How to get accepted on Mercor as an AI evaluator

Mercor pays vetted experts to evaluate the outputs of frontier AI models. The work is async, remote, paid in USD, and the hourly rates beat almost any other remote gig if your domain is in demand. The catch is that Mercor screens hard. Most applicants do not pass. ## What Mercor evaluators actually do You read a prompt, look at the model output, and rate it on rubrics specific to the task (correctness, faithfulness, safety, code quality, reasoning). For developers, the bulk of work is code-evaluation tasks: pairs of model completions where you judge which is better and why, often with a written justification. ## What Mercor screens for Three things show up across most acceptance interviews: 1. Demonstrable depth in at least one technical or domain area (engineering, ML, biology, law, finance, etc.). 2. Written judgment under ambiguity (the application includes short written tasks). 3. Reliability signals (LinkedIn or GitHub profile that confirms what you claim). ## How to position your profile List specifics rather than generalities. A profile that says "Senior Python developer with 6 years at Stripe building risk systems" converts better than "Software engineer with broad experience". For technical reviewers, GitHub history with substantial commits weighs more than degrees. ## The screening flow You apply, complete a short written or coding screen, and if you pass you are invited to a paid trial task. Your first few tasks set your reliability rating; consistent quality unlocks more work and higher rates. ## Typical timeline The full path from application to first paid task usually runs two to four weeks. Domain-scarce profiles (legal expertise, advanced math, niche programming languages) get screened faster. ## What disqualifies fast Generic AI-assisted application copy is the fastest disqualifier. Mercor screens for human judgment; submitting LLM-rewritten answers signals exactly the opposite.