ChatGPT Didn't Push Freelancers Off the Platform. It Trapped Its Best Ones in Place.

Based on the research ofYiu, Seamans, Raj, and Liu, "Worker Repositioning and Technological Change: Evidence from an Online Labor Market," Organization Science, 2026

The intuitive story about AI and skilled labor runs in one direction: the more expert you are, the more exposed you are, and exposure should push you out. A 2026 Organization Science paper by Shun Yiu, Robert Seamans, Manav Raj, and Ted Liu tests that story directly on the supply side of a large online freelance marketplace and finds a sharper, more counterintuitive pattern. Freelancers whose pre-ChatGPT work sat in domains the model could suddenly do, writing, editing, translation, entry-level coding, and similar tasks, mostly did not quit the platform. They repositioned, shifting their bids toward higher-value contracts outside the newly automatable core. The freelancers who struggled to make that shift were not the least skilled. They were the most skilled, the ones with the deepest, most specialized human capital sunk into the category ChatGPT had just devalued.

The Migration No One Predicted

Using proprietary incumbent-freelancer data and a difference-in-differences design that leverages pre-period variation in AI exposure and skill level, Yiu, Seamans, Raj, and Liu show that OpenAI's launch of ChatGPT on November 30, 2022 reduced expected returns to labor in exposed skill domains almost overnight. The freelancer's rational response, in theory, is to leave a market whose returns just fell. That is not primarily what happened. Exposed freelancers reoriented their bidding toward higher-value contracts instead, both by changing which categories of work they pursued and by climbing the price ladder within categories still open to them. The paper frames this as vertical and horizontal repositioning: workers move sideways into new domains, upward into pricier tiers of existing ones, or both.

The twist is in who could not make that move. The authors find that repositioning is systematically less likely among higher-skill-level freelancers, because they face greater adjustment costs. A freelancer who has spent years building a portfolio, client relationships, and platform reputation inside one narrow, now-eroding specialty has more to lose by abandoning it and less transferable capital to bring into an unfamiliar higher-value category. Domain-specific human capital, the very thing that made these workers valuable before ChatGPT, is what constrains their ability to respond to it afterward. The paper inverts the usual AI-and-skill story at the point where it usually stops: exposure alone doesn't predict who adapts. Sunk expertise does, and it predicts adaptation negatively.

Where the Repositioning Actually Went

The aggregate market data from the platforms that publish their own numbers is consistent with a mass upward migration, even though it comes from different companies than the one in the paper's proprietary dataset. Upwork reported that gross services volume tied to AI-related work grew 70 percent year over year in full-year 2023, part of a year in which the platform's total GSV crossed $4.1 billion and active clients rose to 851,000, an all-time high. Fiverr told investors that searches for AI-related services on its marketplace increased more than 1,400 percent over a six-month window in 2023, and that searches for "AI consultant" services specifically rose over 650 percent between January and July of that year. Fiverr responded by standing up entirely new AI service categories to route demand that hadn't existed a year earlier. CNBC's reporting on the shift found individual freelancers commanding premium rates for the new AI-adjacent work, one AI consultant charging $1,000 for ten hours of consulting, well above what comparable general freelance work paid.

None of this proves the specific freelancers in the Organization Science paper are the same people generating Upwork's and Fiverr's AI-category growth. But it is exactly the kind of higher-value destination the paper's repositioning mechanism predicts: money and attention moved rapidly toward new, AI-adjacent, higher-priced categories on the same class of platforms where the original displacement hit. The freelancers most able to chase that migration were not necessarily the ones who had built the deepest expertise in the categories ChatGPT was eroding.

Skill Didn't Protect Anyone. It Just Changed the Trap.

A companion finding sharpens the picture. Xiang Hui, Oren Reshef, and Luofeng Zhou, in a 2024 Organization Science paper studying the same shock, found that after ChatGPT's release, monthly job volume for Upwork writing freelancers fell 2 percent and monthly earnings fell 5.2 percent. Crucially, they did not find that a freelancer's quality track record, measured by past ratings and employment history, protected them from that decline. If anything, their evidence suggested top freelancers were hit at least as hard as everyone else. Read alongside Yiu, Seamans, Raj, and Liu, the two studies describe the same population from opposite sides of the same event: demand for exposed work fell regardless of who was providing it, and the freelancers best equipped by reputation and specialization to have weathered that fall were, instead, among the least likely to escape it by moving.

Expertise was supposed to be the insurance policy against automation. In this market, it was closer to the collateral.

The mechanism is not mysterious once stated plainly. A freelancer with ten years of five-star reviews as a copywriter has a portfolio, a rate history, and a client base that only make sense inside copywriting. Pivoting into higher-value AI-adjacent work, prompt-assisted content strategy, workflow automation, or a genuinely different specialty, means starting over on trust signals a marketplace buyer actually uses to choose between bids. A newer, less specialized freelancer has less of that sunk value to walk away from, which is precisely why they can walk away from it faster.

What This Means for Anyone Managing a Displaced Workforce

The operating lesson generalizes past freelance marketplaces to any organization managing workers through an automation shock. The instinct is to worry hardest about your least skilled, most replaceable people, and to assume your most expert workers will land on their feet because expertise is portable. This research says the opposite dynamic can dominate at exactly the moment automation lands: your most specialized people may be the ones least able to move, not because they lack ability, but because their accumulated value is denominated in a category the technology just devalued. A repositioning strategy built only for junior or generalist workers will miss the group most structurally stuck.

The freelance-platform data makes the stakes concrete. Roughly $4.1 billion moved through Upwork in 2023, with the AI-related slice of it growing 70 percent while a measurable share of writing freelancers saw earnings fall 5.2 percent in the months right after ChatGPT's launch. Somewhere inside that gap between what shrank and what grew sits a specific, identifiable group: deeply skilled freelancers who did not reposition, watching a market reorganize itself around work they were, on paper, more qualified than anyone to do.