Web Research Benchmark Designer

You will design research problems that remain difficult for AI browsing agents, even when those systems have full web access. The work is investigative: you will start with verifiable facts and build questions with clear, objective answers.

You will research unfamiliar subjects, use primary sources, and produce structured evidence that another person can check. Topics may involve dates, people, places, organizations, works, events, records, and quantities.

• Write natural-language research questions with short, stable, objectively verifiable answers.
• Develop independent clues that lead to the answer through web research.
• Search government and institutional databases, archives, registries, and PDF documents.
• Record obvious searches and the results they returned.
• Cite exact pages, tables, and sections to create a complete validation trail.
• Deliver structured research outputs; familiarity with JSON is helpful.

What It Pays and Takes

This is a part-time contractor role for an intermediate-level contributor. You should be comfortable investigating unfamiliar topics from scratch and documenting your findings in detail.

• Pay: $30 USD per completed label.
• Time: 20 or more hours per week.
• Location: Worldwide and remote.
• Language: Fluent written English.
• Education or experience: A master's degree or more than three years of relevant experience.
• Required background: Open-web research, precise sourcing, and experience with LLM evaluation, red-teaming, or benchmark construction.
• Helpful experience: Reference librarianship, archival research, investigative journalism, fact-checking, OSINT, due diligence, KYC, patent or prior-art searching, legal discovery, genealogy, competitive quizzing, or puzzle-hunt construction.
• You should have a high tolerance for structured documentation because the evidence trail is a major part of the work.

How It Works

Apply on OpenTrain with your resume, then complete the application on the hiring site.

About AI Training Work

AI training work uses human-created examples, reviews, and evaluations to improve how AI systems answer questions and complete tasks. Careful researchers are paid to provide reliable evidence and judgments that help measure and improve these systems.