Job seekers find a new source of income: Training AI to do their old roles

The catch? You have to be willing to train artificial intelligence to one day do your job as well as you can.

Welcome to the next gig economy. Instead of driving for Uber or delivering Postmates, a new wave of workers is signing up to school AI. These white-collar contractors review and critique the output of the large language models that power chatbots and other AI tools.

Not just anyone can work for Mercor, an AI startup valued at $10 billion. Applicants have to demonstrate their abilities in the interview process. And you might work on the same project for weeks or even months.

The vast list of subject-area experts Mercor seeks includes astronomers, psychologists, industrial engineers, filmmakers, creative writers, comedians, legal experts, investment bankers and venture capitalists. A dermatologist can make as much as $250 an hour helping a healthcare partner develop its “decision-support tools." Poets who “enhance AI’s understanding of poetic structure, literary nuance and emotional expression" can earn as much as $150 an hour.

Mercor hired over 30,000 contractors in 2025 to work on projects for some of the largest AI companies. Clients have included OpenAI and Anthropic.

Economic uncertainty, tariffs and a wait-and-see approach regarding the extent to which AI can handle tasks have all contributed to a growing unemployment rate, which in November hit its highest level in four years. White-collar workers are finding themselves applying to hundreds of jobs. For many, that now includes gigs with Mercor.

People can earn a $250 referral bonus by getting others to sign on with Mercor, contributing to a flood of job postings on LinkedIn and speculation that the whole thing must be a scam. (It’s not.) But training AI to take over human tasks could be considered a dark irony in today’s anemic job market.

When Katie Williams, 30, first saw a job listed for Mercor, the posting said the company was looking for a video editor. Williams, who lives in Houston, went to school for video editing and has worked in news and social-media marketing. She applied and soon landed an interview, conducted by an unseen AI proctor with a male-sounding voice.

She wasn’t quite sure what the job would entail but the pay was as much as $45 an hour, so she moved ahead.

Williams is now about six months into various projects that have involved watching video clips and writing out captions of everything that’s happening in them, and rating the quality of videos generated by prompts. She has mixed feelings about the work.

“I joked with my friends I’m training AI to take my job someday," she says.

Co-workers in her Slack channel express similar sentiments, she adds. They don’t feel great about training AI but they feel their job prospects are limited.

“Many of the people we work with already see AI as inevitable in their field, but that doesn’t mean humans will run out of meaningful work," a Mercor spokeswoman said in a written statement. “Many of our experts see it as their responsibility to infuse their knowledge and expertise into the models to ensure accurate and thoughtful outcomes."

Applicants undergo an initial video interview conducted by AI, and the company says a small percentage are required to share their screen to show their work.

Once hired, workers must install time-tracking software. That software ensures contractors are working during billable hours and aren’t cutting corners by using AI to critique the AI—which some have been caught doing, according to the Mercor spokeswoman.

After more than 20 years at the same job as an automotive journalist, Peter Valdes-Dapena was laid off in 2024. He spent months sending out résumés for full-time jobs to no avail. He finds freelance work inconsistent and it doesn’t make up for his past salary. Though he saved for his retirement, he’d rather not start dipping in yet.

One day, Mercor popped up in his LinkedIn feed.

The 61-year-old now spends 20 to 30 hours a week critiquing AI’s attempts at writing news articles. He finds the work challenging and says it’s had the pleasant side effect of improving his own writing.

The nature of the work does produce some internal conflict. Valdes-Dapena says journalists will always exist—he thinks people appreciate ideas and writing from humans—but he worries AI could lead to more job losses.

“I didn’t invent AI and I’m not going to uninvent it," he says. “If I were to stop doing this, would that stop it? The answer is no."

Laura Kittel, an academic currently seeking nonprofit and government jobs, learned about Mercor from a friend. She applied and soon got an offer.

The next step was to sign a contract, which was reviewed by The Wall Street Journal. She interpreted it to mean that she was granting royalty-free rights to her existing and future academic papers and any other intellectual property that might benefit an unspecified Mercor client.

“I thought that’s a bit of an overreach," she says.

Kittel tried to amend the contract. “Melvin," a Mercor AI assistant, replied via email saying the contract couldn’t be amended. If she didn’t like it, she could walk.

“It’s very much a feeling of having a lot of your dignity taken away from you," she says.

Mercor says the contract only applies to a worker’s writing, code or other intellectual property if the worker chooses to use it in the course of their project. Anything the worker has created on their own and doesn’t use is exempt from the terms of the agreement.

Sara Kubik, an attorney with her own solo practice, generates supplemental income during slow months by doing contract AI training work for Mercor. She has also worked for two other AI-training companies. She says one project she worked on for Mercor was on behalf of OpenAI.

Kubik says most people who approach her about Mercor are looking to get some work themselves, but some make negative comments. They’ve alleged to her that Mercor is a scam or said she’s training AI to take jobs. Kubik balks at the idea that AI will replace lawyers (though she says some legal assistants might not be so lucky). She finds training AI rewarding and looks forward to AI one day handling some parts of the job, such as screening new clients.

Her contracting work has convinced her there’s a lot of hype and grandiose statements about AI’s potential: “It’s taught me the limitations of AI."

Write to Katherine Bindley at katie.bindley@wsj.com