluation work."

Prepare your examples

Have short examples ready for these questions:

• When you followed detailed guidelines.
• When you caught an error other people missed.
• How you deal with information that conflicts.
• How you decide things when the instructions are unclear.
• A project where accuracy or quality control mattered.

Pre-assessments and assessments

Assessments are structured tests that Mercor uses next to the interview. They check your skills, your judgment and whether you can do work like the real project. Expect a mix of multiple choice, written forms, recorded answers and task-style exercises.

Mercor picks them based on your profile, your resume, your field and the kind of data a project involves. If you pass a recommended assessment, you may be considered automatically for other roles that use the same one.

Before you start:

• Read each instruction twice.
• Run any camera, microphone or screen checks it asks for.
• Use a steady connection and a quiet room.
• Keep longer written answers saved somewhere else before you submit.
• Finish in one sitting when you can, and avoid long gaps.

Many assessments allow a single submission, so read your work through before you press submit. Some opportunities also close without warning, so send it in once you are happy with it.

What the tasks look like

A lot of Mercor work falls under evals, short for evaluation tasks. You read what an AI produced and rate it against a rubric or a quality standard.

A task might ask you to:

• Compare two AI responses and pick the better one.
• Rate accuracy, helpfulness, clarity, safety or instruction following.
• Spot factual errors, unsupported claims or missing context.
• Rewrite a response so it is clearer or more accurate.
• Write an ideal response from nothing.
• Research a claim and cite reliable sources.
• Apply a detailed rubric to coding, legal, medical, financial or technical content.

Each project comes with guidelines and a rubric. The guidelines set the rules, and the rubric gives you a structured way to score. Your job is to apply those standards the same way every time, not to pick whichever answer sounds nicer.

A sample task, rated

Generalist listings on Mercor include comparing AI responses for quality, coherence and helpfulness. Here is a made-up example of that kind of task.

The user asked: "What are some easy ways to save money on groceries without changing my diet too much?"

Response A said to plan meals for the week, check what you already have at home and buy only what you need. It suggested store brands for basics such as rice, pasta, canned beans and frozen vegetables, and comparing unit prices instead of sticker prices. It added that bigger packages save money only if you will use the food before it spoils.

Response B said to always buy in bulk because it is cheaper, to avoid fresh produce because frozen food is always healthier and cheaper, and to make a list but not worry about checking prices.

Response A wins on accuracy, helpfulness, instruction following, clarity and balance. Its advice is practical and realistic, and it admits that bulk buying helps only in some cases. Response B makes absolute claims that are not always true. Bulk items are not always cheaper per unit, frozen produce is not automatically healthier than fresh, and skipping fresh food is unnecessary. Telling the user not to check prices also works against the goal of saving money.

A written justification for this task could read: "Response A is more accurate, balanced and useful. It gives several practical steps: meal planning, checking what is already at home, choosing store brands, comparing unit prices and avoiding waste. It also notes that bulk purchases only save money when the food gets used. Response B makes overly broad claims and gives advice that could lead to wasted money or poor food choices. It also offers no useful way to compare prices."

Why precision matters

Precision is one of the biggest factors in how well you do on Mercor. Mercor says its quality thresholds come from project-specific criteria, including reviewed tasks, negative ratings and approved disputes.

To protect your score:

• Read the full prompt, both responses and the rubric before you rate anything.
• Do not trust a first impression.
• Judge factual accuracy separately from writing style.
• Check that each response followed every instruction.
• Flag what you are unsure about instead of guessing.
• Use the exact rating definitions from the rubric.
• Write clear, specific justifications when they are required.
• Keep formatting, citations and task requirements consistent.

A response can sound polished and still be wrong. Another can be correct and still incomplete. Precision means catching both and applying the rubric correctly.

What happens after you pass

If you pass, Mercor may match you with open projects, and some strong profiles get instant offers for roles that fit their completed assessments and experience.

Once you start work, Mercor generally asks you to track time in its Insightful desktop app. You start the timer when you work and pause it for breaks or when you finish for the day. Mercor also sets weekly work caps to protect quality and leave room for feedback.

A short checklist

• Build a detailed, honest profile.
• Upload a clean, text-based PDF resume.
• Apply only to roles that match your skills.
• Try the free practice interview.
• Prepare specific examples with numbers where you can.
• Read assessment instructions carefully.
• Treat every task like quality control.
• Follow the rubric exactly.
• Submit accurate work with clear reasoning.
• Protect your quality score from the first task.

The contractors who do best are rarely the fastest. They read carefully, apply the rubric the same way each time, explain their choices clearly and avoid careless mistakes.