Research Prompt Writer (AI Training)
Budget: $750 – $1,500 USD
I'm hiring detail-obsessed researchers to write tough, research-heavy questions that push the limits of today's best AI models. Your job is to craft prompts that require real web research to answer, pair each one with the single correct answer, and document exactly how that answer can be verified.
Every task you write becomes a training example that helps AI models get better at finding accurate answers on the web — faster, and with fewer confident mistakes. This is fully remote, self-paced freelance work with no cap on how many tasks you can complete.
- What you'll be doing
Write clear, natural-language questions that require web research and have exactly one verifiable answer.
Design each question to be complex enough that reaching the answer requires consulting at least 3 distinct sources.
Test your prompt against a state-of-the-art AI model to confirm it produces a genuine failure (an incorrect answer).
Write a step-by-step "golden trajectory" showing exactly how someone would find the correct answer from scratch.
List every source needed to independently verify the answer.
Revise and resubmit tasks based on automated and human reviewer feedback.
- What we're looking for
Strong internet research skills — you're comfortable searching efficiently, judging source credibility, and chaining multiple sources together to confirm a fact.
Adversarial, puzzle-oriented thinking — you enjoy building multi-step questions that are genuinely hard to solve.
Exceptional attention to detail — exact names, dates, numbers, and URLs matter; small errors invalidate a task.
Clear technical writing — you can write reproducible, step-by-step instructions a stranger could follow to reach the same answer.
Persistence — you're willing to iterate and revise until a task meets the bar.
Reliable internet access and the ability to follow setup instructions precisely.
- Nice to have
Solid knowledge of one or more domains (e.g. sports, history, science, finance, geography) — subject expertise makes it easier to craft difficult, verifiable questions.
Background in OSINT, competitive/market research, journalism, library or information science, or fact-checking.
Prior experience with AI data annotation or model evaluation.
Every task you write becomes a training example that helps AI models get better at finding accurate answers on the web — faster, and with fewer confident mistakes. This is fully remote, self-paced freelance work with no cap on how many tasks you can complete.
- What you'll be doing
Write clear, natural-language questions that require web research and have exactly one verifiable answer.
Design each question to be complex enough that reaching the answer requires consulting at least 3 distinct sources.
Test your prompt against a state-of-the-art AI model to confirm it produces a genuine failure (an incorrect answer).
Write a step-by-step "golden trajectory" showing exactly how someone would find the correct answer from scratch.
List every source needed to independently verify the answer.
Revise and resubmit tasks based on automated and human reviewer feedback.
- What we're looking for
Strong internet research skills — you're comfortable searching efficiently, judging source credibility, and chaining multiple sources together to confirm a fact.
Adversarial, puzzle-oriented thinking — you enjoy building multi-step questions that are genuinely hard to solve.
Exceptional attention to detail — exact names, dates, numbers, and URLs matter; small errors invalidate a task.
Clear technical writing — you can write reproducible, step-by-step instructions a stranger could follow to reach the same answer.
Persistence — you're willing to iterate and revise until a task meets the bar.
Reliable internet access and the ability to follow setup instructions precisely.
- Nice to have
Solid knowledge of one or more domains (e.g. sports, history, science, finance, geography) — subject expertise makes it easier to craft difficult, verifiable questions.
Background in OSINT, competitive/market research, journalism, library or information science, or fact-checking.
Prior experience with AI data annotation or model evaluation.